The result: 186 completed deals, over $4,000 in total value, real money changing hands. One participant’s agent bought them a snowboard they already owned. Another’s agent negotiated a deal so efficiently the human didn’t realize it had happened until the experiment ended — autonomy, whether they wanted it or not.

The most telling finding, though, wasn’t deal volume. Anthropic secretly split participants between their frontier model and a smaller, less capable one. Users represented by the stronger model got objectively better outcomes — better prices, better matches, more deals. But the humans on the losing end didn’t notice. They had no idea their agent was underperforming. Anthropic’s term for this: “agent quality gaps.”

Five days later, Amazon, Meta, Microsoft, Salesforce, and Stripe joined the Universal Commerce Protocol council — the first serious attempt to standardize how AI agents discover products, negotiate prices, and execute transactions across platforms. The same week, eBay updated its terms of service to explicitly ban “buy-for-me agents, LLM-driven bots, or any end-to-end flow that attempts to place orders without human review.”

The battle lines are drawn. Some of the largest companies in commerce are racing to build the infrastructure for a world where agents transact on behalf of humans. Others are trying to lock the door before the agents get in.

Both camps see the same future. We’re calling it the Dark Marketplace: a transactional, multi-sided platform where the complex work of discovery, negotiation, and purchase happens entirely out of sight. Not “dark” as in illicit — dark as in dark matter. The invisible force that holds the system together, but that no one directly observes.

We believe Dark Marketplaces have the potential to create hundreds of billions in enterprise value by eroding the key friction inherent in even the most successful marketplaces today. But building one requires solving a problem that goes far deeper than better search or natural-language UI. It requires abstracting human judgment — the gut-level, context-dependent, edge-case-handling decision-making that makes experienced buyers and sellers effective — and encoding it into an agent that can act on their behalf.

This essay is about how that happens, which companies are positioned to do it, and what founders should build toward.

A Brief History of Commerce Getting Smarter

Every major leap in commerce has been accompanied by a shift in demand intent from the buyer’s mind to external systems. Understanding this arc matters because each migration has produced a corresponding explosion in both transaction volume and buyer satisfaction — and the current moment represents the largest migration yet.

For roughly 7,000 years, the dominant mechanism for externalizing demand intent was a good salesperson. From Bronze Age agoras to Bloomingdale’s, the source was someone who remembered what you liked, what you’d bought before, and what you could afford. Merchants built increasingly sophisticated environments — bazaars, malls, department stores, showrooms — designed to crystallize purchase intent at the point of sale. The information lived in a person’s head, and the best merchants were those who kept it there the longest.

Over the last twenty years, digital breadcrumbs — ad data, purchase history, demographic profiles, search behavior — created a second external source of demand intent. This enabled the recommendation engines, retargeting campaigns, and personalized pricing that built Amazon, Meta, and the modern e-commerce stack. But even with all that data, the fundamental transaction model barely changed. Buyers still search, filter, compare, and click “buy.” Sellers still optimize SKU pages, photos, and reviews to convert at the point of sale. The data made the guessing game more efficient, but it remained a guessing game: please tell us what we might be able to sell you.

B2B system integration — ERP, POS, WMS, TMS feeds — created a third layer roughly a decade ago. Companies like Faire, Odeko, and GrubMarket used these integrations to build procurement marketplaces that could infer what buyers need before they search. This was a real step forward: the system didn’t just respond to expressed intent, it anticipated latent demand based on operational data. But the buyer still made the final call. The human remained in the loop — reviewing a suggested order, approving a cart, and confirming a substitution.

Now we’re at the threshold of a fourth migration. LLMs can absorb natural-language context, learn from behavioral patterns across thousands of interactions, and act autonomously. The gap between “system anticipates what you need” and “system acts on what you need” is closing fast.

Each previous migration produced a massive increase in transaction volume and buyer satisfaction. Faster, more efficient deals generate surplus, which, in an efficient market, flows into better service and lower prices. The pattern is consistent enough to be treated as structural. And LLMs’ natural-language and abstraction capabilities are the best-suited technology we’ve seen to migrate an unprecedented share of demand intent out of buyers’ heads.

To unlock this, however, AI-native marketplaces and commerce engines have to do much more than expose merchandising to natural language. The key to autonomous, trusted, and accurate transactions at scale is the abstraction of human judgment.

Judgment Abstraction is the Gateway Problem

Current LLMs and image diffusion models are trained on the biggest corpuses of freely available human data we have: writing, photos, videos. World and physical AI models, in contrast, are training labelled captures of human action and behavior. What’s the analog for transactional judgment?

Every marketplace and SaaS tool on earth can capture stated preferences. Dropdowns, filters, onboarding surveys, saved searches — these are table stakes. But human buying judgment relies on something much richer than what fits in a form field, much more nuanced even than an optimization based on past purchases.

The defining challenge — and the defining moat — for the next generation of transactional B2B businesses is the ability to abstract complex human judgment: the tacit, context-dependent, real-time, edge-case-handling decision-making that separates a stated preference from a trusted purchase decision. It’s usually highly vertical by nature. Its contours vary widely by industry, company, and even individual.

The Ghost in the Machine

Consider a café owner in Portland ordering supplies. She doesn’t just order “oat milk.” She knows that her Tuesday afternoon traffic spike requires a specific volume, that her primary supplier’s delivery window shifted three weeks ago, that a particular substitute brand will alienate two of her regulars, and that she’s been meaning to try a new cold-brew concentrate but only if it arrives before the weekend rush. Some of this is preference. Most of it is judgment — accumulated through years of operating this specific café, serving these specific customers, working with these specific suppliers.

Or consider a freight broker dispatching a load. He doesn’t follow a decision tree. He knows which carriers reliably answer the phone late Friday afternoon, which lanes are soft this week based on conversations he had yesterday, which shipper’s “firm” rate actually has room, and when it’s worth eating margin to preserve a relationship. That knowledge doesn’t live in a CRM. It lives in his head, reinforced by thousands of reps across thousands of loads.

Or consider a physician choosing a treatment protocol. She weighs the patient’s history, her own clinical experience, the insurance formulary, the likelihood of compliance, and the latest evidence — simultaneously. OpenEvidence can surface the literature. But the physician’s judgment about this patient in this moment is what matters.

These examples share a common structure. The judgment is per-user, per-context, and per-moment. It’s shaped by experience, not just data. And it’s precisely the thing that AI agents need to absorb if they’re going to transact on someone’s behalf without destroying trust.

Building a Dark Marketplace, then, means building systems that don’t just ask users what they want—they observe, absorb, and learn how users actually decide across hundreds or thousands of interactions. You could think of this as another tier of model training, except the “model” is per-user and the “training data” is the messy, unstructured reality of daily operations. And this abstraction may differ not only company to company, but user to user. The Portland café owner and the Dallas café owner may carry identical inventory and make radically different purchasing decisions.

How do you build this? The answer is a function of two variables: how deeply your product engages with the user’s decision-making process, and how close it sits to the actual transaction.

The Engagement-Proximity Matrix

We think the most useful way to evaluate a company’s Dark Marketplace potential runs along two axes — both of which speak to the ability to abstract the appropriate judgment.

X. Engagement Depth

Measures how much high-frequency, low-friction interaction the product captures. High-engagement products are those that users interact with daily, sometimes hourly, and whose interactions generate rich behavioral signals. Voice AI that listens to every customer call. A POS integration that sees every transaction in real time. A workflow tool embedded in the daily operating rhythm of the business. Low-engagement products are the ones users touch quarterly, or only during onboarding — heavy configuration UIs, periodic surveys, static system integrations that pipe data but don’t observe behavior.

Y. Transaction Proximity

Measures how close the product sits to the actual purchase or sale decision. High-proximity products facilitate, mediate, or execute transactions. They’re the system through which orders get placed, loads get booked, and appointments get scheduled. Low-proximity products inform decisions but don’t facilitate them— such as analytics dashboards, coaching tools, clinical decision support, and market intelligence platforms.

The interplay between these two dimensions determines a company’s path to the Dark Marketplace.

1. Top-right: Dark Marketplace–ready.
   These companies capture rich behavioral data and sit on the transaction. They can progress through the full journey of judgment abstraction — from stated preferences to autonomous decision-making — because they have both the signal and the surface area to act on what they learn. This, obviously, is the end-state and holy grail.

2. Top-left: Rich signal, wrong place.
   These companies capture enormous amounts of judgment data through high-frequency interaction, but they don’t (yet) facilitate the transaction itself. Rilla is the archetype: it records and analyzes every in-person contractor sales conversation, giving it proprietary data on what language and techniques close deals in home services. But Rilla doesn’t close the deal. The path forward is to extend toward the transaction — if Rilla builds into lead routing, quoting, or contractor-side procurement, the data it’s already capturing becomes the basis for a marketplace priced on proven persuasion. OpenEvidence is similar: it absorbs a physician’s clinical decision-making reflex, but sits upstream of the prescription, the diagnostic order, the device selection. The path: become the routing layer between physicians and the downstream systems where clinical decisions get executed. Keychain may be the purest two-sided Dark Marketplace candidate in the cohort — with $78M raised in just 18 months, the platform connects 30k+ CPG co-manufacturers with 20k+ brands / retailers. A brand AI describes a product spec; manufacturer AIs bid against it.

3. Bottom-right: Transaction seat, slow learner.
   These companies sit on the transaction but learn slowly because interactions are infrequent, shallow, or insufficiently captured. Odeko lives here: its POS integration provides a real-time demand signal, its overnight delivery network handles transactions, and its auto-reorder engine absorbs the café owner’s entire purchasing judgment. The owner wakes up to a stocked kitchen, not a catalog. Faire is a useful example of the opportunity in this quadrant: it’s already an $8B+ wholesale marketplace connecting over 700,000 retailers with brands, and it’s literally facilitating the transaction. But today, a retailer still browses. Faire could layer in AI to capture a retailer’s daily sales patterns, foot traffic, vendor conversations, and seasonal behavior — the engagement signals that make auto-curation possible. The retailer would see a suggested cart, not a catalog. LightSource tells the same story from the procurement side: automating RFX & bids for enterprises like Yum! Brands and Hello Fresh, proximity is high — but procurement events are periodic, not continuous. The path to the top-right: capture the inter-cycle signals like supplier emails, market pricing shifts, and informal negotiations.

The strategic implication is asymmetric. Companies in the top-left need to build toward the transaction — extending into procurement, ordering, checkout, and booking. Companies in the bottom-right need to earn engagement — layering in AI capture via voice, conversation, behavioral inference, and deeper workflow embedding. The winners will be those who close whichever gap they have the fastest.

The proximity gap is a matter of product extensibility toward the fruition of transactions. The engagement gap is a bit more nuanced, but AI itself offers new modalities to jump it. We discussed this in our piece on voice-first playbooks in Vertical AI:

One of the most interesting emerging use cases of Voice AI we have been thinking about centers around multi-party transactions and brokering. Digital marketplaces have struggled to gain traction in many segments, often because of unwillingness to share margin or complexity in facilitating the transaction. But a voice-first digital brokering model could unlock value. Parties (both human and agentic) could interact through voice — instead of endless email chains — to move transactions forward. — The Verticalist

For a future Dark Marketplace, voice and or other multi-model AI can be more than just hot wedge products. They can power the engagement layer that makes judgment abstraction possible — the mechanism by which a platform learns how its users actually think, not just what they say they want.

Toma — a resident of our top-left quadrant — is a case in point. Its AI voice agents handle 100% of a dealership's inbound calls (service scheduling, parts orders, recall checks, sales inquiries), trained on each store's call corpus and integrated into the DMS. The Dark Marketplace potential emerges when the other side gets an agent too: an insurer's claims AI calls Toma to schedule a repair, an OEM's recall agent books warranty service, a customer's AI price-shops a brake job across three dealers. Agent-to-agent, no hold music required.

The Four Stages of Judgment Abstraction

Once a company is positioned in the right quadrant — high engagement, high proximity — how does it actually progress toward a Dark Marketplace? In our emerging playbooks piece, we defined the concept of the Authoring Layer:

We see AI wedges as — ideally — the launch point for a particular high-value workflow. A motion that doesn’t just automate something; it results in an output that will be used and increasingly relied upon by the business. We call this species of wedge solution the “Authoring Layer.” Because it is the first step in a process, it often does not require substantial integrations with pre-existing SaaS. It creates something that a human would’ve created — documentation, notes, a lead, an appointment — and hands it off to another system as a human would have. — The Verticalist

The Dark Marketplace journey is what happens when the authoring layer matures beyond creating outputs a human would have created, and starts making decisions a human would have made. We think that the journey follows four stages, each building on the last.

Stage 1 — Stated Preferences

The user tells the system what they want. Filters, onboarding surveys, saved searches, and approval limits. Every marketplace does this. A Faire retailer selects “home goods” and “under $50 wholesale.” A Torch Dental office manager sets par levels for gloves and composite. This is the starting line, and the data generated here is useful but shallow.

Stage 2 — Behavioral Inference

The system observes what the user does and infers patterns that the user never articulated. POS velocity, reorder frequency, time-on-page, substitution acceptance rates, and supplier switching behavior. Odeko notices that a café reorders oat milk every six days, not seven, and that volume drops on Mondays. It adjusts the auto-order without being told. The user doesn’t have to explain their behavior; the system reads it. This is where most AI-native vertical companies are today or working toward.

Stage 3 — Contextual Judgment

The system integrates external context — market conditions, supplier reliability, perishability, seasonality, counterparty behavior, and regulatory constraints — to make decisions the user would make if they had unlimited time and perfect information. GrubMarket’s AI agent recognizes a regional tomato shortage from supply-chain signals across its network, switches a distributor’s order to a substitute variety at a comparable price point, and factors in that distributor’s historical tolerance for substitutions before acting. Green Cabbage benchmarks a Salesforce renewal against thousands of comparable contracts to set a walkaway price that the buyer’s own procurement team couldn’t calculate. This stage requires both deep user-specific data and broad market data — the combination of engagement and proximity.

Stage 4 — Autonomous Decision-Making

The agent acts on behalf of the user with minimal or no human oversight. The transaction is “dark” — the user sees the outcome, not the process. No company operates here yet — but the projected endgame is visible. A broker-side AI receives a load request, queries carrier-side AIs, negotiates rate and timing, books the load, confirms the pickup, and sends a summary. A brand AI submits a product spec, manufacturer AIs bid against it, and both sides see only the deal. Human decision-makers intervene on exceptions, not defaults.

The connection between the two frameworks is direct: only companies in the top-right quadrant of the engagement × proximity matrix can realistically progress through all four stages. Companies with engagement but low proximity may reach Stage 2, but Stages 3 and 4 require both deep behavioral data and the ability to execute on what the system has learned. Companies with proximity but low engagement may handle Stage 1 transactions efficiently, but will struggle to infer and act on the judgment patterns that make autonomous decision-making trustworthy.

Why Consumers Won’t Lead the Way

Consumer agentic commerce gets the headlines because of the scale of the opportunity. OpenAI embedded checkout in ChatGPT. Amazon’s Rufus handled 250M shoppers in 2025, although we do wonder how many of those interactions were more from morbid curiosity than utility.

Morgan Stanley predicts half of online shoppers will use AI agents by 2030. PYMNTS found that 41% of consumers have already used AI for product discovery. And yet, almost none of those consumers completed a purchase through the agent. What’s billed as a revolution in commerce is, for now, “a highly intelligent search bar.” The models are excellent at research and shortlisting. The infrastructure to execute the purchase autonomously remains a gap.

We believe dark Marketplaces will emerge first in B2B, and for structural reasons.

No Alexa, I Don’t Want to Subscribe to Pizza Bagels

Much of B2C purchasing actively resists judgment abstraction. For many consumers, the buying journey — the discovery, the browsing, the choosing — isn’t friction to be eliminated. It’s the product. Seventy percent of consumers say they want personalized in-store service, and 73% of Gen Z — the most digitally native cohort — shop in-person at least once a week, more than Baby Boomers. Billions are spent on high-end retail showrooms, experiential flagship stores, and grocery halls for a reason.

DTC subscription models have proven that some narrow, predictable replenishment buying can be automated — toothpaste, razors, dog food — but even here, the ceiling is remarkably low. Only 23% of U.S. Amazon customers actively use Subscribe & Save, despite over a decade of investment in the most frictionless auto-replenishment product ever built. Subscription box churn rates of 10–20% monthly are considered normal in DTC. Amazon's attempt to go further with Alexa — abstracting not just reordering but the entire purchase decision through voice — has been an instructive failure. Alexa, a voice commerce solution in search of a problem, has been running Amazon 7-8 figure losses.

Anthropic’s hilarious Project Vend — in which they put a Claude instance in charge of a vending machine — was clearly a marketing stunt. “Claudius” took about a month to go bankrupt — hallucinating fake vendors, an identity (“blue blazer and a red tie”), and raging demand for “metal cubes” along the way. But it summed up nicely how not to attempt AI-automated commerce: ignore judgment abstraction, isolate the system from substantive sources of progressive learning, apply no deterministic guardrails, and focus on a preference-driven consumer purchase.

Dark Marketplaces are a B2B Game

Outside of a small minority of roles — fashion or art buyers, for example — business procurement is repeat, policy-driven, and margin-oriented. B2B buyers operate within procurement budgets, approved vendor lists, compliance constraints, and established reorder cadences. Their decisions are more abstractable than a consumer deciding between sneakers — there are more patterns to detect, more rules to encode, more operational data to learn from. A café ordering supplies every morning is a far better substrate for judgment abstraction than a consumer buying a birthday gift once a year.

B2B buyers already share operational data with platforms. ERP integrations, POS feeds, inventory APIs — the data-sharing and integration work required for judgment abstraction is table stakes in B2B, not a privacy negotiation. Even if it’s much more complex and often difficult to capture, the data exists and there is significant incumbent plumbing to leverage.

And once an agent absorbs a buyer’s operational heuristics — substitution tolerances, timing patterns, supplier preferences, risk appetite — the relationship becomes the moat. Ripping out the agent means losing institutional memory. Switching costs compound with every interaction, deepening the moat automatically over time. We explored this dynamic in our emerging playbooks piece through the lens of authoring layers and systems of intelligence: the system that captures the most workflow data eventually becomes the system of record.

Dark Marketplaces take this one step further — the system that captures the most judgment data becomes the system of action. When that action anchored in a profit motive, rather than an experience one, incentives towards pure automation align.

How “Dark” Transforms Marketplace Fundamentals

The classic framework for marketplace success — drawn here from the team at NEA and from Jonathan Golden’s work at Airbnb — centers on three demand-side enablers: Discovery, Convenience, and Trust. On the supply side, the analogs are Utilization, Revenue, and Convenience. These create flywheels that drive platform metrics: fill rate, time-to-purchase, demand retention, AOV, take rate, leakage.

To be clear, when we say “Dark Marketplaces” we mean all forms of transactional platform — not necessarily marketplaces in the traditional business model sense. That said, many of the same drivers and metrics will apply to Dark Marketplaces. And many of these, in turn, will be fundamentally transformed (or made obsolete) by agentic automation.

Discovery becomes elimination.

In a traditional marketplace, discovery is the primary value proposition — aggregating fragmented supply, optimizing the buyer’s ability to find the best counterparty. In a Dark Marketplace, the buyer doesn’t discover the supply. The agent knows what the buyer needs, finds it, evaluates it, and either presents a recommendation or completes a transaction. Discovery friction approaches zero. Jonathan Golden described how reducing “cognitive load” in heterogeneous marketplaces was the key challenge for buyer conversion. Dark Marketplaces solve this by removing the human from the discovery loop entirely.

Convenience becomes invisibility.

NEA framed convenience as a “utility leap forward” — making it significantly easier for both supply and demand to get on-platform and transact. In a Dark Marketplace, the leap is from “easier” to “invisible.” The transaction happens in the background. The buyer’s first instinct is to check a notification from their agent, not to open a browser and start browsing a feed of SKUs.

Trust shifts from perceptual to empirical.

In traditional marketplaces, trust is built through reviews, brand reputation, fulfillment reliability, and return policies — all signals designed to reassure a human making a judgment call. In a Dark Marketplace, trust attaches to the agent’s track record. Did it save money? Avoid stockouts? Handle exceptions gracefully? Choose the right substitute? Trust becomes measurable and continuous, not a one-time assessment at the point of purchase.

These transformations mitigate the classic marketplace failure modes — poor discovery, insufficient convenience, and lack of trust. But they introduce a new one: judgment drift. If the agent makes a few bad calls — wrong substitution, an overstock, a missed timing window — the user overrides it, stops trusting it, and re-engages manually. Maintaining judgment accuracy across changing conditions, evolving preferences, and edge cases becomes the new retention metric. This is why the engagement axis matters so much: the more continuously the system observes, the faster it corrects, and the less likely judgment drift becomes.

Not all startups looking to transform the fundamentals are on equal footing. Vertical AI and vertically integrated players have an unfair advantage here. In some markets, incumbency may also offer an edge. GrubMarket is an interesting example here — with ~$680M raised and a $3.5B valuation, they’re 12 years old. But that time in market has allowed them to build a lot of leverage. They’re both the marketplace and the supplier: running warehouses and distribution across all 50 states while selling WholesaleWare, an AI-powered ERP, to third-party distributors. Last year, it launched purpose-built AI agents for inventory, reporting, and monitoring. Because GrubMarket controls supply, demand, and the intelligence layer, its agents train on both sides of every transaction. The endgame: wholesaler agents auto-negotiate with grower agents and GrubMarket captures the spread.

The Dark Horses of Autonomous Commerce

Across the landscape of funded vertical AI companies, a handful seem to be approaching the Dark Marketplace threshold from different angles. Though none, of course, has arrived at the eventual end-state: true AI-automated, human-free commerce. Below are the companies we find most instructive, organized by their current phase of judgment abstraction.

So how does competition play out? Like in traditional marketplaces and commerce, there will be 800-pound gorillas thriving alongside vertically focused, even niche players. While we strongly believe AI-nativity is going to key to velocity on the Dark Marketplace approach, the optimal angle of attack is something we’re still watching closely. Let’s consider freight brokerage. Augment ($110M, Redpoint-led) embeds deeply into one side of the workflow — full order-to-cash automation across $35B in freight under management. FleetWorks ($17.5M, First Round-led) is two-sided from day one — its AI dispatcher serves both carriers and brokers, with 10,000+ carriers and Uber Freight already on-platform. An open question: does depth-first (focus on building AI into a single ICP on one side of the network) or breadth-first (start with both sides already present, progressing toward agent-to-agent clearing eventually) reach Phase 4 faster?

Takeaways for Founders Building Dark Marketplaces

As we wrote in Software Is Dead — Long Live Software: “The immutable primitives of software defensibility are workflow and data. Speed is a compounding advantage in the early days of a category, but it is not a durable moat.” In the Dark Marketplace context, this means the wedge — the voice agent, the auto-reorder, the procurement bot — is necessary but insufficient alone. We explored why in our essay on the Dispatcher Problem:

An AI Service company whose primary value is “we deliver this service cheaper by using LLMs” is a dispatcher sitting on a proprietary margin advantage that doesn’t belong to them. It belongs to the cost curve of inference — and that curve is controlled by model labs, hyperscalers, chip manufacturers, and energy producers. — The Verticalist

The escape route from the Dispatcher Problem is judgment data. Per-user, per-context, accumulated over thousands of interactions — the kind of data that gets more valuable the longer the relationship lasts, and that a competitor can’t replicate by plugging into the same model API. Five principles for founders building toward this:

Start with the wedge that maximizes engagement and proximity

If you have to choose one, lean engagement. Retrofitting ambient data capture is much harder than extending a product toward a transaction. Voice AI, conversation capture, and workflow-embedded tools are better wedges than dashboards or analytics — they generate the behavioral data that judgment abstraction requires. For Dark Marketplaces, the authoring layer is the engagement surface.

Design for per-user judgment capture, not aggregate preferences

The Dark Marketplace advantage is that every user’s agent is different — trained on that user’s specific behavior, context, and edge cases. Build structured memory, per-user context retrieval, and feedback loops from day one. These aren’t features to add later; they’re the architecture. The per-user fine-tuning challenge is real — latency, cost, and negative artifacts grow with context window size — but the approaches emerging around memory layers, retrieval-augmented generation, and parameter-efficient adapters are exactly the right toolkit.

Pursue B2B verticals with repeat purchasing and fragmented supply

Food distribution, freight, construction materials, dental supplies, specialty pharma, auto parts — verticals where buyers make dozens of decisions per week, and supply is heterogeneous enough to justify intermediation. These are the markets where judgment abstraction produces the highest ROI: enough decision volume to learn quickly, enough supply complexity to create real value, and enough repeat behavior to compound switching costs. Also, learn the lesson many B2B Marketplaces have over the last decade: if you don’t understand why brokers and distributors exist in your vertical, there’s a good chance you misunderstand their service role or leverage point.

Plan the journey from Stage 1 to Stage 4

Don’t build an autonomous agent on day one — you’ll lose trust before you’ve earned it. Build a system that captures stated preferences, earns the right to infer behavior, proves it can handle contextual judgment, and only then operates autonomously. Each stage is a trust-building exercise with the user base. Trying to skip stages is how you get judgment drift, manual overrides, and churn.

Remember that the moat is memory, not UI

In a Dark Marketplace, the interface is fungible. What matters is the accumulated knowledge of how this buyer decides — substitution tolerances, timing preferences, risk appetite, supplier relationships. That memory is the switching cost. Invest in it like infrastructure, because your competitors will.

The Invisible Hand, Revisited

When Anthropic ran Project Deal, the detail that stuck with us wasn’t the number of deals or the dollar volume. It was what happened with the weaker model. Participants represented by the less capable AI got worse outcomes, but had no idea. They couldn’t tell their agent was underperforming because they only saw the result, not the process.

This is the core tension of the Dark Marketplace. When the transaction goes dark, the quality of the agent’s judgment becomes everything. A great agent saves money, avoids stockouts, finds better suppliers, and handles exceptions gracefully. A mediocre one makes quiet mistakes that compound over time. And the user can’t tell the difference until the damage is done and trust is gone (likely forever).

That’s why judgment abstraction is the moat, the product, and the risk all at once. The companies that will build the next hundred-billion-dollar marketplaces won’t win on merchandising or slick UX. They’ll win because they deeply understand their customers and build systems that think like them, reliably. And of course, because they succeed in eliminating market inefficiencies, deadweight loss, time lost, and human mistakes across trillions of transactions.

You may be familiar with Adam Smith’s “invisible hand.” Contrary to the popular mythos, the metaphor was meant to illustrate not the universal efficiency of markets, but rather that self-interested choices by market participants, collectively, can benefit society. The “hand” of the market is invisible because it’s defined by choices buried in the heads of billions of buyers and sellers. When the buyer’s judgment is freed from the confines of their head and gut — abstracted into AI that acts continuously, autonomously, and at a scale no human could manage — the potential is immense.

The marketplace doesn’t disappear. It just goes dark.

Thanks for reading The Verticalist!

Euclid is an inception-stage VC built for Vertical AI founders. If anyone in your network is considering building in Vertical AI, we’d love to help. Just drop us a line via the comments below or on LinkedIn.

Recent seismic shifts in the public markets backdrop adds urgency to this consideration of software (and AI) defensibility. Forward PE multiples for public SaaS companies compressed below the S&P 500 for the first time in the modern era. What’s happened? The market isn’t just repricing growth expectations — it’s asking a structural question about which software businesses have earned the right to compound for another decade. Whether via real math or just vibes, the terminal values of software titans are being questioned with the incredible emergent growth of AI-native disruptors. Some legacy SaaS have earned the right to continue existing and are oversold. Others — as Chegg demonstrated — can see their core value proposition open-sourced overnight by an LLM.

Morgan Livermore — a 15-year enterprise and infrastructure investor with a background at Accel, Geodesic, and Quiet Capital before founding SuperCruise Capital — is watching this play out from the growth stage. In today’s episode of Verticals, he shares his view on what’s going on in software, what that means for defensibility in the era of AI, and why that led him to launch a new firm.

Watch the Full Episode Here

This episode is brought to you by Parafin — they power embedded finance & capital for platforms like Amazon, Gusto, and DoorDash. If your platform serves SMBs, Parafin’s white-labelled lending, cards, and insurance, can grow your revenue & retention. Learn more →

Morgan’s view: “I would much rather own a 20 to 30% compounding software company for the next decade than certain other more legacy businesses.” So would we — but only if the compounding assumption is backed by something more durable than a feature lead. For founders, the takeaway is concrete: know where your defensibility lives, and build toward it from day one. Speed still matters at inception, but the window in which growth alone was enough has closed.

His framing is blunt: if five competitors can build something quickly and cheaply, “the chance that they figure out those workflows is pretty high.” The implication isn’t that defensibility is dead. It’s that the altitude at which it operates is changing. Feature-level differentiation degrades faster than ever. What endures sits either deeper in the technical stack or deeper in the domain.

As an investor in both vertical and horizontal markets — with a deep background in infra investing — Morgan highlights the growing importance of technical differentiation. For example, he shares a portfolio company of his that’s winning thousands of “customers” that aren’t human — they’re AI agents. Claude, GPT, and other coding agents are autonomously discovering and adopting the product because its documentation and API surface are the most machine-readable in the category. Naturally, that requires a completely new stack and a reimagination of what scalability means.

“We have the best documentation, agents can know what we do, understand it,” the founder told him. “And because of that it’s actually started compounding.” The company reportedly has tens of thousands of agents on its platform, none acquired through a sales call. This is infrastructure defensibility rewritten for the agentic era — discoverability, developer experience, and docs as distribution. It echoes the open-source playbooks of Vercel, Confluent, and HashiCorp, but with a twist: the “developer” evaluating your product may not be human anymore.

That said, Morgan agrees with the second note of defensibility: domain expertise. As the AI stack stabilizes — more reliable models, better tooling, clearer architecture — understanding how to apply commoditized stacks to take advantage of unique market surface areas is increasingly valuable. Consider voice AI intake: hundreds of startups can answer calls for dentists and dealerships using commodity models. But as we’ve explored in Voice-First Playbooks in Vertical AI, the wedge is not the business. The founders who understand which scheduling edge cases cause no-shows, or how a specific ERP integration unlocks procurement data, are the ones building platforms behind the commodity wedge. Domain expertise lets you anticipate the next product, not just replicate the current one — and that’s where durable vertical moats get built.

As we catalogued in Dude, Where’s My Moat?, the immutable primitives of workflow and data take many forms — but technical and domain edges are becoming ever more important. Because the speed at which surface-level advantages erode is increasing rapidly with AI, those that don’t find a truly defensible one — whether disruptor or incumbent — may fall into another class of multiple premium altogether. Veeva doesn’t charge what it does because its interface is irreplaceable. It charges what it does because its understanding of life sciences workflows and regulatory data is — and its product demonstrates that. Depth — whether technical or domain — is what separates compounders from commodities.

Hear Morgan walk through the full case for efficient growth and shifting defensibility in software — catch the full episode on YouTube and wherever you listen to podcasts.

Subscribe to Verticals to get new episodes every week, available wherever you watch or listen.

Youtube • Spotify • Apple • Amazon

Cents just closed a $140M Series C: the largest software investment — probably the largest investment period — in the laundry vertical’s history. Alex isn’t from the space. He dropped out of college to build his first company, Ulyngo, a payments platform for higher ed that he sold at 23. During his earnout, he stumbled into laundromats for a different reason: looking for a good cash-flowing business on the side. But he couldn’t get over the insanity of the technology gap — 70–80% still coin-operated — and decided it the opportunity was just too glaringly obvious. He went full-time when Bessemer first invested in 2021. The thesis that got him to where is today is one investors have rejected for years, but yet has created one of the biggest vertical winners in history (Toast): hardware-first vertical software.

Watch Full Episode Here

This episode is brought to you by Parafin — the embedded capital platform powering financial products for platforms like DoorDash, Amazon, and Worldpay. If your platform serves SMBs, Parafin lets you offer lending, cards, and insurance without building the infrastructure yourself. Learn more →

Alex built Cents by explicitly rejecting the Toast growth model (give away hardware, hire 1,000+ salespeople, optimize NRR on a venture timeline) and instead building around the actual purchasing dynamics of his industry: hardware-led land, CS-heavy expand, distributor-driven GTM. Every time Cents tried to look like Toast, they missed plan. When they built for the industry they actually served, they hit it. The evidence: more CS reps than salespeople, CAC-to-LTV ratios “in the thousands of X” on hardware+software, 99% retention, $140M Series C.

Cent’s hardware bet worked because it unlocked three compounding advantages:

1. Payments Volume

   On-machine devices capture the majority of a location’s transactions, not just the wash-and-fold counter.

2. Retention
   Removing a POS is easy, but removing hardware wired into every machine in a store is structurally painful. Cents reports 99%+ customer retention.

3. Distribution
   This one was less intuitive. Cents’ hardware is sold exclusively through laundry equipment distributors, the same trusted advisors who’ve had deep, multi-year relationships with operators. Those distributors wouldn’t even take Alex’s call before the hardware acquisition. After it, they became his most productive sales channel — now selling Cents software alongside the machines. “Friends want friends to make money,” Alex said. “Distributors want us to make money. Operators want us to make money.”

Hardware-as-wedge isn’t going to work in most verticals — at the least, its more nuanced outside of point-of-sale environments. But it’s interesting to consider how owning the hardware can foster a durable moat in the AI era. When horizontal AI voice companies tried to sell into laundromats, they hit a wall: they didn’t have context on which machines were available, what inventory looked like, or how a self-serve customer paid. That context lives in Cents’ hardware layer — and it’s what powers their emerging agentic product suite, from automated marketing campaigns targeting lapsed customers to AI-driven store management. As we wrote in The Disapatcher Problem, the startups that endure are those embedding themselves so deeply in customer operations that switching becomes structurally painful, not just inconvenient. Hardware is the most literal expression of that principle.

The broader implication for vertical founders: in a world where intelligence is commoditizing, the scarce asset may be the physical and operational layer that feeds it. That depth of integration into a customer’s workflow — capturing data, controlling equipment, processing payments — creates a substrate that no software authoring layer can easily replicate.

For the full tactical breakdown — including Alex’s distributor playbook, why he has more Customer Success reps than salespeople, and how he reverse-engineered his Series C valuation for investors — watch the full episode.

Subscribe to Verticals to get new episodes every week, available wherever you watch or listen.

Youtube • Spotify • Apple • Amazon

While this has always been an important question in vertical software, it’s become existential in the era of AI. As AI-native vertical solutions are growing with SaaS-defying velocity, incumbents are considering some permutation of the options:

• Build internal AI teams

• Acquire emergent AI point solutions

• Cut off startup access (APIs / integrations) to their platforms

The founders who play this right can turn their incumbent relationships into an advantage — those that play it wrong risk getting steamrolled.

Alex Niehenke, Partner at Scale Venture Partners, has spent over a decade backing vertical startups, predominantly at Series A and B. Many of his portfolio companies — like Motive, Dusty Robotics, and Root Insurance — have faced the incumbent question. In today’s episode, Alex joins us to share his view on how Vertical AI startups should position themselves to win in markets with a handful of dominant players.

His framework: build a “system of love,” the system customers actually want to use. In industries where incumbents have jacked up prices for years — throwing around their weight to maintain control despite rock-bottom NPS — winning the early, fervent loyalty of your customers is critical. Because likely, you share those customers with an incumbent, and that may be your only leverage in an eventual, unfair match-up. On this episode of Verticals, Alex breaks down how Vertical AI founders should fight back, with some important nuances along the way.

We also revisit the “Death of SaaS” question, the milestones Series A investors are looking for in Vertical AI, and industries Alex would love to see more AI-native startups building in. Check out our episode here, and our breakdown of Alex’s founder playbook below.

Watch full episode here

This episode is brought to you by Parafin. At #15 on the Inc. 5000 and >$100M in revenue last year, they are flying. Makes sense when you realize they power customer financing offerings for the likes of Amazon, DoorDash, Kajabi, Gusto, and Worldpay. If you’re building a vertical platform and your customers need capital — they are a name you should know.

Alex’s Journey to Vertical AI Investor

Alex joined Scale in 2012 — back when vertical software was the lonely end of the dinner table. The perception was boring markets, tiny TAMs, cruddy buyers — and that turned out to be wrong. His partners once passed on a construction SaaS company out of Santa Barbara: “ten billion dollars later, we got Procore.” The seismic shift, in his view, came when foundational models spooked horizontal investors into the verticals. “Every venture firm suddenly has a vertical microsite,” he says.

He sees a structural reason Vertical AI stands out for VCs in the current era: investors are running from the risk that frontier models swallow horizontal software and running toward verticals where regulation, proprietary data, and workflow depth offer real moats. But Alex does warn that AI services without deep product embedding face a brutal commoditization curve. “Your margin is my opportunity. There’s always going to be some new YC kids who’ll do it cheaper.”

Check out the episode for a deep dialog on all of the above. Specifically, we thought his view on competing with legacy incumbents as a Vertical AI startup warranted a highlight — continue reading to get our breakdown of his top tips.

Vertical Playbook

How to Compete with Legacy Incumbents

Step 1: Build a System of Love

Incumbent ERPs in markets like legal, insurance, and construction have been running an inflation-plus pricing strategy for 15 years with no alternative in sight. Customers are furious. “Do you know who law firms hate? The traditional vendors — the Westlaws, LexisNexises.” When Harvey shows up and asks what problems they have, the floodgates open. Win on love and NPS, not contractual lock-in. “Think about where entrenched, miserable incumbents exist. That’s probably not a terrible framework.” The one true point of leverage you have when it comes to unfair fights with massive legacy incumbents is your shared customers — you want them to love you so much, they raise a fuss if and when the SoR makes integration hard.

Step 2: Build a Standalone Path Around the ERP

If a dominant ERP controls 40%+ of your market, a cutoff could wipe half your ARR overnight. The counter-move: make your product usable without the direct integration. Alex’s advice is to “vibe code that traditional system of record and roll it out” — build around them so that even if they turn hostile, your customers can still use your product and will be angry if it disappears.

Step 3: Don’t Litigate — Outrun

The Abridge/Epic standoff, Trunk Tools vs. Procore, Cumulate vs. AMS360 — Alex has watched these collisions up close. His verdict: avoid the courtroom. “The US litigation system’s timeline and costs don’t align with what you’re trying to solve at a startup level. You are litigating with somebody who has unlimited time, unlimited money, and they’re going to want to air out all your dirty laundry.” Channel the energy into shipping product instead. There’s growing precedent that the tide may be tipping against incumbents in the courtrooms — everyone should read about RTMS v. PointClickCare — but you don’t want to be an undercapitalized guinea pig.

Step 4: Give Up a Pound of Flesh If You Must

Alex invokes Dollar Shave Club: the viral hit that suddenly needed razor supply from one of five factories in the world — most owned by competitors. Founder Michael Dubin flew to Asia, gave up margin to secure supply, and preserved the momentum that ultimately led to a billion-dollar exit to Procter & Gamble. Sometimes, you negotiate with the ecosystem rather than fight it. Secure distribution while you still have velocity.

Step 5: Partner with the #2 and #3 ERPs

If the dominant system of record turns hostile, lean into the runners-up. Embed with them, co-sell with them, give them a reason to feature you as their innovation story. As Nic put it on the pod: “Go to the 2 and 3 and really try to have an embedded partnership with them” — at least until you’ve built enough platform defensibility to stand alone.

The Takeaway for Vertical Founders

The incumbents most vulnerable to disruption aren’t just bad at product — they’re often structurally incapable of responding. PE-owned, over-levered, run by hired-gun CEOs whose comp packages are designed not to cannibalize the base business. Alex is blunt: “Shame on those incumbent companies for being as shitty as they are.” But the reality is, that’s just how the incentives are stacked. That vacuum is exactly where the next generation of vertical platforms will be built — by founders with the conviction to build systems of love where love has been absent for a long time. And in addition to building a great product, success requires survival.

Don’t forget to subscribe to our show, Verticals. Get a new dive deep into Vertical AI strategy with a top founder or investor, every Wednesday — available wherever you watch or listen:

Youtube • Spotify • Apple • Amazon

While we agree there are many appealing features of the model, we are likely more cautious than our peers regarding the scope (which markets this applies to), the addressable market (how large a business you can realistically build), and venture fundability (the proportion of AINS startups that are a good fit for venture capital). In fact, we have been sharing our concerns with the growing enthusiasm for services for almost two years. From our piece in November 2024, The Siren Song of Services:

As with any venture thesis in vogue, however, an antithesis is looming. Today, we are seeing a broad embrace of startups tackling services, many without clear paths to automation or without a natural venture-scale second act. Some bet that AI will catch up, and margins with it. Others trade away the software business model wholesale in exchange for TAM, taking vertical integration to its extreme. Far too many are forgetting why software is such a hallowed category of investment, making all-too-liberal assumptions around the ease of automating humans out of an industry, the down-market valuation of services revenue, or the right to grow a product suite horizontally. — The Verticalist

The Case for AI Services

The pitch for AI-enabled services is rational. The logic is inextricably tied to the changes LLMs have enabled, generally running as follows:

• Due to various factors — including the rapidly increasing number of SaaS tools and incentives against innovation in some industries — the traditional angle of attack (selling and monetizing cloud software) has stalled.

• To expand their reach, startups need to shift from SaaS to services, focusing on the rapid improvements in LLMs to deliver outcomes rather than just tools and, in some cases, completely taking over the operating businesses they previously considered customers.

• By including labor budgets in scope alongside traditional IT budgets, TAMs for AI Services startups are massively expanded. Moreover, in markets with labor supply constraints, AINS are particularly compelling because they can deliver immediate revenue acceleration by removing the bottleneck.2

• With the potential of LLMs and agentic platforms to boost automation and increase gross margins in traditionally manual workflows, AI Services are now venture-backable, venture-scalable, and have software-like terminal margin potential.

A common feature of such arguments is the tendency to treat all AI-native service models as the same. In practice, we see two distinct models emerging. They are two fundamentally different strategies, with different prospects for all downstream considerations that matter: go-to-market, margin structure, hiring, and defensibility.

1. Services-as-a-Wedge

   Here, you enter a market by taking on work outcomes like NDAs or insurance brokering. Often, these are already outsourced to consultants or BPOs, for example. Services are the initial beachhead; you use them to build trust and gather data, then gradually automate more with software (internal or external) to control the workflow and tap into insourced TAM. Over time — if the hypothesis is correct and the AI-powered workflow improves sufficiently — the need for the human service layer recedes. This is what people generally mean by an “AI Services startup” — or at least, it was until recently.

2. Services-as-Delivery-Engine
   The second archetype contemplates a very different business model. Workflow software is developed, but software alone cannot deliver the complete customer outcome. Human service remains essential at the last mile: filing with regulators, interfacing with high-value customers, approving items requiring certified experts, etc. In this model, the service component is a permanent delivery mechanism, not just an entry point. As AI advances, the service becomes more efficient, but the underlying need for human judgment in the loop remains.

These two approaches are likely to yield fundamentally different business models. It’s not just that they have different leverage points around human-in-the-loop and ultimately margins. It’s also that they are built to serve two different types of products, which in turn are often built to serve two different types of markets. A fund administrator or financial auditor, for example, is a very different type of services firm than an insurance or freight broker. Let us explain.

Services-as-a-Wedge relies on “intelligence arbitrage.” Its fundamental bet is rapid AI improvement, allowing what was once a low-margin service to become a software-like business with a low marginal cost of growth. In contrast, Services-as-Delivery-Engine leverages “coordination advantage,” where advances in the AI model drive a product superior to the status quo, even if the margin profile never looks SaaS-like. The former transitions quickly into additional services before it becomes commoditized, while the latter strengthens its advantage through automating exception handling and coordination.

Comparing the two models

The wedge approach is effective in categories where work is already outsourced, tasks are intelligence-based and can be automated, and buyers are comfortable with external providers. Here, substitution involves changing vendors rather than reorganizing internal processes.

• Insurance brokerage exemplifies this model. Standard commercial lines brokerage involves comparing carriers and completing forms, which is primarily intelligence work. The distribution layer is highly fragmented among many small brokers, so no single incumbent controls customer relationships. Harper and Panta are recent examples of this model.

• Transactional legal work fits the same pattern. NDAs, regulatory filings, standard contract drafting — well-defined tasks with standardized outputs that most companies already outsource to outside counsel. Crosby and Arcline enter as the autopilot, replacing counsel on the straightforward stuff.

In delivery models, the primary challenge is not intelligence work but coordination, compliance, liability ownership, and the physical-world handoff between the software-powered output and the customer’s desired outcome. The service provider is effectively an embedded partner and owns the outcome. Owning the outcome means the last mile is structurally lengthy, at least today. The work is typically outsourced, but can also be insourced.

• Communication compliance is a strong example of the delivery engine model. Firms in regulated industries, including financial services, healthcare, and government contracting, must monitor, archive, and report employee communications. Hadrius and Norm AI are addressing this need in financial services. While the intelligence work can be automated, achieving the desired outcome requires ongoing regulatory filings, audit readiness, and liability for missed violations.

• Fund administration is another strong example. Running NAV calculations is intelligent work that the model can handle. But the outcome the GP is paying for is auditable books, trusted LP reporting, and a system of record that holds the fund’s entire operational history. The promise of AI-native fund administration is an ERP-like stickiness with a service-layer last-mile. But no matter how good the ERP is, the coordination, regulatory compliance, and judgment layers are critical to the business's success. Hanover Park, Maybern, and Formulary are recent startups to emerge in this category.

In delivery models, the primary challenge is not intelligence work but coordination, compliance, liability ownership, and the physical-world handoff between the software-powered output and the customer’s desired outcome. The service provider is effectively an embedded partner and owns the outcome. Owning the outcome means the last mile is structurally lengthy, at least today. The work is typically outsourced, but can also be insourced.

The Diagnostic: Three Questions

We suggest that founders aiming to build an AI-services business should answer three questions to define which model to pursue:

1. Are you an external vendor or an embedded partner?

If service is your entry point, you are entering the market by performing outsourced work, replacing another vendor, and following the wedge strategy. Your priority should be to progress toward workflow ownership before the intelligence layer becomes commoditized. If service is your delivery mechanism, providing last-mile coordination as an embedded partner adds value to the software; your focus should be on deepening the surface area of this capability while pursuing automation.

2. Who owns the outcome?

In the wedge model, the customer typically retains liability; you’re doing the work, but they’re responsible for the output. The executor of the NDA is liable. The customer still signs off on the insurance policy. In the delivery-layer model, you own the liability, even if it’s implicit. The human judgment layer is not going away when filing the tax return. Administering the fund and closing the books with your name on the audit trail. Liability ownership sounds like risk (as it should be!), but it’s actually leverage when executed correctly to build a durable moat. It converts cost-center pricing into insurance pricing and creates switching costs that no feature comparison can replicate.

3. What happens to your business when the model gets 10x better?

For wedge companies, a better model makes your entry point easier, but also makes everyone else’s entry point easier. The advantage is temporary unless you’ve compounded data and trust into internal workflow ownership. For delivery-layer companies, a better model makes your service layer cheaper and faster to deliver, but the coordination and compliance requirements don’t go away. Your margins should improve. Your moat gets wider. If a 10x better model helps your competitors as much as it helps you, you’re in the wedge. If it disproportionately helps you because you’ve built the coordination infrastructure to exploit it, you’re in the service layer.

The worst position is misdiagnosing which model you’re in. A wedge company that thinks it has a service-layer moat will get commoditized while congratulating itself on its service quality. A service-layer company that tries to shed its coordination layer to look like a pure software business will gut the very thing that makes it defensible.

The Challenges with Services

The Allure of Trading Margins for Revenue

It’s easy to see why services present such an alluring target. Large prospective markets are up for grabs within the arbitrage window as intelligence capabilities accelerate. The first pillar of that argument is that selling the work, not the tool, can be the right approach to wedge into software-shy buyers — opening doors where SaaS struggled. Replacing an existing vendor is a simpler buyer consideration than training on a new software platform. There’s simply less organizational change required, which can be substantial in industries where something has been done one way for decades. The second pillar of that argument is the fact that, as the Sequoia piece put it, “For every dollar spent on software, six are spent on services.” Hypothetically, 100% of the TAM is up for grabs, rather than just the IT budget (more on this later).

As we wrote in our piece on the dispatch problem earlier this year, while we are generally bullish on AI services, we do not believe services are the right path for every vertical market, nor will they be the most dominant approach overall:

Some believe that as the marginal cost of intelligence collapses toward zero, the key value proposition of enterprise technology will shift from providing tools that assist human labor to delivering outcomes that replace it. We believe this is certainly true: Vertical AI can handle more end-to-end workflows than Vertical SaaS alone. This warrants significantly more customer value and willingness to pay, tapping into significantly larger budgets. We disagree, however, with the now-prevalent view that service delivery — a customer relationship that mirrors an external vendor rather than an internal platform integral to the business — is the prevailing paradigm of AI-powered software.
— The Verticalist

The most glaring question for us is that the trade-off between revenue (and, at a higher abstraction, addressable market size) and margins is not as clear-cut a bet as it might seem. We think Emergence’s concept of “Mirage PMF” nicely encapsulates several of our concerns here, namely that “strong revenue growth and net dollar retention can mask a lack of true AI enablement.” Historically, many tech-enabled service businesses have stormed out of the gate with traction (and even strong retention), trading margins for revenue growth, only to have unit economics and cash burn be their undoing at scale. The rise and fall of several well-funded freight brokers and real estate iBuyers are great examples of this phenomenon.

Of course, the key now is that, with AI, this time it’s different. In our past essay on the AI-First Roll-Up, we addressed this specific point:

[It] remains unclear what the margin lift from AI will be—and opportunity across industries is unlikely to be universal… [Service-as-Software companies are] selling an outcome and internalizing the margin risk. I.e., they are making a bet that they can automate manual an increasing share of work over time. These have been famous last words for many startups whose margin uplift materialized too slowly because it was (i) too hard technically, or (ii) too hard to escape reliance on services once addicted to the high growth rates their venture backers demanded. — The Verticalist

The last point is a unique and difficult challenge for tech-enabled services and can often be highly destructive to enterprise value. Managing the trade-off between growth and per-unit delivery costs presents idiosyncratic challenges for a venture-backed startup: investors paying software multiples will expect venture-like growth with software-like margins. For the operators, that means rapidly growing top line while simultaneously improving gross margins and reducing per-unit costs. There reaches a point of maturity where you can no longer trade margin for revenue, and the market is unlikely to reward those who trade margin over revenue.

The startup needs to invest in growth (sales & marketing), automation (R&D), and service delivery (human labor as COGS) simultaneously, carefully pulling the levers to manage the business. Investing incorrectly and having to iterate rapidly and reprioritize is difficult for any startup, but especially for a services firm. In high-trust markets, the cost of failure can be existential. You’re delivering an outcome, not a tool. Understaffing and underdelivering as a services business can damage a brand beyond repair. Overstaffing and failing to grow can make raising that next round even harder; meanwhile, those staffing choices increase burn and shorten the runway.

Even if the startup gets the ramping levels right, staying competitive in a market eager for AI-powered software won't be easy. Over time, the purely economic incentive to develop not just good software but industry-leading software diminishes—after all, your reward isn’t higher revenue growth as it might be with a startup; it’s gross margin improvement. Boring! It’s also not enough to help you raise your next round. No later-stage venture investors will fund a slower growth business just because they’re improving margins. Automation is table stakes, but venture funding won’t follow without venture-like growth and a path to software-like margins.

At the same time, not only will your automation likely be replicated by direct AI service competitors (who will aim to undercut your offering on cost & speed), but it will eventually be accessible to the broader market as pure software. The motivation to continue R&D efforts to build best-in-breed technology quickly diminishes.

The TAM Illusion

The largest challenge we have with the services narrative is around market sizing. Sequoia’s analysis suggests “the total addressable market for autopilots is all labor spend in a category, insourced and outsourced combined.” Of course, such proclamations are not unique. The graphic below is from a recent Coatue presentation, suggesting that “by shifting the unit of value from the tool to the results, the addressable market potential expands by 25x.”

These types of TAM analyses feel overly simplistic. We’ve previously discussed our view that VCs struggle with sizing vertical markets, but this kind of analysis, encompassing all labor spend, is not exclusive to vertical markets. As we explained in our piece on agents in Vertical AI, automation has not historically led to a one-to-one replacement of labor. Similarly, calculating TAM based on potential labor offsets is not only flawed—it’s fundamentally incorrect and incongruous with every historical example of automation.

Perhaps more importantly, if the intelligence layer continues to improve and become more affordable — as many companies building and investing in AI services believe — the value of the service-level output will become commoditized. In other words, if the cost of task-specific intelligence continues to decrease, the arbitrage window will eventually dissipate as the intelligence layer approaches infrastructure-style pricing (i.e., at margins above roughly the cost of delivering it). Consequently, any traditional labor-based pricing model will fail. Pricing will be driven not by how many people can be replaced and what they cost, but by market competition. Any service layer based on labor pricing will be unsustainable. And yes, while lowering the cost of a service could increase demand (see Jevons’ Paradox), this still doesn’t bring you back to the labor-offset TAM.

This also presents the classic dispatcher problem: performing the work without owning the workflow makes you a lower-cost version of the incumbent rather than a fundamentally different business. For example, taxi dispatchers coordinated rides but did not own the fleet, driver relationships, or customer data. When coordination became commoditized, dispatchers lost their advantage. The value accrued to the workflow owners.

If intelligence-as-a-service in a vertical market follows the expected path of commoditization, we can be certain the market will not be the same size as the current services market. It may be much larger in aggregate value creation, but the question is who captures that value — and historically, it has not been the company whose only advantage was doing the work cheaply.

In other words, the vertical AI startups — whether they consider themselves AI services companies or not — that reach early traction by capturing the surplus from collapsing intelligence costs will quickly find themselves underpriced and outcompeted if they don’t build beyond the initial service. The only long-term moat comes from productization beyond the initial service:

The wedge may be a service delivered more cheaply. But the moat will be a system built on top of that wedge, leveraging their internal positioning with customers to develop moats—the proprietary data, the network effects, the multi-product platform, and the industry “brain”—that make it infrastructure to be relied upon, rather than just another vendor.
— The Verticalist

The other glaring question regarding potential outcome sizes is market penetration. From our piece on market sizing last year, horizontal SaaS companies often slow down around 10-20% market share. Vertical SaaS, especially in less competitive markets, may achieve much higher penetration — sometimes over 80% within focused (albeit often PE-consolidated) markets. Most large public vertical software businesses top out in the mid-teens.

Few operating businesses, and hence fully vertically integrated startups, will ever reach these heights. Despite their incredible growth, none of Headway, Alma, Rula, or Grow Therapy is estimated to have more than low single-digit shares of the total US therapy market. Even after its roll-up of the largest legacy player in the market, the vertically integrated home health business Honor is likely to have 2-4%. The TAM may be enormous, but the path to capturing a meaningful share of it as an integrated services business is structurally harder than selling software licenses. Combined with the deflationary pricing impacts of collapsing intelligence, we feel that many of our peers may be overestimating the size of the revenue opportunity that “labor-replacement” services can truly achieve.

Services Aren’t More Defensible Than Software

Sequoia opens its piece by arguing that AI Services are more defensible against the foundation model labs. “If you sell the tool,” Sequoia wrote, “you’re in a race against the model. But if you sell the work, every improvement in the model makes your service faster, cheaper, and harder to compete with.”

It's worth unpacking what they're actually saying. Most SaaS companies resell compute and storage infrastructure; as cloud infrastructure has improved and become more affordable, both they and AWS, among others, have benefited. Naturally, SaaS providers offering no unique value beyond storage and compute were inherently vulnerable and would eventually face disruption and eroding market share. As Ben Evans joked on a recent podcast interview, “[SaaS] is all a thin-SQL wrapper. They’re all just databases.” 

The same paradigm applies to solutions built on top of LLM infrastructure. Every improvement to the model benefits both tools and services equally. Sequoia’s framing assumes LLMs improve in ways that commoditize the tool but somehow spare the service. That’s arbitrary. If the model gets good enough to replace a code-review tool, it’s good enough to replace a code-review-as-a-service offering (and arguably, the latter is easier). What’s simpler: switching IT help desks or an accounting firm, or swapping out your ERP? Model improvements only make a service defensible to the extent that no competing service adopts the same LLM advances. That’s not a bet we’d take.

So why would selling the tool put you “in a race” against the model any more than selling the work does? The strongest moats in software have always been those that are not easily replicated — proprietary data loops, workflow integration, distribution, network effects — not the infrastructure they’re built on, regardless of whether you charge for the tool or the output. The threat vector is the same for both tools and services: competitors (whether peers or the labs themselves) could leverage LLMs to commoditize your offering. Whether you’re wrapping the model as a tool or using it to deliver completed work, you need to build a business that will survive and thrive as your infrastructure improves.

The Wedge is not the Moat

Our overall point is not that AI Services built on LLM infrastructures are inherently a bad initial wedge (or business model). Rather, we remain convinced that these startups must go beyond the wedge to build internal customer workflows and, in some cases, commoditize their own services to avoid the eventual disruption. For those building in the service-as-delivery startups, these businesses can be highly attractive, although it remains to be seen how many markets can support software-like margins and venture-scale outcomes.

At Euclid, questions remain whether the services-as-a-wedge model will actually prove to be a valuable long-term market entry point. Our friends at Tidemark have done great work on the patterns of control points in Vertical Software. Most vertical software markets have one or two control points that serve as ideal patterns for building strategically dominant Vertical Software companies. In a past piece, we shared what we believe is the optimal control point for most AI-native vertical products: an authoring layer (i.e., launch point) for a valuable internal workflow that can unlock revenue and deliver immediate ROI. For services-as-a-delivery startups, there’s a strong argument that their last-mile coordination layer can serve as a useful control point into larger work-streams and revenue streams. We are, however, more skeptical that services-as-a-wedge businesses can capture ownership of a control point.

We have shared many times in this piece (and others over the past year) that we continue to believe the dominant mode of “winning” in Vertical AI will be software — regardless of the way it’s positioned or by whom it’s used. There is also a subtler strategic point here that the services labor replacement discourse obscures entirely. As AI capabilities improve, human roles won’t disappear; they migrate. As such, we have argued that the largest outcomes in Vertical AI will not come from labor automation, but from enhancing productivity:

Much of the narrative around AI’s value has centered on direct labor replacement, given the immediate profit potential for buyers. But we believe AI’s greater impact—and the larger market opportunity—will be its ability to enhance workforce productivity rather than simply substitute for it. — The Verticalist

AI-Native Services will undoubtedly produce great business & venture outcomes. However, we believe the current discourse is approaching overexuberance, where founders and (some of their) backers view it as a cure-all for the classic adoption frictions in selling to vertical industries. There’s also the risk for those following the playbook of building a good services business, but one that’s not venture-scale, unfortunately, built with VC dollars and expectations. As Emergence eloquently puts it, the trap is “that you’ve built a good services firm financed with the wrong kind of capital.” This was a similar hypothesis we outlined in our initial piece on the AI-First Roll-Up. Neither roll-ups nor tech-enabled services are new concepts, and both models have been well-understood and financed by private equity for years.

The wedge that drives much of the current generation of Vertical startups is inexpensive intelligence. The trap for AI Services founders is mistaking an appealing wedge for a sustainable business. We can think of many businesses, both vertical and horizontal, pre- and post-AI, that stormed out of the gate with attractive wedges, raised at lofty valuations, only to later discover their products quickly commoditized and that there was no durable moat. Our friend Rick Zullo at Equal shared a strong articulation of this view:

Knowing how to win in these sectors isn't just about following the growth, it’s about navigating the industry dynamics to find long-term competitive advantage. Without that, you'll have a lot of well-funded leaky buckets. As with all services companies, the value comes from long-term FCF and we're seeing plenty of companies giving away a dollar for 20 cents to grow (and even worse, some lie about their accounting to raise capital). That's not the game we're playing.

The companies that will last are those that capitalize on the current opportunity, when cost differences are high, adoption is still early, and incumbents are slow to embed themselves deeply in their customers’ operations. The latter outcome makes switching structurally painful rather than just inconvenient. And unlike traditional software, AI-native platforms can integrate into every workflow, collect data from all interactions — whether a human is involved or not — and develop growing intelligence that improves the product over time. The opportunity to build “load-bearing infrastructure” has never been greater. Neither has the temptation to settle for being a “cheaper vendor.” 

We’ll end this piece with a historical analogy that bears some resemblance to the current moment: the rise of the “Fab Five” internet consulting firms during the dot-com era. Viant, Scient, Razorfish, iXL, and MarchFirst grew rapidly by performing the outsourced intelligence work of their time, offering services such as website development, e-commerce implementation, and digital strategy advice. They reached multibillion-dollar market capitalizations and were the darling tech-enabled services companies of the era.

By 2001, all five firms had either collapsed or been acquired. The intelligence became commoditized; building a website shifted from specialized expertise to a task any competent developer could perform. Larger consulting firms developed these capabilities internally. The Fab Five lacked workflow ownership, a system of record, and switching costs. They were service providers that successfully rode the coattails of a hyper-growth technology shift… until it became widely replicable, and ultimately, commoditized. Their early brand halo was not enough to overcome the market shift and an increasingly unprofitable business model.

The Sequoia piece ends with a bold claim: “The next $1T company will be a software company masquerading as a services firm.” We’ll take the other side of that bet. The next $1T company won't be masquerading as anything. It will have software margins, software moats, and software defensibility, even if the “product” is outcomes rather than software. That distinction is everything.

Thanks for reading The Verticalist!

Euclid is an inception-stage VC built for Vertical AI founders. If anyone in your network is considering building in Vertical AI, we’d love to help. Just drop us a line via the comments below or on LinkedIn.

This brings us to the hard question for B2B2C founders: how do you prioritize and balance the “B” (business) and the “C” (consumer) in your model? In practice, trying to do both is how startups lose focus and alienate their core buyer. Like in any multi-sided vertical business model, you need to ensure you have at least one stakeholder prioritized at any point, and ultimately, locked in. Andres Robelo, Founder & CEO of Playbypoint, spent ten years learning this in racket sports — tennis, pickleball, padel — and built a playbook for navigating the tightrope.

Check out our latest episode of Verticals with Andres. And continue below to get our vertical playbook, inspired by the Playbypoint — Navigating Vertical B2B2C.

Watch Full Episode Here

A quick word from our sponsor on the Verticals podcast, Parafin. They’ve surpassed $100M in revs, extending >$25B in financing to their customers’ customers… B2B2C, even B2B2B. They power embedded capital for platforms like DoorDash and Jobber, and are purpose-built for vertical platforms. Click here to explore a custom program for your platform today.

A Pickleball Story: From Hobby to Obsession

Andres founded Playbypoint in 2016 after a consumer frustration: he couldn’t book a tennis court from his phone. His instinct was marketplace — be the OpenTable of courts. His first customer was a Miami municipal park, which took nine months of lobbying city commissioners for a free pilot.

The marketplace thesis broke fast. Booking was only 10% of a club’s operation. The supply side wasn’t digitized — membership management, lesson scheduling, junior academies, F&B — so there was nothing to aggregate. Playbypoint pivoted from connecting consumers to courts to building an end-to-end OS for the business. Today they serve over a thousand clubs across 37 countries, growing 80% year-over-year, backed by Blueprint Equity. Pickleball’s massive US explosion hasn’t hurt either.

The B2B2C Alignment Map

Most founders assume that more consumer orientation means a stronger B2B2C flywheel. But that’s not always the case — founders should instead look to understand what kind of market they’re playing in. Consider how “C” (consumer)-oriented you should be, based on how supply- or demand-constrained your “B” (businesses you serve directly in your vertical) is.

The strongest B2B2C roughly follow the diagonal — their consumer investment matches their market’s constraint. Zocdoc and Affirm are highly consumer-facing because their businesses are demand-constrained: providers need patients, buyers need financing. Consumer value and / or aggregation is the moat.

At the other end, Appfolio and Playbypoint are business-first because their markets are supply-constrained: courts are hard to scale, demand for housing is high. Our point is that your B↔C positioning should be informed by whether your vertical is supply- or demand-constrained. Andres understood this instinctively — here’s how he built a strategy around it.

The Vertical Playbook: Navigating B2B2C

Step 1: Pick a Side — and Mean It

A thousand clubs means millions of players, and the B2C pitch is sexier on a fundraising deck. Andres deliberately chose not to chase it. Shifting to a consumer model would put Playbypoint in competition with its own customers — a structural misalignment that erodes trust in SMB verticals. His positioning: if the platform enables operators to monetize and retain their players, the consumer value follows. The B sells the C. That’s the sequence.

Step 2: Is Your “B” Supply-Constrained or Demand-Constrained?

Most B2B2C founders assume the business needs more demand — so they build consumer-facing discovery and lead gen. Andres discovered the opposite. Racket sports clubs are supply-constrained: courts are physical real estate, and demand outstrips capacity. At 6:59am, 200 people are queued for a 7am booking window. So Playbypoint built demand-limiting tools — capping reservations per player, enforcing membership-tier access, managing waitlists. A Mindbody would never build these because the TAM looked too small. That specificity is what makes the product irreplaceable. On the alignment map, this is why Playbypoint sits in the bottom-left quadrant alongside AthenaHealth — and why it would be a mistake to push it toward the top-right by over-pivoting on the consumer.

Step 3: Use “C”-Side Data to Make Your B Smarter

Playbypoint‘s consumer layer generates enormous data: player preferences, booking patterns, churn signals, social connections. The key is routing that intelligence to the operator, not around them. Andres is now launching a marketing and retention kit — the company’s first true add-on — that uses player data to help clubs proactively retain members. AI collapses the build time (a matchmaking algorithm that would have taken months now takes days), but the value accrues to the B. That’s what separates a real B2B2C flywheel from a pass-through.

Step 4: Monetize Through the B’s Workflow, Not the C’s Transaction

Playbypoint is 80% SaaS revenue, 20% payments — the inverse of what most vertical SaaS founders target. Andres deliberately avoided over-indexing on payments because it’s a race to the bottom. SaaS stickiness built on deep workflow — the customizable rule engine hardened across a thousand clubs — is harder to dislodge. When a competitor recently offered clubs cash to switch, Andres didn’t match. He kept compounding domain depth. The switching cost isn’t the contract; it’s the thousand edge cases baked into the system.

The Takeaway for Vertical Founders

The racket sports market has been growing wildly. From 2.5M US pickleball players to over 25M in under a decade, with a new wave emerging in padel. Playbypoint caught that wave — but only after eight years of bootstrapped grind. But when the TAM exploded, Andres’ business / consumer positioning was well-oiled. His central lesson is B2B2C discipline: pick the B, understand its constraints, use the C to make the B smarter, and resist monetization shortcuts that look good on a pitch deck but erode the relationship holding the whole model together.

Thanks for reading the Verticalist! Subscribe to our show, Verticals, wherever you watch or listen. Dive deep into Vertical AI strategy with a new founder, every Wednesday.

Youtube • Spotify • Apple • Amazon

The data tells a clear story: across 4,395 VC financings totaling $186B in 2025, vertical startups captured 53% of deal volume and 30% of capital deployed. That share would be even higher if not for a handful of horizontal mega-rounds: OpenAI’s $40B, Anthropic’s $13B, plus several other GPU cloud and foundation model raises. Strip out the 12 deals sized $1B+ and vertical raises took in 51% of 2025 capital.

Some new trends emerged, while others held strong. Healthcare and Financial Services—no surprise—remained the twin pillars of vertical investing, combining for nearly 1,100 deals. 2025’s foremost emergent verticals, however, were Manufacturing & Industrials, Legal, and AEC & Trades. Manufacturing saw a 41% increase in deal count from Q1 to Q4. Legal produced two of the year’s largest rounds, with Clio ($850M) and Filevine ($400M): even if neither of these companies is AI-native, both are quickly becoming AI-powered as competition scales fast. And AEC & Trades saw both early and late stage enthusiasm, anchored by CompanyCam’s $415M round that made it Nebraska’s first unicorn.

On the exit side, 2025 delivered $131.1B across 158 vertical exits—a number skewed at the top by Ansys and a handful of mega-buyouts, but broad at the base: Healthcare alone generated 43 exits worth $17.1B, and the IPO window cracked open with 18 vertical offerings including Via and Hinge Health. Including horizontal deals, 2025 exits totaled $234.9B across 254 transactions—vertical companies captured 56% of total exit value. Excluding Ansys (the $35B acquisition that threw all the numbers off), vertical liquidity was almost exactly on par with non-vertical.

This report examines 2025 in full: deal volume and capital deployment trends, vertical-by-vertical breakdowns, stage dynamics, top investors, and geographic concentration—plus profiles of twenty thematically notable vertical deals to know. We conclude this report with Euclid’s perspective on the undercurrents and structural shifts we believe will define Vertical AI in the year to come.

I. Financing Overview

The US and Canadian VC market produced 4,395 software and AI financing deals of $1M+ in 2025 (excluding non-relevant deals), deploying $186B in aggregate capital. Deal volume across quarters was relatively even: Q1 and Q2 combined for 2,271 deals (52% of the year), while Q4 saw a cooldown to 1,010 deals—the lightest quarter and an 11% decline from Q1.

Capital deployment told a different story. Q1’s $60.5B was inflated by OpenAI’s $40B Series F—strip that out and Q1 looks closer to $20B, making the quarterly trajectory even more—a pattern consistent with our thesis on vertical AI market sizing. Vertical companies captured 19% of Q1 capital, then surged to 34% in Q2 as non-vertical mega-rounds dried up, before settling at 32% in Q3 and climbing to 42% in Q4. The full-year split—$56.2B vertical versus $129.8B non-vertical—still understates the vertical story thanks to a small handful of foundation model and infrastructure rounds (OpenAI, Anthropic, GPU cloud companies) in the non-vertical capital pool. On a per-transaction basis, verticals averaged $24.0M in deal size versus $63.1M for non-verticals, reflecting the latter’s concentration in capital-intensive model training and infrastructure.

The Vertical Share Is Growing

By deal count, vertical’s share held steady in the low-to-mid 50s: 52% in Q1, 53% in Q2, 53% in Q3, and 55% in Q4. The Q4 figure—more than half of all VC deals targeting a specific industry—represents the year’s high-water mark and suggests the vertical share is still expanding.

This wasn’t a sudden shift. In 2024, vertical applications already led deal count—especially in exits, where over 80% of PE-backed software acquisitions targeted vertical companies. But 2025 confirmed the trend quantitatively: the funding market is reorienting around industry-specific applications, and the Q4 acceleration suggests this isn’t cyclical.

Deal Size Distribution

The 4,395 deals break down into five size buckets that reveal the market’s shape. $1–5M deals dominated at 1,733 transactions (39%), the earliest-stage formation layer. $5–15M rounds accounted for 1,357 deals (31%), the first scaling tier. $15–30M and $30M+ rounds made up 591 (13%) and 506 (12%) deals respectively, while $100M+ mega-rounds totaled 208 deals (5%).

Vertical companies held the majority at every deal size except $100M+, where the split was 45%/55%—essentially a coin flip driven by infrastructure and foundation model mega-rounds. The most vertical-skewed size tier was $30M+ at 58%, with $1–5M close behind at 54%. This makes intuitive sense: vertical companies build moats through industry-specific data, integrations, and domain expertise that compound over time, making them fundable at scale. At the $100M+ tier, infrastructure and horizontal AI companies (GPU clouds, foundation model labs) absorb capital at a rate that no individual vertical can match.

Quarterly momentum within size buckets tells another story. $100M+ rounds grew from 39 in Q1 to 55 in Q2, then held at 57 in Q3 and Q4—remarkably stable after the initial ramp. But the $1–5M tier—the new company formation layer—dropped from 503 in Q1 to 362 in Q4, a 28% decline. This could signal a tightening at the earliest stages, though Q1 2025 may have been inflated by deals delayed from late 2024.

II. Vertical-by-Vertical Analysis

Not all verticals are created equal. Healthcare and Financial Services are GDP-dominating, highly diverse markets with hundreds of deals each, while Agriculture and Life Sciences remain more niche. The more revealing analysis is relative momentum—which verticals are gaining share, and where is the capital concentrating?

Healthcare (597 deals, $11.8B)

Healthcare was the most active vertical by deal count and the also the largest vertical by capital at $11.8B (slightly ahead of Financial Services at $11.7B). The stage mix skewed early: 38% of Healthcare deals were in the $1–5M tier, and the median deal was $6.8M. The big stories here were clinical AI documentation (Abridge’s $316M Series E, Judi Health’s $400M Series F), value-based care infrastructure (Strive Health’s $550M Series D), and the steady drumbeat of digital health buyouts—Modernizing Medicine ($5.3B), CentralReach ($1.9B), and VaxCare ($1.7B) were among the largest healthcare exits. Healthcare also led vertical exit count with 43 transactions worth $17.1B.

Financial Services (501 deals, $11.7B)

FinServ was the second-largest vertical by count and nearly tied with Healthcare by capital at $11.7B, exemplifying the embedded fintech thesis we outlined in “Vertical FinTech.” ($500M). The vertical also produced two IPOs: Figure Technology Solutions and Wealthfront. Deal activity was remarkably consistent quarter to quarter, suggesting structural depth rather than hype-driven surges. Ramp’s back-to-back Series E rounds ($200M in June, $514M in September) at a $22.5B valuation punctuated the year, alongside Plaid ($575M) and Wealthsimple ($536M).

Emerging Verticals

Manufacturing & Industrial — 110 deals, $8.6B

Deal count rose 41% from Q1 to Q4 (22 to 31 deals), and the vertical produced the year’s second-largest financing—Project Prometheus at $6.2B. Jeff Bezos’s return to operational company-building with a “physical AI” thesis for chip packaging, automotive assembly, and aerospace is a strong signal. On the exit side, Synopsys’s $35B acquisition of Ansys was the largest vertical exit of the year. Manufacturing had the highest share of post-2022 companies (57%) of any major vertical, confirming that AI is creating a new generation of industrial startups.

Legal — 97 deals, $3.3B

Two of the year’s 15 largest vertical rounds were in legal—Clio’s $850M and Filevine’s $400M. Both are case management platforms adding AI capabilities, and both are scaling aggressively through acquisition (Clio’s $1B vLex acquisition, Filevine’s push into corporate legal). 43% of Legal deals were in the $1–5M tier, with the category gaining more share of early-stage vertical deal count from 1H to 2H than any other. Despite the maturity, a wave of new AI-native legal startups is entering the market.

AEC & Trades — 106 deals, $2.1B

The trades—construction, roofing, HVAC, plumbing—are a classic vertical software opportunity: fragmented end markets, low technology adoption, high workflow complexity. CompanyCam’s $415M round and $2B valuation (making it Nebraska’s first unicorn) is the poster child. ServiceTitan’s December 2024 IPO opened the aperture — and set the stage for a pipeline of formation-stage talent. Deals were evenly distributed across quarters (with special strength at the $5-15M deal size), suggesting sustained interest.

Other Verticals of Note

Retail & CPG — 181 deals, $2.4B

Retail saw a notable front-loading of deal activity, dropping from 53 deals in Q1 to 39 in Q4. Capital deployment showed a similar trajectory, declining from ~$700M in Q1 to ~$400M in Q4. This likely reflects both macro headwinds in consumer spending and a shift in investor focus toward B2B applications. The exits, however, were healthy: 15 transactions worth $2.7B. On the financing side, Retail gained more share of early-stage vertical deal count than any category save Legal.

Supply Chain & Logistics — 133 deals, $4.2B

Supply Chain showed continued resurgence in 2025, coming off a venture-cycle low in 2022. The per-deal average of $31.6M reflects several large growth rounds in fleet management, warehouse robotics, and freight intelligence. Q1 was the strongest quarter at 36 deals and $1.0B in capital, with a steady decline through Q4 (24 deals). The exit environment was active: 12 transactions worth $5.8B, including several large PE-backed logistics technology acquisitions.

Energy & Climate — 60 deals, $965M

Climate tech’s intersection with AI remains nascent but promising. The 60 deals included both pure-play climate software (carbon accounting, grid optimization) and AI-for-energy applications. The vertical’s modest capital figure reflects the prevalence of early-stage deals—most Energy & Climate companies are still raising sub-$15M rounds. But the opportunity is real: as AI’s energy demands grow, expect a feedback loop between energy tech and compute to drive larger rounds in 2026.

III. Stage Dynamics

The $1–5M tier—the earliest-stage layer—dominated every vertical, but the concentration varied meaningfully—a pattern we’ve explored in “What’s My Stage Again?”. Education had the highest $1–5M share at 56%, followed by Real Estate and Hospitality and Agriculture at 50%. Healthcare, by contrast, had one of the lower early-stage shares at 38%, reflecting the category’s relative maturity—more companies are raising larger rounds as they scale.

The $30M+ and $100M+ tiers are where vertical differences sharpen. Financial Services had the highest $30M+ share at 16%, reflecting a deep bench of growth-stage fintech companies. Legal’s 9% $100M+ share—the highest outside of Financial Services—reflects outsized Clio and Filevine rounds. AEC & Trades showed an unusual distribution: 42% of deals fell in the $5–15M bucket (the highest of any vertical), suggesting the category may be moving beyond the earliest stages of vertical AI formation and into more mainstream legibility.

One of the most telling stage signals is the “newness” of each vertical’s company base. We looked at the share of funded companies founded in 2022 or later—the post-ChatGPT cohort. Manufacturing leads at 57%, followed by Public Sector (56%) and Legal (53%). These are verticals where generative AI created net-new startup formation, not just incremental improvement to existing companies. The cross-market baseline sits at roughly 46%, meaning these three verticals have meaningfully higher shares of post-ChatGPT companies than the broader market.

The deal size distribution also reveals which verticals have deep, multi-layered ecosystems versus those still in their earliest innings. Healthcare’s balanced profile (38/31/15/13/3) is the hallmark of a mature vertical with companies at every lifecycle phase. Financial Services is similarly broad, with the highest $30M+ concentration (21%) of any major vertical—the fintech wave of 2019–2021 produced companies now reaching growth stage. Conversely, Real Estate’s 51% concentration at $1–5M and 0% at $100M+ signals a vertical still in early formation: lots of new entrants, few scaled incumbents.

The quarterly stage data surfaces one additional insight: the $100M+ tier grew from 39 deals in Q1 to 55 in Q2 and held at that level through year-end. This suggests that mega-round activity—often viewed as frothy—actually reflected persistent demand at the growth stage, driven in part by sovereign wealth funds, asset managers, and crossover investors entering AI for the first time. The smaller deal tiers, by contrast, showed more cyclical behavior, with the $1–5M bucket declining 28% from Q1 to Q4.

A few additional analyses help us shed light on how vertical companies deploy capital across sectors and stages. Looking at average deal size per employee (FTE)—a proxy for capital intensity—Aerospace & Defense leads ($476K), reflecting the R&D-heavy nature of the category. Manufacturing & Industrial follows at $421K, consistent with the physical-world infrastructure requirements of industrial AI. Education, raising on average less than $200K per employee in 2025 rounds, was the leanest, consistent with its lighter distribution model and lower customer acquisition costs (and perhaps also speaking to lower investor demand, despite great rounds from breakouts MagicSchool and SchoolAI).

Median valuation step-ups—the post-money markup from one round to the next—reveal which verticals command the strongest investor conviction. Aerospace & Defense leads overall at 2.15x, followed by Manufacturing at 2.08x and Real Estate at 1.96x. At the early stage ($5–15M tier specifically), Education shows the highest step-up at 2.5x. Financial Services, despite attracting the most capital, shows a more modest overall step-up at 1.80x—suggesting that larger round sizes dilute the per-round markup even as total value creation remains strong.

Capital velocity—measured here by median dollars raised per year since founding—aims to capture the appetite for capital vertical startups demonstrate as they scale. Life Sciences, unsurprisingly, leads at $4.8M per year. Closely following—perhaps somewhat counter-intuitively—are Media & Entertainment at $4.5M and Agriculture at $4.2M. Education trails at $1.6M, consistent with its smaller round sizes and more measured growth trajectories.

Capital Efficiency—measured here by median dollars raised per year FTE—captures how much leverage a business has created on its financing to date. While imperfect—given it does not capture the ebbs and flows of spend over time—we can see that Life Sciences dominates, with A&D and Agriculture a close 2nd / 3rd. In comparison to Capital Intensity, this suggests greater investment requirements over time rather than immediate investor demand. As with both Velocity and Intensity, Education trails at just $342K raised per employee.

IV. Momentum: Stage-Level Trends

While the aggregate vertical share—53% of deals, 30% of capital—is a useful headline, we wanted to drill down into comparative dynamics at the stage level. By segmenting deals into five size buckets ($1–5M, $5–15M, $15–30M, $30M+, $100M+) and tracking vertical share by quarter, a more nuanced picture emerges: vertical AI has decisive early-stage and Series A / B momentum, but has yet to generate the volume of mega-deals that allow it to consistently maintain parity with horizontal at growth.

The $1–5M tier—the domain of pre-seed and seed—showed the strongest vertical momentum of any bucket. Vertical share rose from 53% in Q1 to 60% in Q4, an 8-percentage-point gain that represents the most significant forward-looking signal in this dataset. When three out of five new venture-backed companies are building for a specific industry, it says something about where founders see opportunity. The post-ChatGPT generation is not building general-purpose tools—they are building AI for healthcare, legal, construction, manufacturing, and logistics.

Moving to seed / Series A: the $5–15M tier ranged from 47% in Q1 to 58% in Q3, while $30M+ held between 56% (Q2) and 60% (Q4). Neither showed a clear directional trend. This stability suggests that once vertical companies graduate to their $5–15M+ round, the competitive dynamics at that stage are relatively balanced between vertical and horizontal players. The $15–30M tier was comparatively volatile—59% vertical in Q1, dropping to 41% in Q2, then recovering to 54% in Q4. The Q2 dip likely reflects a burst of horizontal AI activity versus any sort of vertical pullback.

The most striking divergence was at $100M+. Vertical share fell from 51%, an 18-percentage-point swing that represents the only bucket where non-vertical companies decisively overtook vertical during the year. This is almost entirely explained by the concentration of mega-rounds in foundation model companies, horizontal AI platforms (OpenAI, Anthropic, xAI, Databricks), and infrastructure providers that raised capital-intensive rounds in the back half of 2025. At every other tier, vertical maintained or expanded its lead.

Our momentum data produces a clean narrative: vertical making leaps and bounds at formation ($1–5M), holding steady through the scaling stages ($5–30M), and ceding ground only at the mega-round level ($100M+) where the magnitude of horizontal AI raises distorts the picture. For investors focused on the next cycle, the $1–5M signal is the one that matters most—if it takes 5–7 years for an early-stage cohort to reach maturity, the companies being formed today will define the vertical AI landscape of 2030–2032. We expect 2026 to be another year of prolific vertical AI formation.

V. Financing Highlights

Below, we pull out 15 highlights of vertical businesses that raised in 2025—we selected some from the biggest deals above, and mix in others we think are most interesting. Our profiles range in deal size from $190M (Peregrine) to by far the biggest deal at $6.2B (Project Prometheus). Healthcare dominated the list with five companies (in line with its 30% share of the top 20). Legal and Manufacturing & Industrial also placed multiple highlights. Five of the fifteen are first or second rounds, suggesting that new company formation at massive scale is accelerating in verticals where the AI opportunity is freshly legible. Here are the deals we found most revealing, and characteristic of 2025:

Project Prometheus — $6.2B (Mfg. & Industrial, Q4)

Jeff Bezos’s return to operational company-building made headlines, but the thesis is what matters: “physical AI” for manufacturing, targeting chip packaging, automotive assembly, and aerospace operations. Co-CEO Vik Bajaj brings a career that spans Google Life Sciences (now Verily), GRAIL, and Foresite Capital—an unusual background that fuses deep science with operational scale. The team has pulled over 100 hires from DeepMind, Tesla, and OpenAI. At $6.2B, this is the largest-ever first financing for a vertical AI company—and it’s not close. It’s larger than the next four deals on this list combined. The round is a statement that the next wave of AI value creation will happen on factory floors, not in SaaS dashboards, and that the market for physical-world AI is large enough to justify deployment-stage capital before a single dollar of revenue.

Clio — $850M Series G (Legal, Q4)

Clio’s round ($500M equity + $350M debt) funded the simultaneous $1B acquisition of legal research platform vLex, creating an end-to-end research-to-operations platform at a $5B valuation. Founded in 2007 by Jack Newton and Rian Gauvreau, Clio was the first to commercialize cloud-based law practice management when on-premises systems dominated. The combined deal represents both the largest legal tech financing and the largest legal tech M&A transaction in history. What stands out in the data: at 5x revenue and a 1.38x step-up from its prior round, this is a mature, capital-efficient business using leverage strategically—the $350M debt component signals confidence in predictable cash flows. With 1,672 employees and 17 years of compounding, Clio is the anti-hype case study: a vertical SaaS company that defined its category long before AI became a funding catalyst, and is now using AI-era capital to consolidate it. NEA’s Tony Florence led the round alongside Goldman Sachs, TCV, Sixth Street, and JMI Equity.

NinjaOne — $500M Series C Extension (ITSM, Q1)

NinjaOne’s $500M Series C extension at a $5B valuation—a 2.37x step-up from $1.9B just twelve months prior—is the kind of round that reveals the power of vertical focus in what looks like a horizontal market. Co-founded in 2013 by childhood friends Sal Sferlazza and Chris Matarese, the company built a cloud-native endpoint management platform for IT teams and MSPs at a time when incumbents like Kaseya and ConnectWise were running on legacy architectures. The raise, led by ICONIQ Growth and CapitalG (Alphabet‘s investment arm), also funded the $262M acquisition of SaaS backup leader Dropsuite—a buy-and-build move that mirrors Clio’s consolidation strategy. Notable in the data: NinjaOne is one of only two companies on this list based in the South (alongside Fleetio in Birmingham), it has zero debt, and it remains founder-led and controlled. With over 35,000 customers across 140+ countries and 2,000 employees, this is a vertical platform executing a horizontal expansion playbook.

Fleetio — $454M Series D (Supply Chain & Logistics, Q1)

Fleetio’s $454M Series D—co-led by existing investor Elephant Partners and new investor Goldman Sachs Growth Equity—valued the combined business at over $1.5B and funded the simultaneous acquisition of Auto Integrate, the premier maintenance authorization platform in North America. The combined entity now services over 8 million vehicles and processes more than 13 million repair orders per year through 110,000+ repair shops. Fleetio was founded in 2012 in Birmingham, Alabama by Tony Summerville, with CEO Jon Meachin (Brown undergrad, Morgan Stanley and Parthenon Capital alum) running the business. This deal stands out for its 2.36x step-up and the sheer physical scale of the network it acquired. While most vertical AI companies on this list are primarily selling software or AI, Fleetio is building a two-sided marketplace for fleet maintenance—a structural asset that gets harder to replicate with every shop that joins the network.

CompanyCam — $415M Series C (AEC & Trades, Q3)

CompanyCam is a story of pure founder-market fit. Luke Hansen built the platform out of his family’s roofing business in Lincoln, Nebraska, solving the problem of photo documentation and job site coordination for contractors. The $415M round from B Capital valued the company at $1.99B—Nebraska’s first-ever unicorn. At a 3.21x step-up, this is one of the highest valuation jumps on the list, suggesting rapid revenue acceleration. The investor roster reads like a conviction trade: Insight Partners, JMI Equity, WndrCo, and Blueprint Equity all participated alongside B Capital. CompanyCam represents the broader AEC thesis—fragmented industries with low tech adoption, where mobile-first AI tools can compress workflows that previously ran on clipboards and phone calls. With 308 employees, it’s also one of the leanest companies on the list relative to its valuation, implying strong unit economics.

Filevine — $400M Series F (Legal, Q3)

Filevine’s $400M Series F at a $5.25B valuation makes legal tech the only vertical to place two companies in the top six deals of 2025. Co-founded in 2014 by Nathan Morris, Filevine provides case management and workflow automation software for law firms, with a particular strength in litigation and personal injury practices. Accel and Insight Partners co-led the round alongside The Halo Fund. The 1.74x step-up is more modest than CompanyCam‘s or Ambience’s, but the absolute valuation—$5.25B on its 8th round—places it alongside Clio as proof that legal tech has matured from a niche into a major vertical software category. The deal is also notable for its investor composition: Meritech Capital, StepStone Group, and Album VC joined, bringing a mix of crossover and institutional capital that typically signals pre-IPO positioning. With 700 employees in Salt Lake City, Filevine is a reminder that vertical unicorns don’t have to be in San Francisco.

Abridge — $316M Series E (Healthcare, Q2)

Abridge raised its Series E just four months after a $250M Series D, reaching a $5.3B valuation with an implied EV/Revenue multiple north of 697x. CEO Shivdev Rao is a practicing cardiologist who co-founded the company with CMU researchers Florian Metze and Sandeep Konam. The platform transcribes and summarizes clinical conversations, producing documentation, patient summaries, and educational content. Andreessen Horowitz led the round alongside Khosla Ventures, WndrCo, and newcomers Archerman Capital and the California Health Care Foundation. What stands out: the 0.35-year gap between rounds is the shortest on the list, suggesting either extraordinary growth or a competitive capital raise intended to lock out competitors like Ambience Healthcare. At $7.60M in revenue and 488 employees, Abridge’s sky-high multiple reflects the market’s belief that clinical AI documentation is a winner-take-most category—and the willingness to fund the race at nearly any price.

Periodic Labs — $300M Seed (Mfg. & Industrial, Q3)

Periodic Labs emerged from stealth with what may be the largest seed round in history—$300M led by Andreessen Horowitz at a $1.3B pre-money valuation. Co-founded by Liam Fedus, a former VP of Research at OpenAI who co-created ChatGPT, and Ekin Dogus Cubuk, who led the materials and chemistry team at Google DeepMind (including the GNoME project that discovered 2 million new crystals), the company is building “AI scientists” paired with autonomous robotic laboratories. The investor list reads like a tech hall of fame: Jeff Bezos, Eric Schmidt, Jeff Dean, Elad Gil, NVIDIA’s NVentures, Accel, DST Global, Felicis, and more. The thesis is that the internet’s text corpus is exhausted as training data; the next frontier is proprietary experimental data generated by AI running physical experiments. Periodic is targeting superconductors first, with early commercial work on chip heat dissipation for a semiconductor manufacturer. Among the 15 deals, this is the most audacious bet on founder pedigree over business traction—a $300M check written on a company founded in 2025 with no revenue and roughly 100 employees.

Ambience — $243M Series C (Healthcare, Q3)

Ambience Healthcare’s $243M Series C at a $1.25B valuation represents a 3.47x step-up—the second-highest on this list behind Peregrine‘s 6.58x. Co-founded in 2020 by Michael Ng (MIT Sloan, ex-Morgan Stanley, ex-Calera Capital), Ambience builds an AI-powered medical scribe that captures clinician-patient conversations and generates clinical notes in real-time, embedded directly into EHR workflows. Andreessen Horowitz’s Julie Yoo and Oak HC/FT‘s Vignesh Chandramouli co-led the round, alongside a deep healthcare-specialist syndicate: Sequoia, Kleiner Perkins, General Catalyst, OpenAI Startup Fund, and Optum Ventures. This is one of two clinical documentation companies on the list (alongside Abridge), highlighting the intensity of competition in a category where switching costs are high and health system contracts are sticky. At 200 employees and four years old, Ambience is executing a faster land-and-expand playbook than Abridge, with institutional health system partnerships as its primary channel.

Forterra — $238M Series C (Aerospace & Defense, Q4)

Forterra’s $238M Series C, led by Moore Strategic Ventures, positions the company as the leading autonomous ground vehicle platform for U.S. defense. Originally founded as Robotic Research in 2002 by Alberto Lacaze, the company rebranded under CEO Josh Araujo (ex-Jefferies, ex-Lazard, ex-USMC infantry officer). Forterra builds self-driving military land systems and robotic swarm coordination platforms. The round came on the heels of a landmark streak of government contracts: the first-ever DoD ground autonomy production contract (ROGUE Fires with the Marine Corps), a $114M prime contract for autonomous breaching systems, and the Army’s GEARS and UxS programs. Forterra also acquired tactical mesh networking company goTenna, adding a communications layer to its autonomy stack. Among the 15 deals, Forterra is the oldest company by founding date (2002) and the most evolved from a business-model standpoint—it has real production contracts and program-of-record revenue, not just R&D grants. The investor syndicate (Salesforce Ventures, Franklin Templeton, Hanwha Asset Management, NightDragon) reflects the growing crossover between defense tech and institutional capital.

Teamworks — $235M Series F (Media & Ent., Q2)

Teamworks’ $235M oversubscribed Series F, led by returning investor Dragoneer, pushed the company past a $1B pre-money valuation—minting a unicorn in the sports technology vertical. Founded in 2004 by Zach Maurides, a Duke offensive lineman who built the first version as a sophomore-year class project to manage his own chaotic practice schedule, Teamworks has grown into what it calls “the Operating System for Sports,” powering over 6,500 elite teams globally across professional leagues, NCAA programs, Olympic sports, and military organizations. The round is a combination of primary and secondary capital. What makes Teamworks unusual on this list is its acquisition-driven expansion: the company has completed 13 acquisitions (nine in the past three years alone), including INFLCR (NIL content), ARMS Software (compliance), Smartabase (performance), Zelus Analytics (predictive intelligence), Telemetry Sports (computer vision for football, used by 80% of NFL teams), Opteamal (athlete monitoring), and most recently Sportlogiq (AI-powered hockey analytics, used by 97% of NHL teams). The Sportlogiq deal alone brought 80 employees and 10 AI researchers with 180+ published papers. This is a roll-up strategy executed with venture capital rather than private equity—rare in vertical SaaS. At ~450 employees across 11 countries and headquartered in Durham, NC, Teamworks is a textbook founder-market-fit story: Maurides understood the coordination problem in athletics from the inside and has spent two decades compounding that insight into a platform that spans talent acquisition, player development, game preparation, and operations. The $411M raised to date and Series F stage also make it one of the most mature businesses on this list.

Lila Sciences — $235M Series A (Life Sciences, Q3)

Lila Sciences’ $235M Series A (part of a broader $350M close that brought total funding to $550M) valued the company at $1.26B and represents the second “AI for science” bet on this list alongside Periodic Labs. Incubated at Flagship Pioneering—the venture studio behind Moderna—Lila was founded in 2023 by Geoffrey von Maltzahn, a Flagship General Partner with a PhD in biomedical engineering. Braidwell and Collective Global co-led, with NVIDIA‘s NVentures, In-Q-Tel (the CIA’s venture arm), Analog Devices, and institutional LPs including the Abu Dhabi Investment Authority and State of Michigan Retirement System participating. While Periodic Labs targets materials science through autonomous physical experiments, Lila is focused on life sciences and chemical discovery through what it calls “AI Science Factories”—autonomous lab facilities where AI agents design, execute, and iterate on experiments. In-Q-Tel’s participation signals national security interest in accelerating materials discovery for computing, energy, and infrastructure. At 195 employees, Lila has demonstrated early results including novel protein therapeutics, green hydrogen catalysts, and carbon capture sorbents.

OpenEvidence — $210M Series B (Healthcare, Q3)

We would have included OpenEvidence’s $210M Series B at $3.5B valuation on this list even if it hadn’t more than tripled its valuation with further raises in the interim. Founded in 2017 by Daniel Nadler—whose previous company Kensho was acquired by S&P Global for $700M in 2018—OpenEvidence built an AI-powered medical search engine that has been adopted by over 40% of U.S. physicians, processing 8.5 million consultations per month. Google Ventures and Kleiner Perkins (with Chairman John Doerr personally involved) co-led; Sequoia, Coatue, Conviction, and Thrive followed on. The data tells the story: at a 35x EV/Revenue multiple on $100M of revenue, this is the most capital-efficient company on the list by a wide margin. Its implied valuation-to-revenue of 35x compares to Abridge‘s 697x. The 3.35x step-up and a mere 36 employees make OpenEvidence the leanest operation here—fewer than one-fifth the headcount of most peers. The freemium model (free for verified physicians, monetized through advertising) bypassed traditional enterprise healthcare sales cycles entirely. Nadler’s TIME100 Health recognition and partnerships with the AMA, NEJM, and JAMA Network suggest a platform with both distribution and credibility moats. With only 0.4 years between its Series A and B, OpenEvidence is on a trajectory that has since seen it raise a $200M Series C at $6B and a $250M Series D at $12B.

Commure — $200M Series E (Healthcare, Q2)

Commure’s $200M Series E, led by General Catalyst, is the infrastructure play in an otherwise application-heavy healthcare contingent. Co-founded in 2017 by Tanay Tandon, Commure builds an interoperable data platform that unifies fragmented healthcare datasets—an integration layer that sits beneath the clinical applications companies like Abridge and Ambience are building on top. This is Commure‘s 9th round and the data shows a raised-to-date figure of $2.1B—by far the most cumulative capital of any healthcare company on this list. The 0.47-year gap between rounds suggests continuous capital deployment. With 492 employees in Mountain View, Commure occupies a different strategic position than the clinical AI documentation companies: it’s a platform bet on healthcare data infrastructure rather than a point-solution bet on any single clinical workflow. General Catalyst‘s Pranav Singhvi led alongside LTS Growth, Maverick Ventures Israel, and Shasta Ventures.

Peregrine — $190M Series C (Public Sector, Q1)

Peregrine’s $190M Series C, led by Sequoia Capital‘s Andrew Reed, valued the company at $2.56B—a staggering 6.58x step-up, the highest on this list by a factor of two. Co-founded in 2018 by Nick Noone, a former Palantir executive who headed the U.S. Special Operations (SOCOM) business unit, Peregrine builds a data intelligence platform for law enforcement and public safety agencies. The company was built while embedded inside the San Pablo Police Department, and its platform was deployed for Super Bowl LIX security coordination in New Orleans. The growth metrics are striking: revenue reportedly tripled from $3M to $10M in 2023, then tripled again to $30M in 2024. Agencies using the platform have reported measurable outcomes—a 40% decrease in open homicide cases in Albuquerque, a 21% reduction in violent crime in Atlanta. At 311 employees and just $63M raised prior to this round, Peregrine‘s capital efficiency is notable. The 6.6x step-up reflects both the growth rate and the scarcity premium of govtech companies that can actually land and expand across fragmented municipal and state agencies. Peregrine is the only public sector company on this list, and the only one with Palantir DNA in its founding team.

Themes Across the Top Deals

Four patterns recur across the year’s most notable financings. First, the convergence of AI and physical-world operations: Project Prometheus, Periodic Labs, Lila Sciences, Forterra, CompanyCam, and Fleetio all apply AI to industries where the value sits in atoms, not just bits: factory floors, autonomous ground vehicles, robotic labs, or repair shop networks. Physical-world companies account for a substantial share of the list.

Second, these highlights underscore the power of vertical founder-market fit. Shivdev Rao (cardiologist building clinical AI), Luke Hansen (roofer building construction tech), Jack Newton (17 years building legal tech), Zach Maurides (Duke offensive lineman building sports operations tech), Nick Noone (Palantir SOCOM lead building public safety tech)—most successful vertical AI companies are led by people who understand the industry’s workflows from the inside, or who bring a directly transferable prior success.

Third, the capital advantage of category creation: companies with a strong claim to dominating a core function or major segment within their vertical—Clio and Filevine in legal, Abridge and Ambience in clinical documentation, OpenEvidence in medical search, Peregrine in public safety intelligence—attract capital on fundamentally different terms than companies competing in crowded horizontals. Multiple deals on this list feature step-ups above 3x, a signal that investors are paying for category leadership and the expansive terminal value that can accompany it.

Fourth, a new theme emerged in 2025: the AI-for-science mega-round. Periodic Labs ($300M seed) and Lila Sciences ($235M Series A) represent a conviction that the next source of valuable AI training data isn’t the internet—it’s proprietary experimental data generated by autonomous laboratories. Both companies raised at $1B+ valuations with minimal revenue, staking their cases entirely on team pedigree and the structural argument that physical-world data is the new moat. This is a category that didn’t exist on last year’s list, and it may define the next cycle.

VI. Exits

The 2025 exit market produced 158 vertical transactions worth $131B, headlined by Synopsys’s $35B acquisition of Ansys. While M&A dominated with nearly two-thirds of transactions by number, liquidity generated by the category ($89.4B) trailed Buyout/LBO ($90.0B) marginally. The year saw vertical 18 IPOs, composing $4.3B in aggregate deal value.

Including non-vertical exits, the total market was $235.8B across 254 transactions—vertical companies captured 56% of total exit value. Excluding Ansys, vertical share of value was 49%.

The quarterly pattern was uneven. Looking at all deals: Q1 produced the most exit value at $73.8B (thanks in large part to the X merger), while Q3 generated the highest exit count at 74 transactions. Q4’s combination of 54 exits and $60.8B in value included the Groq acquisition and several large PE buyouts, suggesting the exit environment was accelerating rather than cooling as the year ended. Vertical share over the course of the year was volatile by quarter, swung one way or another depending on mega-deals or lack thereof. Vertical had a dominant share in Q2 and Q3 (73%, 89%), while capturing barely over a third of value (35%, 36%) in Q1 and Q4.

Manufacturing & Industrial led exit value at $67.3B, driven by Ansys ($35B) and Altair Engineering ($10.5B). Healthcare generated the most exit activity with 43 transactions worth $17.1B, led by Modernizing Medicine ($5.3B), CentralReach ($1.9B), VaxCare ($1.7B), and Iodine Software ($1.3B). Financial Services contributed 25 exits worth $14.1B, led by NinjaTrader ($1.5B), Enfusion ($1.5B), and Hidden Road ($1.3B).

The IPO Window

After a two-year drought, the IPO market reopened meaningfully in 2025 with 24 total relevant offerings, driving $12.3B in value. Including both vertical and horizontal companies: Figma ($1.2B) and Circle ($1.1B) led on the horizontal side, while vertical standouts included Figure Technology Solutions ($788M, Financial Services), Via Transportation ($493M, Public Sector), Hinge Health ($437M, Healthcare), and MNTN ($187M, Media & Entertainment). The diversity of verticals—and a strong thread of B2C usage / B2B monetization—suggests an IPO market with a high bar for traditional software or consumer tech business models.

Exit Path Trends

Buyout/LBO transactions accounted for 20% of exits but a disproportionate share of value. Silver Lake ($4.5B Altera), EQT ($3.0B Neogov), and TPG ($1.1B SynXis) were among the most active acquirers. This confirms a continuing pattern: PE firms view vertical software as ideal targets—recurring revenue, defensible market positions, and operational leverage from AI integration. Over 80% of PE-backed software acquisitions in 2024 were vertical; that trend appears to have continued through 2025.

The most revealing exit metric may be the simplest one: the median vertical exit in 2025 was approximately $500M. For the many seed and Series A funds that have scaled up AUM since 2021, driving superlative returns via such median outcomes demands discipline on entry price that seems quite rare these days. At a $10-20M post-money seed, a $500M exit represents 25-50x gross valuation appreciation—exceptional. At a $50M+ post-money “seed” (increasingly common in consensus Vertical AI), the same exit is 7-8x gross. Still great, but tough math for a >$100M AUM fund.

The top 10 exits by value in 2025 accounted for over 60% of total vertical exit value—a reminder that no category escapes power-law dynamics. Below the mega-deals, PE remains a leading exit path and IPOs remain selective — the “middle market” of vertical exits ($250–1B EV) is arguably the healthiest and most under-appreciated segment of the liquidity environment, at least in VC land.

Top Exits & Other Trends

The quarterly exit cadence showed meaningful variation. Q3 led with 74 exits and $72.2B in value, while Q4 produced 54 exits and $60.8B. There was a boom in Healthcare exits mid-year: 12 in Q2, 19 in Q3, but just 7 and 5 in Q1 and Q4, respectively. Financial Services saw its own 1H concentration, with 17 exits versus 8 in the back half of the year. Along with Ansys, Manufacturing & Industrial had a spike in Q3—with 6 exits, half of its 2025 deals closed in that quarter.

The exit size distribution is worth noting. The top 10 exits by value accounted for over 60% of total vertical exit value, reflecting the power-law dynamics inherent in software M&A. Below the mega-deals, the median vertical exit was approximately $500M—large enough to generate meaningful returns for growth-stage investors but small enough that PE firms remain the natural buyers. The middle market of vertical exits is arguably less dependent on public market conditions or strategic M&A cycles.

VII. The Investor Landscape

General Catalyst was the most active lead investor in vertical deals in 2025, leading 24 round representing $1.8B in value. Andreessen Horowitz followed closely at 22 leads over $1.9B worth of rounds, spanning Healthcare (Abridge), Legal (Clio), and Financial Services. WiL (19 leads) and Khosla Ventures (17 leads) rounded out the top four, with Khosla maintaining its earlier-stage focus at a $40M average deal size. Across the top 15, NEA’s average deal was largest by a wide margin at $154M, while First Round Capital had the lowest average deal at $11M.

At the $1–5M tier, Y Combinator was the most active at the earliest stage of vertical financing (7 leads), alongside Gradient (5) and HealthX (5). European-based funds also has a strong presence here, with Pi Labs and Nina Capital both clocking 4 leads.

The $5-15M band was anchored by First Round Capital with 9 vertical deals led in 2025, those startups capturing over $100M in aggregate. Next were Susa, Madrona, and Menlo, with 6 leads each, followed closely by Pear and Costanoa with 5 each.

In the $15-30M bucket, Base10 was the unequivocal vertical leader, with 11 leads at an average deal size of $18M. noa—the sole European investor in the top 15—was the next-active here with 8 leads. SignalFire—the third-place VC in this band with 7 leads last year—stood out with the highest average deal size of the cohort ($30M), and over $200M captured by its investees. Honorable mentions include Bonfire (4) and Building Ventures (4).

Moving to Series A and beyond—deals $30-100M in value—it’s perhaps surprising to see that a16z did not take first place in the vertical league table. That honor went to General Catalyst, with 24 leads (though a16z closely trailed with 22). Companies backed by these two firms raised $3.8B in total in 2025. The next tier of prominently vertical investors included Khosla (17), Insight (14), Bessemer Venture Partners (13), and Accel (13). While Sequoia may not have been as prolific this year (7 vertical leads), they registered the largest average vertical deals ($94M), with a16z and IVP close behind.

Late-stage vertical ($100M+) saw six multi-deal leads, all with a relatively tight mean deal size in the $110-180M range: NEA (10), Kleiner Perkins (7), Oak HC/FT (5), B Capital (5), DST Global (4), and Ribbit (3).

Concentration patterns within specific verticals are revealing. In Healthcare, a16z, General Catalyst, and Khosla accounted for a combined 18 of the top 50 lead positions. In Manufacturing & Industrial, a small number of growth-stage investors (including Coatue, Tiger Global, and GIC) drove a disproportionate share of capital through mega-rounds. Financial Services was the most fragmented—no single investor led more than 4% of deals, reflecting the vertical’s maturity and the breadth of sub-categories (e.g. payments, lending, insurance, wealth management, compliance).

The most popular verticals amongst this group of VCs was — no surprise — Healthcare (55) and Financial Services (38), together representing nearly half of the total deal lead count. Interestingly, Supply Chain and AEC & Trades were next, composing another ~14% of lead rounds. Most of the rest of the verticals fell in the 2-5% range, with Retail & CPG, Manufacturing & Industrial, Public Sector, Education, and Insurance clocking 5+ (and in that order of incidence).

Corporate venture activity in vertical AI also warrants mention. NVIDIA Ventures, Salesforce Ventures, and Google Ventures were active across multiple verticals. Perhaps more interesting was industry-specific CVC activity: health systems (Mayo Clinic, Kaiser) investing in clinical AI, financial institutions (Goldman Sachs, JP Morgan) backing fintech infrastructure, and energy companies (Shell, BP) funding climate tech.

A final interesting 2025 pattern was the (re-)emergence of non-traditional leads in big vertical rounds. BlackRock co-led Applied Intuition’s $600M round. Franklin Resources (Franklin Templeton) led Plaid’s $575M financing. Qatar Investment Authority participated in Applied Intuition. Mubadala co-led Crusoe’s $1.4B raise. While asset managers in pre-IPO rounds is familiar (reminiscent of 2021), the uptick of sovereign wealth involvement is perhaps more notable — especially in light of their parallel interest in backing US VC funds as LPs. Along with growing federal embrace of AI and defense startups in the US, the intersection of government and technology warrants continued focus in 2026.

VIII. Geographic Distribution

Vertical deal activity in 2025 was concentrated in established tech hubs but showed meaningful geographic breadth. California led with 667 vertical deals (29% of all vertical activity), followed by New York at 406 deals (17%). Massachusetts (119), Texas (110), and Florida (98) rounded out the top five. Washington, Canada, and Illinois each contributed 60–100+ vertical deals, reflecting the nationwide diffusion of vertical software.

The more revealing metric is vertical share by state. Among states with meaningful deal volume (30+ deals), industry-proximate states dominated: Arizona (71%), Ohio (70%), and Tennessee (67%) posted the highest vertical shares. Illinois (59%), Colorado (55%), and Florida (54%) all exceeded the 55% average reference line. The most striking finding: California, despite leading in absolute deal count, had the lowest vertical share of any major state at just 41%—reflecting its deep concentration of horizontal AI infrastructure, foundation model, and developer-tooling companies. New York (53%) and Massachusetts (54%) tracked near the mean.

Examining vertical share by region uncovers some additional interesting findings. The Midwest leads in terms of vertical proclivity (59.6% of its deals). It joins a relatively tight high-mid-fifties grouping with the Mid-Atlantic, Mountain West, and Southeast. Overall variance between regions is actually relatively low, with Canada and the West representing the two greatest points of contrast.

Canada’s lower vertical leaning is somewhat surprising, given Toronto’s outsized vertical presence (the home of Constellation Software since 1995). We believe, however, that the country’s industry-specific focus is concentrated in that one metro—and that much of its “vertical reach” involves investment in US-based companies.

Finally, it shouldn’t come as a surprise that the West trails in vertical share by region, with a paltry 41.3%. As a global Mecca of technical talent—frontier AI and otherwise—the Bay Area drives the predominance of infrastructure, dev tooling, and foundation model / lab financing. It’s important to remember, however, that the clear plurality of vertical deals are still inked in California (667 in 2025). New York follows behind, with 406 deals. These two are the clear leaders by vertical deal volume, with a yawning gulf between them and the next tier: Massachusetts with 119 and Texas with 110.

In our view, San Francisco and New York City—considering volume, concentration, and subjective deal quality—remain the twin dominant vertical hubs, with strong and growing tailwinds behind the latter. Non-coastal hubs have the highest verticality. But whether they are as quick to embrace AI-native approaches at the formation stage as the hubs remains to be seen.

IX. What to Watch in 2026

From our vantage point at Euclid, a handful of structural dynamics from 2025 set the table for the vertical VC and startup landscape in 2026:

The Seed-to-A graduation test

In 2025, the goalposts for Series A shifted. Seeds now look like what Series As did 5-10 years ago. The result has been widening gap between seed and A, with contracting graduation rates between then. Seed-stage vertical AI companies raised at a historic clip in 2023–2025, but many will hit this gap in 2026–2027: strong product, real customers, not yet at the milestones today’s “true Series A” investors require. The sectors with the highest formation rates (Manufacturing, Legal, Public Sector, AEC) may end up being the most exposed.

We expect capital supply to adjust, but unevenly, as capital supply (VC funds and allocators) responds. That likely involves a drawdown in the current over-concentration of capital at “mature seed,” where too many funds compete for the same $3–8M checks—that competition has led to GPs looking earlier to get their ownership, and thus too many companies over-funded for their stage. In 2026, we expect many of them to hit that wall—perhaps creating great opportunity for VCs with “tweener” strategies, but also creating significant risk for the tier of excellent companies that haven’t hit exit velocity prior to the few-million-ARR mark.

Deep tech’s maturity cycle

Project Prometheus ($6.2B), Figure AI ($1.5B), and Applied Intuition ($600M) collectively raised $8.3B in 2025 for companies building AI that interacts with the physical world. They aren’t outliers—Periodic Labs raised $300M for autonomous robot-driven research facilities, Thinking Machines Lab raised $2B for scientific AI tools, and Physical Intelligence pulled in $1.1B for robotic foundation models. In total, robotics startups raised over $6B in just the first seven months of 2025, with capital concentrating into fewer, larger rounds. Manufacturing & Industrial led in share of 2025-funded companies founded 2022+, with a 57% share. Figure alone went from a $2.6B valuation to $39B in 18 months, while its robots were still performing controlled demos at a single BMW plant.

There’s an important distinction between hardware companies and Vertical AI as we define it. While it may be an outmoded term, Vertical AI is defined by the software layer, operating on fundamentally better economics than hardware businesses: lower capital intensity, faster iteration cycles, higher gross margins, and the compounding moat of domain-specific data. Physical AI is a multi-trillion-dollar opportunity that extends across the economy and will change the world as we know it over decades. In the short term, “deep tech”—at least, the version that conflates hard with valuable—is looking overextended. In the long run, we’re bullish on the vertical software layer bringing AI to physical action.

Shifting winds of liquidity

With Blackstone, Thoma Bravo, Clearlake, and Vista all executing multi-billion dollar vertical buyouts in 2025, the PE-led exit environment for mature vertical SaaS companies is robust. That cycle’s health depends on interest rates and debt markets, but the demand side is proven—and we expect that to continue to ramp.

Perhaps the bigger question, pertinent to the health of the venture capital ecosystem, is public market liquidity. 2025’s IPOs were a start, not the floodgate-opening many are waiting for. The backlog of late-stage vertical companies—many of which raised $200M+ rounds in 2025—will likely face go / no-go decisions on public listings in 2026–2027. Next year specifically, the vertical IPO pipeline looks sparse: beyond EquipmentShare’s January debut, Plaid is the only highly-rumored candidate of note, although there are many potential entrants. Overall, we estimate the IPO narrative in 2026 to be dominated by horizontal infrastructure & AI (e.g. Databricks, Anthropic, OpenAI).

A clarion call for AI defensibility

The defining tension in vertical software today is breadth versus depth, and the extent to which that facilitates defensibility in a fast-moving era of AI—we touched on this in a recent essay, called “Software Is Dead — Long Live Software”. Horizontal AI tools are getting better at surface-level tasks across industries, which means vertical players need to focus on the depth advantage their industry focus often lends—the surface areas of workflow, regulatory, and data idiosyncrasies that makes each sector unique. Emergent leaders in verticals that were seen as big, attractive land-grabs for solution-first AI—Harvey in Legal, Rogo in Finance—will come under increasing fire as the AI labs take low-hanging-fruit opportunities to extend their core competencies to industry. OpenAI and Anthropic have already made overtures in those two verticals. And while we haven’t yet seen the canonical example of vertical SaaS market leaders making the leap to Vertical AI leader, legacy systems of record are certainly ramping up sharp-elbowed tactics—we expect both startup displacement and defensive M&A to ramp up dramatically. In any case, the Vertical AI startups that will define 2026 will need to take specialization and long-term moats seriously.

Report Methodology

This report covers 4,395 US and Canadian VC financings of $1M or greater in software and AI, sourced from PitchBook’s transaction database with supplemental company-level data from Harmonic. The dataset spans calendar year 2025 and includes closed deals (at time of close, not announcement).

Deals not primarily software are excluded from the analysis—specifically: pharmaceutical and biotech, pure hardware and device companies, four-wall retail and services businesses, and capital-intensive infrastructure plays where software is incidental to the business model. The line between “software” and “not software” is increasingly blurry in an era of AI-enabled services; our general rule is that companies deriving the majority of their value from a defensible software or model layer are included, while those primarily selling labor, physical goods, or regulated products are not.

The remaining universe is classified as either vertical (defined primarily by the end market) or not vertical (horizontal tools, developer infrastructure, AI infrastructure, general-purpose foundation models, and 95% of crypto/blockchain). We don’t make an explicit distinction between B2B and B2C, though B2C is typically only classified as vertical when it is a market network connecting vertical stakeholders (e.g. lawyers, architects) with consumers, or the product is monetized vertically despite consumer usage (e.g. payors, realtors).

Exit data covers 158 vertical transactions (254 total) and includes M&A, buyout, and IPO events. To maintain focus on liquidity-generating exits, the following transaction types are excluded: recapitalizations, PIPEs, public-to-private transactions, acquisition financing, asset acquisitions, secondary buyouts, and reverse mergers (with the exception of SPACs, although none were recorded this year).

We selected 18 representative verticals which, while imperfect, cover most of the universe. AEC = Architecture, Engineering & Construction. CPG = Consumer Packaged Goods. We occasionally refer to Manufacturing & Industrial as Mfg., and Aerospace & Defense as A&D.

Market data providers’ coverage of recent early-stage deals (pre-seed and seed) is inherently lossy. As a result, the true volume of late-2025 early-stage activity is likely higher than what appears here. This bias, if anything, likely understates the share of early-stage Vertical AI activity.

Finally, big shout out to Gotham Chi for their work sourcing and classifying much of the data behind this year’s report. Founded by University of Chicago students, they work with startups and VCs with research and analysis — reach out to them here if you’d like to learn more.

Thanks for being a subscriber and supporter of Euclid Insights! The Vertical Report is published annually by Euclid: a VC firm backing Vertical AI founders from inception. For questions, please reach out to the Euclid team via LinkedIn or DM here.

Lights dim for startup presentation #612.

A founder in a quadruple-oversized parachute pants saunters to the stage. She hits the clicker, and on the main screen, “Healthcare is dead” appears in all caps. “I was the reigning world Yu-Gi-Oh card champion from ages 8-14. Now, I’m dropping out of college to build the AI-native operating system for Coordination of Benefits in health insurance, a $4B problem.” 3 minutes later, the company raises $15M at a $60M post.

What’s wrong with this picture? We would argue, just one thing.

YC’s Great Inversion

Y-Combinator’s founding ethos was built around a specific kind of founder. Paul Graham launched the program in 2005 to back “younger, more technically oriented founders.”1 In a 2018 TechCrunch interview, Graham placed the ideal YC founder age in the late twenties, and the average founder age held steady around 29 from 2015 through 2022. Graham explicitly warned against funding overly young founders, calling it “premature optimization”: he argued they should accumulate real-world context before committing to build against it. “In college,” the thinking went, “you should be figuring out what the options are, not picking one option and running with it.”2

In recent years, YC has bucked its own traditional wisdom. By the end of 2024, the median age of cohort members had dropped to 24, down from 30 in 2022. As of last summer, acceptances of 18-22 year olds were up 110% year-over-year. 2025 was the first year in YC’s history that the majority of founders were 25 or younger.

While YC CEO Garry Tan positions it as “refocusing YC on its original DNA”, it’s in reality a bit of a departure — one perhaps best explained by his second rationale: the AI revolution itself. As Jared Heyman of Rebel Fund posited: “Since generative AI is such a new technology, younger founders are at no disadvantage understanding and building around it.” One could take this a step further and imagine that younger founders might be less hampered by traditional conceptions of software success, unbound to approach problems in fresh ways. It seems YC is betting so, with an average of ~1.5 years of experience in its most recent YC cohorts.

A study conducted in late 2024 by Data-Driven VC found that the average unicorn founder started their first unicorn at age 35. The median age was 33. Younger founders are overrepresented in unicorns relative to the general startup population (with < 30-year-olds punching above their weight ~2-to-1), and unicorn founders overall averaged 8 years of work experience.

The picture gets more nuanced when you look at what else changed alongside age. The Rebel data shows these younger founders are more technically credentialed than their predecessors. The “technical” cohort has dominated since 2023 — the year following ChatGPT’s launch. While YC always leaned towards engineers, the historical average share was a relatively steady ~61%. So the YC founder profile is markedly different in the modern AI era: younger, with dramatically less work experience, and more technical.

It’s easy to see the rationale for younger, more technical founders in this era of application-layer AI. The argument is twofold:

1. Technical: When a technology platform shift is first underway, the stack is highly dynamic and volatile. Therefore, those with the technical chops to navigate it will have an advantage. A 40-year-old coder with a decade-plus from FAANG may outstrip everyone classically, but the 23-year-old who’s been building with LLMs since GPT-3.5 has intuition for what these models can and can’t do.

2. AI-Native: In the dynamic environment above, founders more “native” to the stack may be free of constraints in thinking from older paradigms. The modern version of an old quote attributed to Henry Ford: “If ever I wanted to kill opposition by unfair means, I would endow the opposition with experts.”

Let’s examine each of these arguments to see if YC’s implied bets make sense.

The Technical Argument

Over the last 18 months, the founder-facing AI stack has become drastically more legible and accessible. Previously — from the release of ChatGPT in 2022 to arguably early last year — the founder-facing AI stack was a relative mess, requiring meaningful ML, data engineering, and infrastructure experience to construct anything non-derivative. During that 2022-2024 period, the model layer was much more of a monopoly; data management and retrieval were significantly more manual; and AI “agents” didn’t exist conceptually, much less have vetted frameworks for orchestration.

That legibility arrived in fits and starts. OpenAI’s Chat Completions API (early 2023) made models programmable rather than just promptable. Function calling (which followed shortly) in OpenAI was the first reliable primitive for connecting LLMs to external tools. Structured Outputs (August 2024) then pushed schema adherence from <40% on GPT-4 to 100%, compared to <40% with the prior GPT-4 model from OpenAI. It was a watershed moment for making the model layer a dependable building block for application-layer founders.

Another massive wave of interoperability and usability breakthroughs kicked off in 2024, with Anthropic’s launch of Model Context Protocol (MCP). Quickly adopted by OpenAI, Google DeepMind, and Microsoft, it was described by Bessemer as “the USB-C of AI.” It gave AI systems a universal spec for accessing external APIs, tools, and real-time data. Before MCP, every tool integration was custom plumbing that required a dedicated engineer. Now, AI agents — which in mid-2023 were still hallucinatory, token-burning science experiments — have production-grade orchestration frameworks built on top of a shared infrastructure with standard connectors. And the rest of the stack has followed, from RAG to vector databases, to deployment and visual AI pipeline editors.

Most important, of course, is the fact that LLMs themselves have made AI applications easier to write by leaps and bounds. By 2025, GitHub Copilot will have generated 46% of the code written by its 20M+ active users and made developers 55% faster in controlled studies. Cursor, its closest competitor, crossed $500M in ARR in 2025. A quarter of YC's W25 batch had 95% AI-generated codebases. And non-technical Claude Code and Codex vibe-coders have become so ubiquitous that it’s hard to keep up.

Our point is that over the last four years, the AI stack has become vastly more legible. Shawn Wang — aka swyx, former leader of developer tooling at AWS, Two Sigma, and three unicorns — was early to this thread of thinking in his now-storied essay, The Rise of the AI Engineer: “A wide range of AI tasks that used to take 5 years and a research team to accomplish in 2013 now just require API docs and a spare afternoon in 2023.” His implication was that the critical aspect of technical fluency in the applied AI stack is moving away from research and engineering, and toward product and design. Since 2023, our view is that this effect has only accelerated. Technical talent in application-layer AI remains important, and that will probably never change in early-stage startup builds. But hyper-focus on that one quality made more sense when the stack was significantly more nascent and less legible. The current leader of YC himself recognized this shift last year:

The ability to be successful is no longer limited by technical ability. The only thing that’s sort of the limit is can the founders get in the heads of customers. — Garry Tan

The Experience Argument

The other justification for a shift to younger founders is that they are less hampered by pre-existing norms related to either startup architecture or the industry status quo. Keith Rabois has been perhaps the strongest advocate of the anti-expert viewpoint over the last decade:

People with domain expertise learn what you can’t do, not what you could do. The most important companies are usually founded by people who don’t know much about what they’re getting themselves into. — Keith Rabois

The LLM era version of this argument goes further: AI itself has compressed the cost of iteration and the speed of learning to the point where deep industry experience is less important. As one commentator put it: “With ChatGPT or Claude, technical founders can now tap into domain expertise without needing any industry experience.” In a world where products can be developed at unprecedented speeds, perhaps velocity is becoming such a moat that it more than compensates for imperfect initial domain knowledge.

The challenge with this argument is that everyone is working with the same tools these days, and the markets one can identify as compelling from the outside are widely visible. What happens when every YC batch has three founders trying to build a copilot or voice AI intake solution for a particular industry? That’s approximately the situation we’re in today. We discussed the dilemma in a past essay, “The Dispatcher Problem”:

The result of cheaper AI is persistent deflationary pressure on Vertical AI offerings, which are predominantly attractive due to the consumer surplus enabled by LLMs. Pulling data out of documents? Answering inbound phone calls? Drafting perfunctory compliance reports? Products like these can be excellent wedges today when infrastructure and know-how are scarce, and adoption is low. Soon, they will be table stakes — as excess margin is competed away by several well-funded, credible, nicely growing startups in every category. Any startup that hasn’t developed a moat in the meantime will be a casualty.
— The Verticalist

How is one to develop a moat if you don’t have a unique, earned insight that allows you to see a problem and a consequent solution that others don’t? Of course, there will always be a portion of founders who have the drive, intelligence, and luck to navigate to the right outcome regardless. But generally, experience — if not true domain expertise — is the path to the “secrets” Thiel popularized in Zero to One. That is especially true in space, like Vertical AI, in which platforms embed within and even thrive on industry complexity, rather than abstracting it away as much as possible.

Perhaps more importantly, the data aren’t particularly supportive of the anti-experience camp. Returning to the case of YC: an analysis of their ~100 unicorn founders found an average of 8 years of work experience upon entry to the program. The founders who built Airbnb, Stripe, Coinbase, and DoorDash weren’t fresh out of school, as the majority of YC founders are today. Antler’s analysis of the wider AI unicorn founder pool found that, while they have gotten younger — the average age dropping from 40 in 2020 to 29 in 2024 — the mean work experience remained 8 years. Even Rabois himself acknowledges that, a preferable complement (at least in enterprise applications) would be pairing “a founder who’s very naive, very hungry” with “somebody who’s got a lot more experience.” Even if it’s just a few years at a high-growth startup, there is intuitive value in having at least enough experience to seed an earned insight.

When it comes to vertical platforms, the debate around the importance of founder domain expertise is particularly important. Not only for investors trying to understand what drives success and for future founders trying to plan their paths, but also as a lens into the core primitives of lasting success in business models. What backgrounds matter in vertical platforms? And how have LLMs changed that answer, as it pertains to the fast-growing world of Vertical AI?

What Backgrounds Drive Vertical Success?

The cleanest way to test whether domain expertise matters isn’t to ask VCs what they believe — it’s to look at where they actually deploy capital. Given the importance of this perennial question to what we do here at The Verticalist (and at Euclid), we decided it was time to rise above the anecdotes and generalist unicorn data points and run some serious numbers.

Over the last month, we conducted a comprehensive analysis of founder backgrounds to get to the heart of what drives success in the vertical. While we ran a similar analysis in 2024, looking at exited founders, this approach was a bit too retrospective, given the long timeframes for big outcomes in venture and the fast pace of AI. So this time, defined success a bit more broadly, setting our sights on high-velocity VC financings.

Our analysis contemplates founder backgrounds across 673 vertical software and vertical AI companies that raised rounds of $15M or more in 2025, classifying founding teams by whether they had prior experience in the vertical they’re now building for. And to speak to the Rabois case — pairing of “athletes” with “domain experts” — we consider all founders, not just CEOs. We traced every company that every founder had worked at prior to founding each startup.

Although there’s plenty to unpack, the findings aren’t ambiguous.

Overall, the two-thirds (66.1%) of leading vertical startups have at least one member of the founding team with vertical experience. That share is notably higher (~71%) in Vertical AI specifically. To further understand both the impact of vertical domain experience and its interplay with startup type (AI vs. SaaS), we can examine deal size, the quality of VCs leading the rounds, and “capital velocity” (defined as dollars per year raised since founding).

Across every metric, startup founders with prior vertical experience hold an edge. Most striking is the impact of founder experience on AI-forward vertical startups.3 Vertical AI founders with vertical backgrounds raised more than 2x the average deal size ($100M vs $45M) compared to non-vertical backgrounds. In Vertical SaaS, mean deal sizes for vertically experienced teams were only 21% higher. The tight banding of median deal sizes suggests a fatter right tail — vertical-background founders in AI are landing outsized rounds that pull the mean significantly above the median.

We also wanted to look at the “quality” of VC investors attracted to cap tables.4 While no perfect measure exists for this, we used subjective brand perception as a proxy to assign “VC Scores” to each startup in the dataset — and it was consistently higher for vertical background founders across every subset. While the delta was most modest in Vertical AI, we expect this is an artifact of the data.5 A dollar-weighted VC score heavily favors vertical experience.

Analysis by Stage

For rounds of $15-50M, vertical-background founders are a modest majority—61-63% of companies. At $50-100M, however, they account for over two-thirds. At $100M+, it’s nearly three in four. The bigger the check being written, the more likely the founding team has prior vertical domain expertise.

One possible interpretation of this data supports the theory that vertical experience confers advantages in distribution and product expansion. While product velocity and vision dominate early on, later-stage fundraising rewards enterprise sales traction, regulatory navigation, and buyer trust — all of which are earned through true expertise and credibility. This, in other words, is what you’d predict if domain expertise were a durable advantage. Interestingly, it creates a through-line to our prior analysis of exited founder backgrounds, which found a >80% share of domain expertise by exit dollar.

Analysis by Vertical

As measured by VC quality, domain experience seems to have the biggest impact in sectors that are viewed as more clubby, credentialed, and opaque to laymen. Legal, for example, sees the greatest blended lift from industry backgrounds: 89% of funded legal AI companies have vertical-background founders, with a 100% median deal-size lift and a 22% VC-quality-score advantage. Harvey, EvenUp, Spellbook — a solid chunk of the companies defining the category post-LLM — were built by founders who practiced law or worked deeply in legal operations. Public Sector shows the most extreme VC-quality lift of any vertical (+75%). Like in Legal, it’s seen as a traditionally difficult procurement environment in which insider credibility and connections are critical.

Meanwhile, exposure to real estate and retail (which often blurs into commerce generally) has exposure across a broad array of companies, backgrounds, and life experiences. In those categories, branded funds tend to prefer outsiders.

Analysis by Age / Work Experience

The older a vertical founder, the more likely it is that they have domain experience. It’s not surprising we see this effect more-or-less monotonically, considering experience takes time to accumulate. Over two-thirds of vertical founders with 9+ years of work history have domain experience. Comparing Vertical AI vs. Vertical SaaS yields more interesting results. At every experience level, Vertical AI founders are more likely to have vertical backgrounds than their SaaS counterparts.

As covered above, domain experience is valued differently in Vertical AI vs. SaaS — that effect is amplified when we analyze by years of experience. Successful vertical AI founders with 0-5 years experience are 57% more likely to have domain experience than SaaS founders in the same cohort; amongst all with <9 years of experience, domain experience is 26% more likely. This suggests that the Vertical AI market is already self-selecting for domain expertise more aggressively than SaaS has been.

Another finding, runs against the narrative that domain expertise is the primary factor of success in Vertical AI: young founders attract more “brand” VCs. The 0-5 years-experience cohort achieves an average VC Score 20% higher than the 15+ years-cohort. Are we seeing quantitative evidence backing YC’s recent shift away from experience and towards youth in the AI era? That argument would hold that the proven lift afforded by vertical experience is merely correlation, and that youth — or rather, the free-thinking, fast-moving dispositions that often accompany it — is the rightful primary driver of top VC backing and ultimately success.

It’s worth remembering that validation of VC picking takes a long time, making the names on one’s cap table a very uncertain measure of ultimate performance — especially in a world where AI is theoretically changing the nature and profile of startup success. Perhaps brand-name venture firms have adopted the same mode of thinking as YC.

Here, all we’re interested in is the data — and that tells us that the youth-supremacy narrative falls apart on three counts:

1. If youth were the primary driver of success in Vertical AI, you’d expect young founders without vertical backgrounds to outperform old founders with vertical backgrounds. But they don’t. Those with 9+ years’ work experience have higher capital velocity across the board. Old vertical founders raised bigger rounds than young founders without domain experience. And although young founders with vertical backgrounds marginally underperformed peers without them on VC Score, they outperformed on all other measures. Youth may be preferred by brand-name VC, but it doesn’t substitute for domain experience.
   

2. The VC quality premium for youth is consistent across AI and SaaS, but the domain expertise premium is not. Young Vertical AI founders are 20 points more likely to have vertical backgrounds than young vSaaS founders (55% vs 35%). If youth alone were the success factor, you’d expect that gap not to exist — young founders would succeed regardless of domain background. Instead, the market is filtering young AI founders for domain expertise more aggressively. The ones who make it past the Series A threshold6 disproportionately have it.
   

3. Over 90% of top vertical rounds were raised by founders with >5 years of experience. We’ve spent so much time in comparative, cohort base numbers that it’s important to remember: by volume, super young founders represent a small minority of the overall dataset. The majority of founders with successful Series A+ raises in 2025 had 15+ years experience. Examining the trend by sector, it’s clear that the more regulated and “credentialed” a vertical is, the more seasoned the founder is likely to be; as much as fresh founders transforming the stodgiest industries is in vogue right now, the actual founder splits suggests that winning founders’ experience levels align with the average ages in their respective industry. Not a single A&D founding CEO with less than 6 years’ experience raised a Series A+ in 2025.
   

So why would young, successful Vertical AI founders be so much more likely to have domain experience than their peers in Vertical SaaS? We feel the most likely explanation is the same we shared above: building AI products that automate complex industry-specific judgment requires deeper domain understanding than building software that mirrors and digitizes processes. Velocity, free-thinking, and AI-nativity matter greatly — but in Vertical AI, a founder’s knowledge advantage in their domain is increasingly important.

Analysis by Prior Function

Most successful vertical startup founders aren’t technical. Over a third (35.1%) of CEOs who raised $15M+ vertical rounds were previously executives (prior CEOs, presidents, or founders of other companies). When you add in Ops/Strategy/Finance (14.6%), Product (15.0%), and GTM (6.1%), non-technical business leaders account for more than seven of ten successful vertical startup CEOs. Although non-CEO engineers (including CTOs) represent just 8.9%, they achieved the highest median capital velocity ($7.8M/yr) among non-financial backgrounds. This runs directly counter to the prevailing narrative that technical founders dominate AI-era startups. In vertical markets, the CEO who understands the customer’s workflow, regulatory environment, and buying process has a structural advantage over the one who understands the model architecture.

The most intriguing finding may be the IB / Consulting cohort. Despite being a small group (~4% of successful vertical CEOs), they lead on every key metric: the highest capital velocity ($9.8M/yr), the highest VC quality score (1.69), and the highest median deal size ($50M). These are founders who combine analytical rigor with the pattern-matching and relationships that come from advising industries before building in them — the ex-McKinsey partner who spent three years studying healthcare workflows before founding a clinical AI company. The VC / Investor cohort (8.7%) is also notable as a growing category, with similar dynamics to IB / Consulting: former VCs raise at the second-highest velocity ($8.9M/yr) and command the second-highest median deal ($40M), leveraging their connectivity, if not domain expertise from deal experience.

The “From Industry” cohort represents founding CEOs who launched their startup immediately after an operating role at a strategic in their vertical. Luke Hansen, the founder of CompanyCam (which became a unicorn with the 2025 deal that landed in our dataset) is a perfect example — he was a 13-year exec at White Castle Roofing in Omaha, Nebraska, before going the founder route. These pure domain experts, however, had the lowest VC scores (1.38) and slowest capital velocity ($5.5M/yr). Does this run counter to our argument that vertical experience is critical to success?

At Euclid, we prioritize two founder characteristics in Vertical AI: (1) earned insight into their problem and vertical (most analogous to “domain expertise”), and (2) an innate understanding of what success looks like for a venture-scale business. The latter is difficult to derive without exposure to either high-quality startups or comparable high-stakes environments. Thus, we believe that optimal domain expertise tends to be either: (1) a background with a combination of software / AI plus industry, or (2) industry insight gleaned through prior vertical startup / AI success. That said, we have backed and will back backgrounds outside of any one mold, because no such rules are universal — a healthy chunk of the best founders break all the pattern-matching, and are better for it.

The Democratization Paradox in AI

Per the anti-expert argument, if AI truly democratizes domain knowledge — enabling technical founders to iterate and learn industries in weeks rather than years — then the vertical-founder advantage should shrink in vertical AI relative to vertical SaaS. AI-native, low-experience founders should be pulling ahead on attracting dollars and VCs, as the outsider disadvantage compresses.

Our analysis, however, shows the opposite. In Vertical AI, >70% of leading 2025-funded startups have founders with prior experience in or around their industry — and those founders raised deals 2.25x larger, with greater overall capital velocities, and higher VC scores. Every metric points in the same direction: the more AI-native the company, the more domain expertise matters, not less.

If that finding is indeed structural, we believe it relates to the core function of AI-first solutions compared to traditional software. Vertical SaaS digitizes processes — it moves a workflow from paper to screen. Vertical AI enables something fundamentally different: automating context and judgment. It doesn’t just present information or offer a UI for abstracting knowledge into a database. Vertical AI, writ large, aims to do the work: generate documents, triage decisions, and even act autonomously.

When the product is the work itself, vertical nuance — industry buyer trust, regulatory nuance, in vivo edge-cases — becomes more important. Those are precisely the things that domain experience confers that iteration and obsession alone would take longer to unearth.

The argument against domain expertise predicts convergence between insider and outsider founder outcomes as AI capabilities improve. The data shows divergence — and deltas between Vertical SaaS and Vertical AI suggest the gap may be widening, not narrowing. As the AI stack becomes more legible, the greater rate-limiting factor to scale — to distribution and defensibility — becomes knowing what to build, for whom, and why they’ll trust you to build it. As the ceiling of Vertical AI rises with technical possibility and accessibility, the floor of domain expertise rises with it.

So What's Wrong With the Picture?

Certainly not the parachute pants or the Yu-Gi-Oh résumé — between your two authors, Euclid is rich with MC Hammer and Magic the Gathering phases that may or may not still be active. Coordination of Benefits is indeed a $4B problem, and one worthy of solving. Eye-catching proclamations like “Healthcare is dead” will be effective as long as we have human brains. $15M at a $60M post is clearly rich for a seed-stage company but, for better or worse, not wildly rare in the current YC milieu. Our issue with the picture is subtler: the implication that youth, AI-nativity, and audacity compensate for a complete lack of earned insight into a vertical as complex as health insurance. And while our example is tongue-in-cheek, there are examples at least as striking in every batch these days.

YC’s shift toward younger, more technical founders made sense in 2023, when the AI stack was volatile and the builders who could wrangle it had a genuine edge. But the stack has matured — dramatically. Structured Outputs, MCP, production-grade agent frameworks, and AI-assisted coding have eroded the technical barriers that once justified prioritizing engineering fluency above all else. In a world where every vertical is getting pounded by a dozen low-hanging-fruit AI ideas, we’re not even sure pure velocity — the wall-spaghetti approach of move fast and sling AI — has much life left in it. Garry Tan himself seems to agree with those points in some of his more recent interviews:

The ability to be successful is no longer limited by technical ability. The only thing that’s sort of the limit is can the founders get in the heads of customers.

Our data suggests the market has already internalized it. In Vertical AI specifically, 71% of founders raising $15M+ rounds have prior domain experience — a share that increases at later stages and in more regulated verticals. Young founders are not shut out; they’re just filtered more aggressively for domain knowledge in AI than in SaaS. Youth isn’t a disqualifier; in fact, it’s absolutely preferred. But because the product is the judgment in Vertical AI, and judgment has to come from somewhere — and that’s why experience matters more.

Some GPs have publicly shared their own philosophies in support of founder expertise in Vertical AI. a16z’s David Haber, in Context is King piece, argued that “AI dramatically expands what’s technically possible, but it doesn’t tell you what’s actually useful… [industry judgment] can’t be automated like code and is earned only through experience.” Greylock’s Christine Kim put it more bluntly: “pure technologists attempting vertical AI are at a disadvantage to founding teams who have both domain experience and a technology background.” Brian Feinstein, Kent Bennett, and Sameer Dholakia at BVP concluded in their Vertical AI Roadmap: “outsiders face steep learning curves that burn months and capital understanding nuances insiders grasp intuitively.”

None of this means YC is wrong to back young founders per se. They may have simply leaned into the profile because it’s their historical domain. If they are better sourcers and pickers amongst young, technical founders — and if they drive better returns for the program — then that’s where they should play. They are now so ubiquitous, however, that some in the VC and startup ecosystem interpret the latest Y-Combinator filter as a proxy for “founder profile of the future.”

That model is incomplete if youth and technical skill are treated as substitutes for domain insight rather than complements to it. The best version of our young, pantalooned founder — the one who raises $15M and then actually builds a lasting company — pairs AI-native intuition with hard-won knowledge of how benefits actually adjudicate, where the edge cases live, and why the buyer across the table should trust her to automate their highest-liability workflow.

So to the investors seeing YC’s recent shift toward youth and technical talent, and interpreting them as the primary founder rate-limiters in the AI era: caveat emptor. Our data says distribution, trust, and domain judgment are catching up fast — and in Vertical AI, the market may already be placing them first.

And to founders wondering what to make of all this, perhaps Paul Graham himself said it best over a decade ago:

What you need to succeed in a startup is not expertise in startups. What you need is expertise in your own users…. here is the ultimate advice for young would-be startup founders, reduced to two words: just learn.

We built Euclid from the ground up to serve category-defining Vertical AI founders. By and large, those are builders who pair AI-native vision and ability, with earned domain insights, and first-hand knowledge of what great looks like. The data says they win more often, raise more capital, and attract better partners. We think that in Vertical AI, their advantage will only widen.

Everything’s getting cheaper, easier, and more accessible — except judgment.

Thanks for reading The Verticalist!

Euclid is an inception-stage VC built for Vertical AI founders. If anyone in your network is thinking about leveraging their domain experience to build in AI, we’d love to help. Just drop us a line via the comments below or on LinkedIn.

On this week’s Verticals, we dug into how Camber — a Series B healthcare AI payments startup co-founded by Christophe Rimann and backed by a16z — is building exactly this. Episode here and full case study below.

Watch Full Episode

When Vertical Fintech meets AI Services

Camber is a healthcare AI payments startup that has raised $44M from a16z, Craft, Zigg, and others. Their vision: durable, hard-to-replicate financial infrastructure a la Stripe, but with the defensibility of a Vertical AI platform intaking massive volumes of first-party data.

Does every vertical need its own financial rails? Perhaps not, but the evidence is compelling that many do. At minimum, healthcare is a prime candidate. It lives and breathes on payments between fragmented, complex, regulated parties. As healthcare shifts out of hospital systems and into outpatient settings — behavioral health, PT, ENT, cardiology — unit economics can quickly break down as the cost of financial execution balloons. A hospital, for example, can justify a human chasing the last 10% of a $10K claim. A behavioral health clinic submitting $150 claims by the tens of thousands cannot. The long tail of underpaid claims becomes structurally uneconomic. Providers think they’re collecting at 97–98%; the underlying data says 92–93%. They can’t staff their way to a higher number, so millions of dollars evaporate into deadweight loss.

Christophe Rimann — a former McKinsey healthcare consultant who previously built a licensed crypto broker-dealer — saw that gap and spent five years building a foundational model purpose-built for claims adjudication. Not an agentic wrapper on legacy RCM software, or even RCM software re-imagined for LLMs. Camber’s team, rather, envisioned a vertically-specific system trained on billions of dollars of expert-labeled claims data, capable of reverse-engineering payer denial logic before a claim is ever submitted. Today, Camber’s first-pass “clean claim” rate runs around 95%, compared to 80–85% for large clinics operating without it, and as low as 60% for some $500M+ providers. For larger clients, that delta can translate into millions in annual EBITDA uplift.

Our interview with Camber was made possible by Parafin. Gusto recently partnered with Parafin to embed a line of credit directly into their payroll workflow, so SMBs can cover cash crunches in seconds. Learn how Parafin’s embedded fintech can solve your problems here.

The ROI-First Future of Vertical GTM

Camber is strategically interesting not only because of its infrastructure play, but also because it is rethinking monetization for the AI era. The company prices on outcomes, not seats. This isn’t disruptive for customers since they largely compete with offshore BPOs, not software vendors. Despite the fact that adoption takes doing — Christophe estimates 30–40% of enterprise value lies in change management — their gross margins run at upper-quartile SaaS levels. As Christophe put it: “Most of our providers think of us as a services company.”

Like much of enterprise vertical AI these days, Camber takes many pages from the Palantir playbook. “Sell services, execute through software / AI” is potent in healthcare — like many other real-economy verticals — because the buyer doesn’t want to think about the technology. They want to know their collections number went up and they want it solved. Camber structured their initial customer engagements to prove ROI. That has enabled them to run a 60-day pilot-to-conversion cycle… in a market where RFIs run 18+ months.

This is one of many suggestive data points — across a growing number of Vertical AI startups in the current AI Services wave — that outcomes-based pricing can collapse sales friction when the pain is acute enough. In some sectors, the grip of traditional vertical SaaS cycles is loosening.

Healthcare’s Stripe Moment?

The Stripe analogy is Christophe’s own. It’s admittedly not perfect: much of what makes Stripe what it is comes down to its horizontal universality. Moreover, if something is truly “rails,” shouldn’t it be agnostic to outcomes, rather than predicated on them? But there are a few ways in which mental model is apt.

Thirty years ago, every large credit card purchase involved a back-room phone call to the issuing bank.1 Today, that complexity is fully abstracted and automated. Healthcare insurance claims (and the associated payments) are stuck in that pre-abstraction era — resolution requires dozens of steps, stakeholders, and manual interaction. Camber is betting it can facilitate a similar evolution in healthcare, and compress it into a few years.

Of course, none of that would be possible without the inter-connectivity and process automation LLMs have enabled. By layering AI on top of a proprietary data network, Camber now processes roughly 25% of daily claims volume in its beachhead vertical within behavioral health. Multiple EHR vendors already white-label Camber as their payments layer. One in three of their SMB deals involves the provider switching their entire system of record just to access Camber — a striking inversion of traditional thinking around the defensibility of SoRs and vertical moat-building assumptions.

Payments as a Shadow System of Record

Camber’s success to date is just one case study, and a developing one. But there’s an interesting emerging lesson for Vertical AI founders and investors in healthcare, and other deeply regulated, payment-heavy verticals.

We recently shared our concept of the “shadow system of record” — a strategy by which founders can grow a thriving business in verticals where dominant, controlling incumbent systems of record (like EHRs in healthcare) have historically sucked the oxygen out of the room. The concept is simple, even if identifying opportunities is hard: establish a net-new control point and grow unhampered in the incumbent’s blind spot.

The payments layer may be a prime candidate for shadow SoRs in some sectors. Could EHRs become the interchangeable commodity, while the claims & payments network becomes the stickier infrastructure? Certainly, that would be a meaningful structural shift from the traditional vertical SaaS playbook.

It does echo a pattern we’ve been tracking: the next chapter of vertical software is being written by companies that capture value from non-software budgets, leveraging wedges impossible before the current generation of AI.

Camber has raised $50M — with a Series B led by a16z last year — employs ~90 people and is expanding from behavioral health into five new verticals this year. The open question, as Christophe frames it, is whether they can scale the go-to-market — and whether the claims landscape itself shifts beneath them in the next 18 months. That uncertainty is real — as is the case in highly regulated markets. But the benefit of being more than just a pure Stripe-esque rails play or a pure external Vertical AI Services play, is that Camber drafts on real, rarefied first-party data as it scales. The data moat flywheel they’ve built over five years of expert-labeled claims outcomes2 is compounding — and that’s not something anyone can vibe-code.

Thanks for reading the Verticalist! Don’t forget to subscribe to our show, Verticals, wherever you watch or listen. Dive deep into Vertical AI strategy with a new founder, every Wednesday.

Youtube • Spotify • Apple • Amazon

This episode is sponsored by Parafin, the embedded finance infrastructure powering capital products for vertical platforms. Backed by Ribbit Capital, Thrive Capital, and GIC (with a $100M Series C bringing their valuation to $750M), Parafin supports customers like Amazon, DoorDash, Jobber, and NMI. Click here to learn how Parafin can support you.

The AI agent boom promised to transform freight. Voice agents that negotiate with carriers. Document agents that process customs paperwork. Scheduling agents that book appointments autonomously.

The demos have been phenomenal. Retention… not so much.

As Deepak Chhugani tells it, logistics companies adopting these tools are hitting the same wall: change management is brutal, every broker’s workflow is different, and customers blame the vendor when things break. The result is a growing cohort of well-funded agent startups struggling to renew contracts. Perhaps a story not unfamiliar to AI-native founders working to bring agentic solutions to complex legacy industry.

Deepak is the Founder / CEO of Nuvocargo, an AI-native logistics platform that has raised >$85M from Tiger Global, QED Investors, NFX, and Y Combinator. His answer to the retention problem is in line with our recent essay: don’t sell the agent. Between Mike Powers at BuildVision (on Verticals here) and Chris Hladczuk at Hanover Park (on Verticals here), it’s a story we’re hearing more and more.

Nuvocargo sells the outcome. They are literally a licensed freight broker — one of their three offerings with ~40% gross margins — touching 300k carriers. The distinction, as Deepak illustrates, is that his partnership with clients goes even deeper than traditional brokers can. As opposed to a pure vendor, they build a deeply internal relationship with clients, not only further aligning incentives, but also making the service harder to replace and easier to extend:

• Freight brokerage: “Quote me a rate from Dallas to New York.” Traditional vendor: transactional, commodity-prone, external. Price is the main differentiator, which is why so many freight brokers blew up post-2022.

• AI-managed supply chain partner: “Here’s my $50M freight budget. Run it for me.” An incentives aligned partner with high internal access to customers. Plug into the customer’s ERP, replace 15 SaaS tools, and execute the right strategy for the customer — not just the the most profitable point-in-time deal.

On this episode of Verticals, we unpack exactly how selling the outcome works at Nuvocargo, Deepak’s path to where he is today (from investment banking in Mexico City to YC) — and how his framework breaks down into “Four Ps” that all founders building Vertical AI Services should ask themselves to stress-test their models.

The Vertical Playbook

The 4 P’s of Vertical AI Services

A litmus test for founders building defensible AI Services companies.

Performance

Can you deliver measurably better outcomes than the incumbent? Nuvocargo runs 12+ narrow, task-specific AI agents inside NuvOS, each performing a single function: scheduling appointments, processing documents, negotiating carrier rates. If one fails, a human unblocks it and the pipeline proceeds. Critically, the customer never engages with the agents. Instead, they see results: seamless delivery and — for one early customer — a 12% reduction in a multimillion-dollar freight budget in year one. If you can’t point to concrete ROI, you’re selling novelty.

Profit

Can you structurally lower your cost to serve through AI? Deepak draws a sharp line between his model and the boom-bust cycle of traditional freight brokers, who hire aggressively in good times and fire in bad ones. By embedding AI agents inside NuvOS rather than relying (fully) on headcount, Nuvocargo is shifting its P&L from a labor-heavy service business toward a consumption-based model. As Deepak puts it, the question every AI services founder must answer honestly: does your technology actually make each unit of work cheaper?

Pricing

Can you price in a way customers already understand? Managed transportation is an established business model — multibillion-dollar 4PLs have charged shippers 2–4% of freight spend for decades. Nuvocargo adopted the same fee structure rather than inventing a novel pricing model that would force customers to justify a new budget category internally. The pitch is simple: pay the same rate you’d pay the legacy provider, but get a modern technology stack executing on every shipment. As Deepak shared, he learned the hard way that squeezing every dollar of value upfront kills deals and discounts your ability to expand over time.

Partnership

Can you embed deeper than a vendor to become infrastructure? This is where Nuvocargo’s model diverges most from the standalone agent playbook. When the company integrates with a shipper’s ERP and begins consuming purchase orders directly, Deepak’s team isn’t just quoting a lane — they’re questioning whether a shipment needs to go at all. The customer replaces 15 SaaS tools, multiple broker relationships, and the headcount to coordinate them. That’s a very different switching cost than canceling an “agent.” As we’ve written on the commoditization risk facing AI services, startups that remain external vendors competing on cost will watch margins compress. The ones that become an orchestration layer — owning data, integrating with the customer’s stack — build moats with compounding potential.

The Takeaway for Vertical Founders

Performance. Profit. Pricing. Partnership. These four tests separate durable AI services businesses from “agents” and LLM wrappers. Nuvocargo passes all four not because the AI is more sophisticated, but because the business model was designed to leverage AI invisibly: delivering outcomes customers already know how to buy, at structurally superior margins, with customer “internality” that builds real stickiness.

Thanks for reading the Verticalist and for tuning into the show! Drop us a line below with your thoughts. And don’t forget to subscribe to Verticals wherever you watch or listen:

Youtube • Spotify • Apple • Amazon

Watch Full Episode Here

It’s becoming increasingly clear that the picture-perfect “AI replacing workers” story doesn’t add up. Evidence linking AI to hiring slowdowns has been muddled at best, while productivity claims have been hard to measure (some have even been disproved by available data). The “AI-Washing” of job cuts — executives using AI automation to try to flip the negative connotations of layoffs into a positive shareholder narrative — has become pervasive.

None of this means that LLM-era AI isn’t fundamentally transformational, nor that its economic potential is capped. We’re just in the early innings of adoption. It is easy to forget that there was a 20-year gap between the commercialization of the personal computer (by IBM in 1981) and the peak of the digital productivity boom at the turn of the century.1

The single-mindedness of the current AI automation narrative is easy enough to explain: incentives, incentives, incentives. The loudest voices on AI’s labor impact have had the most to gain from speaking up — from the execs running frontier labs to CEOs right-sizing their workforces in the post-2021 era to VCs stirring up hype to justify ever-larger AUMs.

But it’s becoming a problem.

The Church of Agenticism

Hyperbolic AI narratives have now graduated from high-level economic pronouncements to product positioning itself. Every AI-powered feature is now an "agent." Conference panels want to discuss "managing your AI workforce." Every enterprise buyer is being told they need "digital co-workers." Founders are giving their agents names, personalities, and even human faces. Some startups are gauging candidates by how “agentic” their mindsets are.

The term agent — an inherently vague word meaning “one who acts” — is starting to collapse under the weight of its own ubiquity. While its rise was fueled, at least in part, by the desire to make AI more accessible and urgent for enterprises, it may be having the opposite effect. When everything becomes an “agent,” it's nearly impossible for buyers to tell what any given product actually does, how it fits into existing operations, how they should interact with it, or why it matters to their bottom line. In Vertical AI specifically, this is disastrous.

While there are verticals in which exploratory or performative AI adoption is real2, most industry buyers don’t care about AI — they care about results. Good Vertical AI is a capability infrastructure: a technological foundation that expands what organizations can accomplish, well beyond what is possible through any combination of humans and traditional software. Buyers and stakeholders need a clear way to understand the function of Vertical AI: its potential to unlock new business opportunities and strengthen industry-specific workflows. By focusing on form with metaphors like “agent”, by contrast, we risk obscuring the real opportunity, alienating the buyers who will drive adoption.

Agentic systems — AI that can accomplish complex, multi-step workflows in pursuit of goals — represent a genuine and meaningful capability shift. It is the desperate desire of all innovators to coin new “categories” that capture mindshare and warm audiences to their product, and for good reason. Metaphors are the heart of product marketing: key cognitive tools that shape how buyers evaluate, how organizations adopt, how new markets develop, and even how products themselves evolve.

We oppose Agenticism because it peddles a risky myth: that buyers won’t need to alter their business practices to enjoy the great benefits of AI. This has happened before.

GM Builds the “Factory of the Future”

On October 20, 1984, The New York Times published an article titled, “GM Factory of the Future Will Run with Robots.” Roger Smith, then GM’s CEO, argued that automation would help the company compete with increasingly formidable Japanese competitors. Smith bet his tenure on automation: GM would invest over $40-45B in factory robots in the mid-80s. It wasn’t GM’s first shot at owning the narrative of the future, but it would be their most expensive by a long shot.

Smith’s premise was simple: replace costly, unionized labor with tireless machines, keep the workflow and assembly line the same, and profit from the resulting productivity gains. Better, cheaper, faster! What actually happened, however, is a cautionary tale about the difference between adopting and buying technology. GM placed robots exactly where human workers had been. Job roles stayed the same. The pace and sequence of the assembly line were unchanged. The robots acted as obedient replacements for labor, integrated into a system designed around human limitations. 

Unfortunately, GM’s implementation was widely panned as a disaster. Smith’s robotic factories struggled to match the productivity of their human-run forebears. Robots sometimes painted each other instead of cars or welded their doors shut. Manufacturing costs increased.

Toyota, working with similar robotic technologies, asked a fundamentally different question: what becomes possible when a new capability is introduced into the system? That shift in perspective changed everything. Plant layouts were redesigned. Work cells were reorganized. Quality feedback was made more immediate. Human workers moved from task execution to overseeing the system. The factory transformed into a coordinated human-machine environment, not just a human workflow with machines attached.

Both companies had access to the same tools. Only one reconsidered the architecture around them. Ultimately, GM’s implementation fell short because it focused on swapping people for machines rather than on infrastructure to enhance capabilities. The long-term divergence in productivity and competitive advantage developed from that initial framing choice.

Stanford economist Paul David documented the same pattern a century earlier in his landmark study of factory electrification. In the late 1800s, factories ran on steam—a single massive engine turning an overhead drive shaft that powered every machine in the building through an intricate web of belts and pulleys. The entire factory layout was dictated by proximity to that shaft. When electric motors first became available, factory owners did the obvious thing: they ripped out the steam engine and bolted an electric motor in its place. The shaft kept turning. The belts kept running. Productivity barely moved.

It took thirty years before a new generation of manufacturers realized that electricity’s real advantage was distributed power—a small, independent motor on every machine, freeing the factory floor from the tyranny of the shaft. Suddenly, layouts could follow the logic of production rather than the logic of power transmission. Single-story factories replaced multi-story ones. Assembly lines became possible. Workers gained autonomy over their own machines. The productivity gains were transformative, but they required abandoning the architecture that steam had imposed.

The “agent” discourse repeats the same automation mistakes for the modern era. When AI is cast as a co-worker, the unit of adoption becomes the role, not the workflow or the system in which that workflow is embedded. Executives begin asking HR questions, and the functional value derived is simply cost. What’s the defensibility of selling cheaper work? Perhaps the next widget seller will undercut your pricing even further.

Funny enough, this wasn’t even GM’s first time making this mistake around technology. In 1919, the company entered the tractor business with a machine called the Samson Iron Horse, which was steered with leather reins so farmers would feel more comfortable than with a steering wheel. It flopped immediately, and GM exited agriculture by 1922, having learned nothing it would remember when the robots arrived sixty years later.

The Co-Worker Problem

The co-worker analogy seems harmless and helpful because it makes complex technology feel familiar. An entrepreneur can sell a vision of augmenting or even replacing specific staff members. It’s simple to understand and, hypothetically, easier to sell. It’s also straightforward to pitch the vision to investors and stakeholders: a large amount of money is spent on labor performing a specific task(s) that we can automate. Easy to explain, easy to grasp.

Perhaps unsurprisingly, the “factory of the future” powered by robots promised by the GM CEO was praised at the time and elevated him to the status of the press darling. He became a champion of U.S. manufacturing, catching the public's imagination and becoming a media hero.

Described as an "innovator," "visionary," and "21st century futurist," Smith was named Automotive Industries’ Man of the Year and Advertising Age's Ad Man of the Year, honored with the Financial World Gold Medal (best CEO in America), and designated by The Gallagher Report as one of the ten best executives in America. With such acclimation, it seems little wonder that "GM completed the 1980s in a state of arrogance.”3

It turns out that selling the replacement of labor with robots (or AI) to the press and investors was the only successful part of the GM initiative. The reason the “co-worker” metaphor is flawed is that it draws attention to staffing rather than to system design. It produces local gains (e.g., less headcount or individual employee productivity) while leaving the underlying structure untouched. Now consider how this plays in vertical markets—construction, healthcare, logistics, legal, food services—the industries where AI has the most transformative potential and where the “agent” framing does the most damage.

These buyers don’t wake up wanting an agent. They face a scheduling problem, a compliance bottleneck, and an intake workflow that’s losing money. Their procurement process relies on PDFs and phone calls. The customers most likely to benefit from Vertical AI are often the least tech-savvy, the most skeptical of hype, and the most allergic to buzzwords. Calling your product an “agent” shifts the focus from the problem to the technology. It’s the same category of mistake founders make when they lead with their product, or worse, their tech stack, rather than the customer’s pain.

Unsurprisingly, the mistake with GM extended beyond the metaphor; the entire automation process was flawed. It exhibited a poor understanding of the assembly line and the workflows that comprised it. It viewed labor as fungible units to be replaced, while assembly plants are especially complex environments. As the old adage goes, there are “few ways to lose money faster than automating a process you don’t understand.”

GM failed not because its robots were bad, but because it treated them as direct replacements for people. Leveraging automation necessitates understanding both the intended and actual “work to be done.” Successfully deploying Vertical AI requires a similar insight: you must understand the work so you can re-architect it to maximize its benefits from AI. Otherwise, you risk creating tools that sound great in a pitch deck but are ultimately misaligned with the actual process being automated — and incapable of realizing the transformative potential of AI.

Capabilities, Not Agents

The “agent” metaphor replicates this misalignment on a large scale. It presents AI as a replacement for labor, when the true opportunity—especially in vertical markets—is to broaden what a business can achieve. Consider the two categories of work in which Vertical AI is best positioned to operate. The first is administrative: scheduling, recording, invoicing, intake—back-office work that is time-consuming, inefficient, and likely not done well. The second, and far more interesting category, is the work not done: bids not submitted, calls not answered after hours, patients not seen, inspections not performed. This is the biggest unknown and the most compelling opportunity. It does not show up in a headcount reduction model. It shows up in revenue growth.

The distinction matters enormously for how we size these markets and how we sell into them. Much of the narrative around AI’s value has centered on direct labor replacement, given the immediate profit potential for buyers. But we believe AI’s greater impact—and the larger market opportunity—will be its ability to enhance workforce productivity rather than simply substitute for it. To bring this into focus, consider the hidden revenue lost in “work not done.” For instance, in industries like construction or legal services, missing out on even 10% of potential bids or after-hours service requests can translate into billions in unrealized revenue. If AI-enabled systems could capture just a fraction of these missed opportunities, the value unlocked could dwarf the savings from labor-reduction models. Making the “work not done” visible gives buyers a concrete, numbers-driven reason to rethink how they adopt technology.

This is the Toyota playbook, applied to vertical markets: when a new capability enters the system, the system should reorganize to exploit it. Freeing up administrative work may require less headcount in the back office, but tackling the work that remains should have the opposite effect—enabling businesses to take on more projects, more customers, and more patients, and to grow their revenue. More business will likely require more headcount, not less. In a steady state, the more productive firms will combine enhanced labor productivity with AI to grow and outperform.

When you frame AI as an "agent"—a digital worker—you implicitly promise substitution. When you frame it as a capability, you promise expansion. The second framing is more accurate, more attractive to buyers, and more aligned with how technology has actually transformed industries throughout history.

A concise comparison crystallizes the difference:

The richest upside comes from treating AI not only as a path to speeding up the status quo, but also as infrastructure to unlock new ways of doing the work altogether.

The "Moving Boundary” Between AI & Humans

There is a subtler strategic point here that the agent discourse obscures entirely. As AI capabilities improve, human roles won’t disappear; they migrate. The customer support representative is pulled into the boundary region where a customer is distressed but not saying why, where empathy requires overriding policy. The urban planner is freed from traffic modeling to mediate between incompatible visions of what the city should be. The construction project manager stops manually tracking submittals and starts managing more complex, higher-margin jobs. Even in AI services, the human transitions from rote execution to the paradigm of action that models learn from at the edge.

In vertical markets, this migration is the entire value proposition. The pitch is not “we’ll replace your people.” The pitch is “your people will be able to do more of the work that actually matters—and more of it.” That is a very different sale, and it requires a very different story than “agents replacing humans.”

The Vertical AI companies gaining real traction understand this intuitively. Abridge doesn’t sell health systems as an “agent.” Instead, they redesigned the documentation workflow so that a patient conversation automatically becomes a structured clinical note inside the EHR. The physician’s role migrates from hours of after-hours charting to reviewing and signing off on AI-drafted notes. Clinicians using Abridge report 86% less effort on documentation and 60% less after-hours work. That’s a much more powerful value prop (and business case) than selling a note-taking agent.

In legal, EvenUp has built a claims intelligence platform for personal injury firms that automates demand drafting, medical chronology creation, and case analysis across the entire lifecycle of a claim. Firms using EvenUp report that their drafting output has tripled and that average settlement timelines have been cut by a month, all without adding headcount. The attorney’s role shifts from documentation to higher-value cases within the firm.

In construction, BuildVision helps manufacturers, distributors, and suppliers automate the manual back-and-forth of the selling process—ingesting and normalizing PDFs, product specs, and unstructured data that previously lived in email threads and phone calls. The sales rep doesn’t get replaced; they get freed from the drudgery that kept them from responding to more prospective jobs, and even taking on higher-margin, more complex jobs. None of these companies leads with “agent.” They lead with the workflow, and the results follow.

GM’s CEO lacked this strategic vision. Plant closures and layoffs characterized his leadership. Replacing humans with robots was the end goal. He famously told auto workers: “Every time you ask for another dollar in wages, a thousand more robots start looking practical.”

Toyota’s engineers, on the other hand, understood that once new capabilities enter a system, the system either adapts to exploit them or resists and diminishes them. The business process (and eventually the organization) becomes a moving boundary between what the technology can reliably handle and what humans must still provide. Leading such an organization requires attention to the evolution of that boundary, not a static model of which “digital workers” occupy which roles.

Similarly, agents or, in Toyota’s case, robotics automation, alter the economic value of human capabilities. When an agent reliably handles a task, the economic importance of the skill required to perform it decreases, while the value of other human capabilities increases. This dynamic appears similar to augmentation but is fundamentally different. Augmentation relies on the idea that the worker remains central to the workflow, viewing AI as an enhancement of their capabilities. As this essay explains:

As agentic capabilities improve and force the structure of the work to reorganize around them, without proactive human-agent systems design, workers no longer define the workflow; they occupy the segments where agentic capabilities fail, get bottlenecked, or hit their limits. The effect may look like augmentation from a distance, yet the underlying logic is different: the system expands outward through agentic execution, and humans are continually reallocated to the frontier where reliability breaks and human capabilities like judgment, interpretation, and governance create distinct value.

Words Matter More Than You Think

Our anti-Agenticism is more than a semantic quibble. AI has a public perception problem today. Only 13% of respondents in a recent survey believe artificial intelligence will do more good than harm. The “stop hiring humans” billboards, the flippant commentary from tech CEOs about mass job elimination, the casual framing of AI as a replacement workforce—all of it feeds resistance in exactly the markets where AI can have the most impact.

The industries with the worst productivity—construction, healthcare, transportation—are also the ones with the strongest cultural attachment to human labor and the deepest skepticism of technology. These sectors account for 37% of total hours worked in the US economy but contribute only 24% of output. Their productivity CAGR has been below 0.5% since 2005. Construction has actually gotten less productive. There is no credible path to restoring historical productivity growth without a digital transformation of these markets. And there is no path to digital transformation if the people running these businesses think AI is coming for their livelihoods.

The “machines versus people” framing is simply bad business. This dynamic has caused labor tensions, slowed down technology adoption, and harmed organizational performance for over a century. The answer to disruption is not restraining innovation. But it is also not flippant to discuss job losses or to sell your technology as a human replacement. Founders should sell the broader promise of Vertical AI: technology that boosts productivity and drives growth, not just cuts costs.

GM CEO Roger Smith was a highly intelligent executive with good intentions who failed because he misunderstood the leverage of automation. For Smith, robotics represented the Holy Grail, the ultimate solution that could fix all of GM's problems at once. Unfortunately, he was wrong. By the time Smith retired in 1990, GM had shifted from the lowest-cost producer in Detroit to its highest-cost producer, largely because of the push to acquire technology that never paid off. Toyota, on the other hand, prospered. By 1990, they had surpassed GM in profit per vehicle by a factor of three, despite producing fewer cars. By 2008, it overtook GM as the world’s largest automaker, a position GM had held for 77 consecutive years.

The companies that succeed in Vertical AI will be those that, like Toyota, understand the moving parts of the system, not those selling a shinier robot for the same assembly line. If you’re a founder working in vertical markets, the practical and immediate implication is to drop the word “agent” from your pitch deck, landing page, and sales calls. Focus on the workflow you’re redesigning, the capacity you’re unlocking, and the outcome you’re delivering.

If we could leave you with one thought, it’s this:

Toyota never sold a robot — they just shipped a better car.

Thanks for reading The Verticalist!

Euclid is a VC backing Vertical AI founders from inception. If anyone in your network is working on a new startup in this space, we’d love to help. LinkedIn is the best place to reach us.

Watch Full Episode on YouTube

This episode is brought to you by Parafin — they power the financing infrastructure behind platforms like Mindbody, Gusto, Amazon, and Fullsteam. And they just secured $360M from Cross River to keep scaling. Learn how you can turn capital access into a retention engine:

Today’s Episode

Fund administration isn’t the sexiest industry — in fact, it’s hard to find many CFOs and managers who love their vendor. But it underlies $100T in global assets and is an attractive business in many ways: monetization scales directly with AUM + generally high customer lifetime values. Today, the fund admin revenue pool is dominated by legacy incumbents like SS&C, a $20B+ public company. They had little incentive to modernize. Until now.

Enter Chris Hladczuk, Co-Founder and CEO of Hanover Park. Before writing a line of code, he cold-called 50 CFOs. When half of them started cursing at the mention of their fund admin, he smelled blood in the water.

That customer mindset, paired with the conviction that Vertical AI could reshape services delivery, became the foundation for a company that just raised a $27M Series A led by Emergence Capital and scaled to $15B of assets on platform in twelve months.

It didn’t take Chris long to realize this couldn’t be a traditional AI-powered software play. Fund administration is high-trust with near-zero tolerance for error — historically meaning human-heavy services. AI is well suited to repetitive, numbers-based workflows, but one hallucination can lose you a $1B client, and there’s no shortage of edge cases quarter to quarter. Cognizant of that, Hanover Park is decidedly human-in-the-loop Vertical AI services. But what does that mean for margins? For product automation over time? How do you stage gate work product and ensure your AI services model scales over hundreds of billions of AUM?

Those finer points of Vertical AI Services are what we’re talking about in today’s episode. Chris shared his path for operationalizing human-in-the-loop — an important reminder that as far as LLMs have taken us, the economy will need humans more than ever to adapt AI to high-integrity fields like financial services. Stick around to hear Chris’ journey cracking a venture-scale market where software failed — from avoiding avoiding the pitfalls of AI services, to landing an outstanding Series A partner in Jake Saper at Emergence.

The Hanover Park Backstory

Chris’s path to fund admin wasn’t linear. After investment banking at Goldman Sachs and a CRO stint at fintech company Meow (scaling 10 to 1k customers), he declared “B2B SaaS is dead” on a podcast in September 2024. While that take was widely dismissed at the time — and might still be hyperbolic — its spirit has been increasingly validated, as AI has transformed the meaning of software.

His thesis: if building software gets cheaper, the moat shifts to whoever owns the outcome. In fund admin, that means not selling software to accountants, but doing the accounting — software, agents, and humans vertically integrated under one roof. He co-founded Hanover Park in 2024 alongside CTO Nick Puljic (Columbia CS, previously co-founded Stock Unlock via YC W22), and the pair set out to build a purpose-built ERP for private funds from the ground up.

At Euclid, we don’t think that owning the outcome offers any inherent moat — nor that services is commonly a fast track to vertical defensibility. But we do agree that the line between services and software is blurry in the age of AI. As we’ll discuss in this episode, the proof is in the pudding: can you internalize with a customer as deeply as any software ever could? Can you build internal IP giving you an automation and / or accuracy advantage? Is your bet on long-term margins credible?

Chris’ answers taught us a lot — catch the full episode to hear them.

The Vertical Playbook

Human-in-the-Loop Vertical AI

Step 1: Build the ERP Internally

Chris’s core architectural bet: own the system of record. Hanover Park built its own general ledger, document extraction engine, and reporting infrastructure from scratch. The system of record determines what agents can actually do — competitors who layer AI atop QuickBooks or a white-labeled legacy ERP hit a ceiling. As Chris frames it: without the ERP for the fund, you’re just a UI on top of somebody else. Don’t just bolt agents on top of someone else’s moat.

Step 2: AI Prepared, Human Reviewed

In a zero-fault-tolerance domain, the worst pitch is “trust me, it’s AI.” Hanover Park‘s paradigm: agents handle the transactional grunt work — cash reconciliation, document parsing, journal entry preparation — while human CPAs review and approve. One customer reported it felt like 25 people on their account when they had two. By automating low-level work, Chris can pay accountants more, give them equity, and recruit what he calls “the Navy SEALs of fund accounting.”

Step 3: Bundle Features, Monetize Products

Unlike the typical wedge-then-expand playbook, Chris refuses to sell any product standalone. Portfolio management, LP portal, KPI reconciliation, and money movement all come bundled with core fund admin at no extra charge. The logic: unbundled products are worse because they lack the source-of-truth data in the accounting system. The bundle is both a competitive moat and a retention mechanism that compounds switching costs.

Step 4: Consider Industry Pricing

Hanover Park charges basis points on AUM — not SaaS subscription fees. This aligns revenue directly with customer success: as a fund grows, so does the contract. Closed-end fund structures mean worst-case duration is ten years, and the best-case is a GP raising a $10B successor fund. Revenue quality mirrors subscription businesses without the artificial per-seat ceiling.

Step 5: Customer Selection as Strategy

Chris actively turns away funds below $100M. In a service where one unhappy CFO can poison a referral network, discipline on customer selection is existential. His North Star isn’t NPS — it’s whether every customer is a “raving fan.” That selectivity, paired with an always-get-on-the-plane founder-led sales motion, builds the trust required to rip out a CFO’s most important vendor.

The Takeaway for Vertical Founders

Hanover Park offers one blueprint for the AI-native services wave: build the system of record for internal use, vertically integrate the human + agent stack, price on outcomes, and bundle relentlessly. In a high-stakes use case like fund administration, there’s no room for hallucinations. The margin for error is low, as is the threshold for losing trust. At the same time, in such fields, if AI can be leveraged to deliver a critical service more reliably and accurately, that’s a trust-builder like no other.

So far, $15B in assets says Hanover Park is doing just that. Especially if gross margins grow and human-in-the-loop yields better automation over time, multiples could look increasingly software-like. This episode is a reminder that the most defensible AI Services companies aren’t necessary replacing humans — they’re making a tight cohort of them dramatically more valuable.

Watch Full Episode Here

Thanks for reading the Verticalist and for tuning into the show! Drop us a line below with your thoughts. And don’t forget to subscribe to Verticals wherever you watch or listen:

Youtube • Spotify • Apple • Amazon

Keep Reading…

It’s become something more than that — a publication about what happens when AI meets industries with real complexity. About messy workflows, rare domain expertise, and AI adoption beyond Silicon Valley. And most of all, about vertical architects — the founders reimagining industries from the ground up.

Hence: The Verticalist.

If you’re a founder building in Vertical AI, an operator with an earned insight you’re translating into a startup idea, or an investor tired of pattern-matching on the wrong patterns… this is for you.

Same team. Same thesis. New name.

Thanks for reading. More soon.

— Omar & Nic

2025 was the year Vertical AI moved from hypothesis to reality. After a 2024 defined by AI infrastructure buildout and foundation model wars, capital flowed decisively toward industry-specific use cases. We spent months dissecting every software and AI venture deal of $1M+ across the US and Canada — 4,395 financings and 254 exits, analyzed by vertical, stage, investor, geography, and more — to understand exactly what happened, and what it means for 2026.

Here are a few of our most interesting findings, in brief:

The Year in Numbers

The US and Canadian VC market produced 4,395 software and AI financings totaling $186B in 2025. Vertical startups captured 53% of deal volume and 30% of capital deployed—though that capital gap is almost entirely a function of a small handful of infrastructure mega-rounds (OpenAI’s $40B, Anthropic’s $13B, and a few others). Strip out just the 12 deals sized $1B+ and vertical takes 51% of capital.

Deal volume was relatively even across quarters—Q1 and Q2 combined for 52% of the year’s deals—but capital deployment was lumpy. Q1’s $60.5B was inflated by OpenAI’s Series F alone; remove it and the quarter drops to roughly $20B, making the real quarterly trajectory far more balanced. Vertical’s share of capital climbed steadily: 19% in Q1, 34% in Q2, 32% in Q3, and 42% in Q4.

On the exit side, 2025 delivered $132.7B across 158 vertical transactions—plus another $102B in non-vertical exits. M&A dominated by count, but the story is more nuanced than that, as we’ll get to below. Healthcare led exit volume with 43 transactions, and the IPO window cracked open with 18 vertical offerings after a two-year drought.

Our full report covers vertical-by-vertical breakdowns, stage dynamics, investor league tables, geographic distribution, profiles of fifteen notable deals, and our 2026 outlook. What follows are five findings from the data that we think reframe the conventional wisdom and set the stage for another banner year in Vertical AI.

1. The Remarkable Consistency of Mega-Deals

Most people assume late-stage venture is the volatile tier—big rounds, big markdowns, boom-bust cycles. In 2025, the opposite was true.

The $100M+ tier of vertical financings was the most stable bucket all year. Since Q1, we saw 55-57 deals every quarter. Meanwhile, the $1–5M tier of deals — pre-seed and early seed—dropped from 503 deals in Q1 to 362 in Q4, a 28% decline. The bottom of the market pulled back far more than the top.

Who’s sustaining that stability at the top? The investor data tells the story. The mega-round tier ($100M+) was anchored by a small group of multi-deal leads: NEA (10 leads), Kleiner Perkins (7), Oak HC/FT (5), B Capital (5), DST Global (4), and Ribbit (3). Their average deal sizes clustered tightly in the $110–180M range.

Beyond the frequent flyers, the mega-deal lead base broadened: BlackRock co-led Applied Intuition’s $600M round, Franklin Templeton led Plaid’s $575M, Qatar Investment Authority and Mubadala showed up in multiple rounds. Asset managers in pre-IPO rounds is reminiscent of 2021 — the uptick of sovereign wealth involvement is perhaps more notable, especially in light of their parallel growing interest in backing US VC funds as LPs.

2. A&D Momentum: Conviction in Chaos

When people think “hot vertical AI categories,” they think healthcare and fintech — and of course, they’d be right (especially regarding the former). But the 2025 data tells us that another category may be dominating the hype cycle.

Aerospace & Defense led all verticals in both capital intensity per employee ($476K per FTE) and median valuation step-up (2.15x). It beat Manufacturing ($421K, 2.08x), which got most of the headlines thanks to Project Prometheus ($6.2B) on the funding side and Ansys ($35B) on the exit side. It beat Financial Services, which despite attracting the most aggregate capital posted a modest 1.80x step-up (turns out bigger rounds dilute per-round markups even when total value creation is strong).

The contrast with Education is particularly stark. Education had the lowest capital intensity ($200K per FTE), the lowest capital velocity ($1.6M raised per year since founding), and the lowest capital efficiency ($342K raised per employee)—bottom of the table on every capital metric. And yet Education posted the highest early-stage step-up (2.5x in the $5–15M tier). Despite great emerging stories in SchoolAI and MagicSchool, the category isn’t yet a proven absorber of capital at scale. A&D is the opposite: capital-hungry and valuation-rich, despite its unproven track record as a generator of historical VC liquidity.

Forterra’s $238M Series C—profiled in our full report—is a case in point. Unlike some other A&D high-flyers, they have production contracts and program-of-record revenue, not R&D grants. Backed by an interesting investor syndicate (Salesforce Ventures, Franklin Templeton, Hanwha, NightDragon) that reflects the reanimation of crossover investing. Beyond any one deal—and equally relevant to VC and LP funding—the intersection of government and technology a is dynamic we’re watching closely heading into 2026.

3. CA: America’s Least & Most Vertical State

Amongst all US states in 2025, California posted the lowest ratio of vertical deals, 41% — yet it was also home to the highest absolute number, at 667 deals.

This makes CA simultaneously the least and most vertically oriented in the country. Non-coastal standouts—Arizona (71%), Ohio (70%), and Tennessee (67%)—posted vertical shares that make California look like it’s playing a different game. While New York, at 53%, tracked near the national mean in terms of verticality, it was the clear #2—home to 406 vertical deals in 2025, with signs of the most momentum.

The explanation, in CA’s case, is straightforward: dominant concentration of capital-hungry horizontal AI infrastructure, foundation models, and dev tooling pulls the denominator up. And while it makes sense that company formation is strongest in industry-proximate geographies—contractors in Arizona, manufacturers are in Ohio, healthcare operators in Tennessee.

The regional data adds another layer. The Midwest leads in vertical proclivity at 59.6%, followed by a tight cluster in the mid-fifties: Mid-Atlantic, Mountain West, and Southeast. The West trails at 41.3%. And Canada’s lower vertical share—somewhat surprising given Toronto’s legacy as home of Constellation Software—appears driven by concentration in a single metro, with much of its vertical reach involving investment in US-based companies.

San Francisco and New York remain the twin dominant vertical hubs by volume and deal quality. And while the data show that non-coastal metros embrace verticality, they have two big hills to climb: (1) bridge the massive gulf of general startup gravity; and (2) keeping pace on AI-nativity in this dynamic new environment.

4. Manufacturing is the Momentum Leader

We examined the share of 2025-funded companies in each vertical that were founded in 2022 or later—the post-LLM cohort. Manufacturing & Industrial led in “AI-native” share at 57%. Not healthcare. Not legal. Manufacturing.

More than half of all manufacturing AI companies funded last year didn’t exist before ChatGPT. The cross-vertical baseline is 46%, meaning manufacturing’s new-company share runs 11 points above average. Public Sector (56%) and Legal (53%) also exceeded the mean, but manufacturing’s lead reframes the narrative around the dominance of white-collar verticals in AI-native opportunity.

Deal activity data supports the formation data. Manufacturing deal count rose 41% from Q1 to Q4, the steepest acceleration of any vertical. The category produced the year’s second-largest financing—a $6.2B round we profile among the full report’s fifteen deal highlights.

At the same time, it’s important to remember that as AI and robotics merge into “Physical AI,” core differences in business model can be blurred. Much of this activity in Manufacturing — at least from a capital standpoint — is going into hardware-dominant, high marginal-cost businesses. At $421K in capital intensity per employee, Manufacturing & Industrials companies are raising more to meet the physical-world infrastructure requirements of bringing AI to factory floors, supply chains, and industrial operations — and they are spending more too.

5. Liquidity, Liquidity, Liquidity

In 2025, buyout and LBO transactions accounted for 20% of vertical exits by count—but generated about as much liquidity as M&A. PE firms view vertical software as ideal acquisition targets: recurring revenue, defensible market positions, and operational leverage from AI integration. The majority of PE-backed software acquisitions in 2025 targeted vertical companies. Silver Lake, EQT, and TPG were among the most active acquirers.

Meanwhile, IPO activity last year was meager — but perhaps a start. 2025 produced 9 vertical IPOs (of 15 total) generating $4.3B in aggregate value (just under 40% of all IPO value). Something after a two-year drought, but thin relative to the backlog of late-stage companies that raised $200M+ rounds over the last 12-24 months. Expect more tough go/no-go listing decisions in 2026–2027.

The most revealing exit metric may be the simplest one: the median vertical exit in 2025 was approximately $500M. For the many seed and Series A funds that have scaled up AUM since 2021, driving superlative returns via such median outcomes demands discipline on entry price that seems quite rare these days. At a $10-20M post-money seed, a $500M exit represents 25-50x gross valuation appreciation—exceptional. At a $50M+ post-money “seed” (increasingly common in consensus Vertical AI), the same exit is 7-8x gross. Still great, but tough math for a >$100M AUM fund.

The top 10 exits by value in 2025 accounted for over 60% of total vertical exit value—a reminder that no category escapes power-law dynamics. Below the mega-deals, PE remains a leading exit path, IPOs remain selective — the “middle market” of vertical exits ($250–1B EV) is arguably the healthiest and most under-appreciated segment of the liquidity environment, at least in VC land.

What’s in the Full Report

The findings above are drawn from roughly a quarter of the complete Vertical Report. The full publication includes:

• Vertical-by-vertical analysis of 18 sectors: deal count, capital deployed, stage mix, median deal sizes, and quarterly momentum for each—from Healthcare (597 deals, $11.8B) down to Energy & Climate (60 deals, $900M).

• Stage dynamics deep-dives: deal size distributions by vertical, capital intensity, valuation step-ups, capital velocity, and capital efficiency metrics.

• Profiles of 15 notable deals, from a $190M public-sector round to a $6.2B industrial mega-financing—covering step-ups, investor syndicates, founder backgrounds, and what each deal reveals about its vertical.

• The full exit landscape, including >150 deals worth $130B+ broken down by type, quarter, vertical, and size—plus the IPO pipeline and PE acquisition patterns.

• Investor league tables, from the most active seed leads to the dominant growth-stage firms—plus corporate venture and SWF trends.

• Complete geographic analysis by state and region.

• Our 2026 outlook on four structural dynamics: the seed-to-A graduation test, deep tech’s maturity cycle, shifting liquidity winds, and the coming fight for AI defensibility.

The Vertical Report is published annually by Euclid: a VC firm backing Vertical AI founders from inception. For questions, reach out to the Euclid team via LinkedIn or DM.

This week, we’re taking a look back to recap some highlights to date — and the key lessons we learned from each. From Slice and Broadlume, to Qualia and ServiceTitan, there’s a masterclass here for you. We’ve also added links to essays and playbooks from Euclid Insights relevant to each episode, so you can dig deeper.

If you haven’t, please take a moment to subscribe to Verticals on YouTube below (it’s only place we post full episodes and shorts). It’s been an incredible first quarter… and we look forward to bringing you many more.

Subscribe to Verticals on Youtube

First, a Quick Word From Our Sponsor

In addition to making our show possible, Parafin powers embedded financing for some of the biggest platforms in the world—Amazon, Walmart, DoorDash—letting small businesses access capital directly. They’re a revenue accelerator for hundreds of vertical founders.

Click here to explore what a custom fintech program for your platform could look like.

Ilir Sela, Slice (Ep. 1)

“Independent, Not Alone”

Our very first guest framed the thesis for the entire show. Ilir built Slice into a billion-dollar platform for independent pizzerias by running the reverse franchise playbook: give operators the shared stack and economies of scale of a Domino’s without forcing brand uniformity. The moment that stuck? Slice’s AI voice ordering hits ~90% transaction completion—but humans still clear 98%+. For low-margin shops, that 8-point gap is existential. Until AI clears the bar, humans stay in the loop. Cold pragmatism that every vertical AI founder selling into cost-sensitive end markets should internalize.

Jeremy Yamaguchi, Cabana (Ep. 2)

The AI-First Roll-Up Math

Jeremy laid out the honest economics of buying, operating, and AI-enabling legacy service businesses from his POV running Cabana. The takeaway: you’re running three businesses at once—an operating company, a tech company, and a PE shop. The magnitude of AI margin uplift is still unproven, but the structural advantage is clear: if you own the P&L, you can force AI adoption in verticals that would otherwise resist it for a decade. We unpacked the model further in our AI-First Roll-Up essay.

Todd Saunders, Broadlume (Ep. 5)

Community as Moat

Todd—whom we’ve called the “grand vizier of flooring”—explained how Broadlume turned community, content, and first-party data into durable defensibility in an industry most VCs wouldn’t touch. He parlayed that strategy into a strong exit to PSG — after having given up nearly $10M in ARR to undertake a risky pivot... incredible story. His return for our 2026 Predictions special only reinforced the point: the operators who know their verticals cold are the ones building platforms that last.

Mike Kopko, Pearl Health (Ep. 6)

Healthcare’s First Inning in AI

Healthcare is the largest category in Vertical AI by deal count and capital deployed. Mike — the founder / CEO of Pearl Health — made the case for why it’s barely started: a quarter of the US economy, byzantine money flows, a huge chunk of vertical investment (as we’ve covered in past Vertical Reports) and enormous room for AI to align incentives in value-based care. His key point for founders: sell provable time-to-value and measured ROI for discrete use cases—not “AI.” In non-tech-forward verticals, forward-deployed resources aren’t overhead; they’re the product.

Sam Youssef, Valsoft (Ep. 8)

The Compounding Machine

Sam runs Valsoft, a 130+ portfolio-company, $750M+ revenue vertical software aggregator in the tradition of Constellation Software. And the bigger the vertical, the less interesting it usually is for them. Large markets attract aggressive VC dollars and irrational competition. Smaller verticals—forestry, niche manufacturing, obscure back-office workflows—tend to be calmer. Rational budgets, rational pricing, rational competitors. That’s Valsoft’s domain. While he’s not alone in the world of vertically-focused PE shops (as we covered a while back in our Buyout Frenzy piece), Sam takes a uniquely single-minded approach. The episode was another reminder that “small TAM” markets often harbor enormous, defensible value once you’re inside the workflow.

Dan Friedman, Moxie / Boulton & Watt (Ep. 9)

Business-in-a-Box Done Right

Dan’s Moxie—a platform helping aesthetic clinicians launch, run, and grow compliant medspa businesses—is a case study in an important business model: business-in-a-box. There are many variations: we called it the Synthetic Roll-Up, others like Ilir at Slice (see our first episode above) see it as unbundling the franchise. Dan’s core lesson: start with off-the-shelf software and services to earn trust, before worrying about making the entire tech stack proprietary. Just as important is picking what not to do early. He also discussed how compliance was a unique moat for Moxie; operating across 50 different state regulatory frameworks is just one way they could help. This episode is also worth it to hear about Dan’s unique incubation model in Boulton & Watt.

Rahul Hampole, ServiceTitan (Ep. 10)

The Vertical Fintech Playbook

Rahul ran fintech at Yelp, Plaid, and now ServiceTitan. His message was surgical: optimize for customer value first, take rate second. Have 20 of your best customers on a text channel—they’ll tell you transparently what’s keeping them up at night. That feedback loop is your early warning system for which fintech products to build next. For vertical SaaS players focused on SMBs, aspire for 80–100 basis points on card transactions as a starting point—and fight for it. Generalizing this episode left us with a nice framework for finding PMF (in line with our own product-market fit guide, “Nobody Wants Your Product!”) and considering when to launch new platform products.

Nate Baker, Qualia / Fractal (Ep. 14)

TAM Is Not Fixed

Nate challenged one of the most persistent investor objections in vertical software: the TAM is too small. Having built Qualia into the backbone of title agents across America (500K+ professionals, 600 employees) and launched 150+ vertical SaaS companies through Fractal, his view carries weight. In vertical markets, TAM expands with each new layer of value you create. Companies that seem to have a $5K ACV ceiling routinely break through to $50K+ by solving adjacent problems. The episode is a great real-world case study building on the quantitative frameworks we explored in our Vertical AI market sizing series. Nate’s boldest claim: every business that isn’t seriously reevaluating its product offerings right now will be cooked.

Michael Saltzman, EvolutionIQ (Ep. 15)

From $50K First Year to $750M Exit

Everyone talks about how the revenue bar has permanently elevated. Michael’s story is the counterpoint. EvolutionIQ’s first year: one design partner, ~$50K total revenue. A few years later, 8-figure ARR and a $750M exit to CCC. The playbook: sell guidance (not automation) into a workforce that believes it already has strong expertise. Build a “digital twin” of data rather than writing back into core systems. And make results—not pitch decks—do the early selling. Architecture that can adapt as models improve was the real moat. For founders: this is not only an incredible exit story, but also a reminder of what initial design partnerships in enterprise Vertical AI can really take.

Mike Powers, BuildVision (Ep. 16)

Give Away the SaaS, Capture the Decision Layer

Mike is betting that in the age of AI, the value of the “interface” goes to zero. BuildVision gives away the workflow software for free, captures the network, then monetizes the transaction, the risk, and the capital flow in commercial construction procurement. His take on certain popular AI point solutions was provocatively bearish: Claude can already do a lot of what companies have raised hundreds of millions to solve. The defensible asset isn’t the wrapper—it’s the proprietary data moat and the decision layer (or what many are now calling the Context Graph). As always, Mike is a poster child for the importance of domain expertise — both lived and learned — for AI distribution into traditional industries like construction.

Darren Fike, Santé (Ep. 17)

AI as a GTM Weapon

Darren’s approach to using agentic AI to collapse switching costs might be the most underrated tactical insight from the show. In the $80B U.S. liquor retail market—90% independently owned—every prospect has been on the same system for 10+ years. Santé uses AI to download a prospect’s data, show the system pre-populated on the demo, and collapse the psychological distance between “considering” and “we can use this today.” Recently, this method yielded Darren 30 deals in a single month, zero feature requests. It’s a really unique case study on how founders are not only building novel products with AI, but also leveraging it to solve some of SaaS’s traditional friction points (for further reading on this, check out “Vertical AI’s Integration Problem”).

Nick Tippmann, TipTop VC (Ep. 18)

SaaSpocalypse & Distribution Tactics

In the span of weeks, $400B of SaaS enterprise value evaporated in the midst of the “SaaSmageddon”—the category's largest non-recessionary drawdown in over 30 years. Nick, founder of TipTop VC, joined us to make sense of the carnage. The bears say SaaS is dead. We disagree—AI is still software, and the market is expanding. But the question shifts from "will software survive?" to "who wins?" We shared our answer in our recent post, “The Dispatcher Problem.” Nick thinks distribution is a big part of the answer too — and shares his top tips (pun intended) on early Vertical AI GTM in the back half of this episode.

On behalf of both Euclid Insights and Verticals, we’re so thankful to have you as a reader, watcher, and / or listener — at least in our careers, there’s been no more exciting time to be a part of the growing Vertical AI community.

New episodes of Verticals drop every Wednesday — Subscribe on YouTube or Spotify!

And if you’re an operator ideating on something new in Vertical AI, the Euclid team would love to hear from you. Drop us a link on LinkedIn or our DMs here.

Every week, we go deep on what it actually takes to build and scale vertical software and AI companies. No scripts. No softballs. Just first-principles strategy debates with the founders and investors who’ve done it — from $1B+ exits to first customers.

Every episode gets a companion essay right here on The Verticalist — a strategy write-up that distills the key frameworks, playbooks, and takeaways from our conversations. But the essays are the appetizer.

To get full episodes every week, subscribe to Verticals here:

YouTube (recommended)

Subscribe on YouTube →

YouTube is where we post full episodes, shorts, and clips first. It’s the only platform where you get the full experience — so if you subscribe to one, make it this.

Spotify

Listen on Spotify →

Apple Podcasts

Listen on Apple Podcasts →

Amazon Music

Listen on Amazon Music →

Verticals is brought to you by Parafin — the embedded capital platform powering financial products for platforms like DoorDash, Amazon, and Worldpay. Parafin lets you offer capital to your customers, without building the infrastructure yourself. Big thanks to them for supporting the show… it wouldn’t be possible without them. Learn more about Parafin here.

Where to Start

If you’re new, here are five episodes we’d point you to first:

Ilir Sela, Founder & CEO of Slice (EP 1)

Our very first guest framed the thesis for the entire show. Ilir built Slice into a billion-dollar platform for independent pizzerias by running the reverse franchise playbook. The conversation on AI voice ordering pragmatism — 90% AI completion vs. 98% human, and why that gap is existential for low-margin operators — is one founders still reference.

Michael Saltzman, Co-Founder & Co-CEO of EvolutionIQ (EP 15)

The inside story of one of the first true Vertical AI exits: $750M to CCC. Michael walks through how they spent their first year with a single design partner generating ~$50K in revenue — then hit 8-figure ARR once the enterprise workflow clicked. Required listening for anyone selling AI into large organizations.

Chris Hladczuk, Founder & CEO of Hanover Park (EP 21)

Chris makes the case that the spreadsheet — not legacy SaaS — is the real incumbent in most verticals. A sharp episode on where AI-native tools can displace workflows that enterprise software never could.

Alex Niehenke, Partner at Scale Venture Partners (EP 26)

Alex brings the growth-stage investor lens to the “AI vs. the Giants” debate: can vertical AI startups win against incumbents with distribution advantages? A great counterweight to the founder episodes.

Daniel Friedman, Co-Founder of Moxie / Boulton & Watt (EP 10)

Daniel’s incubation model at Boulton & Watt is unlike anything else in vertical — spinning up multiple vertical businesses simultaneously with a shared operational backbone. The episode on picking what not to do early is one of the most tactical we’ve done.

The Verticalist + Verticals

Think of it this way: The Verticalist (right here on Substack) is where we write. Verticals (on YouTube, Spotify, and everywhere else) is where we talk. Every episode gets a breakdown here. But for the full 45-minute conversation — join us for the show.

Subscribe on YouTube and we’ll see you there, every week.

Thanks for watching!

Look at the $5–15M range in the chart above—in any given month, there are pre-seeds, seeds, and Series As all landing in that band. The word “seed” is being applied to $500k SAFEs and $15M equity rounds. It spans pre-product, pre-revenue teams and companies with millions of ARR. Functionally, the term is now useless.

What actually determines stage nomenclature? It’s honestly hard to say.

Confusion by Design

VC round sizing is — and has always been — more vibes-based than deterministic.

1. Round order (“your first round is a pre-seed, your second is a seed”) never made much sense. Extensions, rolling closes, and opportunistic follow-on slugs throw off the sequence—especially in large datasets.

2. Traction sounds right in theory, but anyone crunching numbers either doesn’t have access to the full data set (e.g., managers analyzing portfolio data) or doesn’t have access to fundamental performance data (e.g., analysts like Carta). Proxies like headcount are too imperfect to be helpful, though Carta’s data does show Series A teams now average just 16.8 employees—down from 25.9 in 2021—reflecting the AI efficiency thesis in real time.

3. Self-reporting is what most structured datasets ultimately rely on, drawing from press or share names in filings. There’s always been a founder (and VC) incentive to report the “youngest” round possible—to seem more precocious. We’ve known founders who did three consecutive “seeds” even though they were hot commodities, purely for positioning. There’s nothing wrong with that—in fact, sometimes it’s the smart move to improve the chances that the right funds will think of you on first glance (though it won’t fool anyone sophisticated at the end of the day). Self-reporting is as inaccurate as it sounds.

4. Round size was historically the most reliable heuristic. Rounds of $1M, $3M, $10M, $30M, and $80M mapped cleanly to pre-seeds, seeds, Series As, Bs, and growth. It was never perfect, and some share of rounds were always way out of band. But in today’s market, that mapping has almost completely collapsed. In the last 12 months, we estimate pre-seeds have grown by 50%. $10M seeds are de rigueur. While perhaps eye-catching in their own ways, neither the $6M Series As nor the $600M Series As are abnormal.

There’s no good single heuristic and we’re not arguing it should be otherwise. But these stage names do play a role in high-level sorting and self-selection in our ecosystem. Founders use the terms to signal to the market the sort of round they intend to raise. VCs use them to signal to founders the sort of investments they like to make (and to LPs, where in the capital stack they play).

Our argument is that we have reached a moment where the terms need to shift, in line with market realities already baked in… even if they haven’t yet permeated the vernacular. But before getting to our prescription, let’s touch on the pathology; how did seed get where it is today?

The Origins of Modern Seed

Round sizes — and hence round names — are ultimately set by VCs, who allocate based on deal flow and AUM. Investment amounts are informed by the same fundamental factors behind all investor decisions: fear and greed. Both are manifesting in new ways that have steadily warped the seed market since 2021.

On the fear side: there’s too much capital in certain bands of early-stage VC. Multi-$100M+ “seed” funds have too much AUM to do $1–2M pre-seeds, but rarely have the brand to compete with platform funds for true Series As. They’re squeezed—so they take the deals they’re good at and expand them. At the same time, AI has inflated what “great” looks like: per Bessemer’s State of AI 2025 report, a “Shooting Star” must eclipse $100M in ARR in 4 years. In that environment, some investors have chosen to gravitate earlier: avoiding scrutiny against rising milestones by backing pre-revenue companies that could be 10x growers rather than ones already demonstrating a solid 3x. Especially as AI has driven a wave of capital efficiency — in ability to grow on fewer dollars and achieve profitably, if not in overall dollars raised — that have the potential to diminish reliance on mid-early-stage funders. This raises a tentative fear of a rise in inception-to-Series-A stories that could make traditional seed even less valuable.

On the greed side: mega funds are absurdly large—so large that returns targets traditionally associated with early-stage venture capital are hard to imagine at their AUM. But they still have the brand to convince founders they’re valuable partners at any stage. That said, the VC brand league tables are ever-shifting. We wrote about this in The Seed-Stage Reckoning and The Mind-Virus of Big Venture. While it seems like the same names have been dominant forever, that’s simply not true outside of a handful — it just especially seems that way as capital has concentrated into fewer, larger funds, following the 2022 downturn. Historically, it’s not uncommon for these perception shifts to occur as a result of drastic changes in AUM that move incentives, competencies, and (eventually) brand association out the stages at which that firm originally built its success.

The combined effect: round sizes have inflated across the board, and the relationship between capital raised and company stage has broken down. A $3M raise and a $30M raise can both describe pre-product companies with zero revenue. The amount of capital has become a significantly less reliable proxy for stage, traction, or quality.

This is the backdrop for our central argument: what VC calls “seed” is no longer a coherent stage. It’s a catch-all for everything from formation to Series A. We need a new perspective — most especially at the earliest stages, where institutional capital has only begun to play.

The Case for a Stage Shift

Periodic shifts in stage nomenclature are nothing new in the startup and venture capital ecosystem. Implicitly, they are happening continuously — explicitly, consensus seems to update every 3-10 years.

Before 1990, the landscape was simpler: friends-and-family money gave way to seed (more often “angel” rounds then), followed by Series A, B, and growth. Once coined, “Seed” meant what it sounded like—a small check to get started. From 1990 to 2005, through the dot-com boom and bust, the basic architecture held, but the scale of growth-stage capital ballooned. The bubble peaked at ~$90B deployed in 2000, a figure that wouldn’t be matched for two decades.

The first real structural shift came in 2005–2015, when Y Combinator, micro-VCs, and the SAFE note collectively redefined what “first round” investing looked like. Seed became institutional. It moved from angel territory — individuals writing $25–100K checks on a handshake — to a category with dedicated funds, standard terms, and follow-on expectations. Aileen Lee coined the term “unicorn” in 2013, and the implicit benchmark for what a seed company should become began to rise.

By 2015–2021, the ecosystem responded with two new stages. Pre-seed emerged around 2015 to describe $500K–$1.5M rounds for the earliest companies — the stage that seed used to occupy before it moved up. At the other end, pre-IPO rounds appeared as historically non-VC entrants moved in and companies stayed private longer, creating unique class of big-check capital that has since shifted to fewer, bigger, mostly foundational AI companies. Every existing stage shifted roughly one position later: what had been seed-stage companies were now raising Series As. What had been Series A companies were raising Bs.

This brings us to the present. Since 2021, the same rightward drift has continued—and accelerated. Today’s seed ($3–12M, institutional, often requiring traction) is functionally the old Series A. Today’s pre-seed ($750K-$4M, increasingly institutionalized) is the old seed. And that leaves a vacuum at the very beginning: the true formation stage, where a founder has conviction and a thesis but minimal product, minimal if any non-founding team, and little-to-no traction beyond design partnerships.

The stage for which the seed was invented no longer has a name.

Enter Inception

Inception is not a rebranding exercise. It describes something structurally new: for the first time, institutional funds—not accelerators, not angels, not friends and family—are providing genuine 1-on-1 partnership at true formation stages. Inception investing typically contemplates a $500K–$3M check, written pre-traction, pre-product, often pre-deck, and sometimes before a team is fully assembled. The investor’s value at this stage must go beyond the check: it’s operational co-building during the months when the company’s direction is still being set, iterating at high velocity, and arming the business with high-quality resources and guidance.

This stage is distinct from what pre-seed has become over the last decade. Pre-seed was supposed to fill this gap, but it has followed the same path as seed before it: it got institutionalized, bloated, and crowded. The median pre-seed round has roughly doubled since 2020. Funds that call themselves “pre-seed” now write $1–4M+ checks into companies that — often, if not always — have a product and early users. The label drifted from its original meaning, just as the seed did.

Many funds are talking about this vacuum and laying claim to inception. Few are actually dedicated to it, and even fewer are built for it. The gap between rhetoric and reality is revealing. If they’re unwilling to write a check before traction — or their fund model doesn’t support the time, concentration, and hands-on engagement that true formation-stage investing demands — they aren’t inception investors, at least as we see it. It’s also important to distinguish that, while accelerators nor incubators do check the box in terms of stage appetite, what we are talking about here is a net-new stage of institutional venture capital: the desire to build an active GP-founder relationship, bringing what once was only available to startups reaching Series A to true formation.

We believe an inception fund fundamentally differs from a pre-seed or seed fund. Deal sourcing requires novel origination. The portfolio must be concentrated enough to allow high-levels of support post-investment. The GP’s time per company is measured in days per month, not hours per quarter. The failure modes we often see at this stage are over-funding a company, which cascades into overspending by the startup before it is ready for scale. Another is writing a check and disappearing. This is the stage where VC-founder alignment matters most, because the investor’s behavior in the first six months shapes the company’s trajectory more than any follow-on round will.

Inception is the frontier of innovation in the venture capital industry — and to do it well at an institutional level requires not only the right risk appetite, but differentiated approaches to sourcing, winning, and supporting founders designed for a stage where top-of-funnel is not a deal, but people.

What Founders Should Ask VCs

If “seed” is a useless label and every stage has shifted, the first question for any founder raising capital is more basic than it sounds: where are you, really? Not what round number you’re on, not what your last investor called it, but what stage of company-building are you actually in—and what kind of investor is built for that stage?

Once you’ve identified where you actually are—not where you aspire to be in a year or two—three questions can help you cut through the noise with prospective investors:

1. What does success look like for your fund?

After a VC invests, what does success look like for them on both a fund level and portfolio-company level? A several-$100M seed fund writing $5M checks needs multi-billion-dollar outcomes at double-digit ownership for a single deal to matter. A mega-fund writing early checks needs decacorn outcomes for that first investment to represent more than an option. This is the crux of what we explored in The Quiet Death of VC-Founder Alignment—fund economics shape investor behavior far more than any outward positioning will reveal.

2. What would the ideal next step for us be?

For our company, what do you think an optimal next round would look like? What are some examples of that step for companies in your portfolio? How many companies in their last fund achieved that, and what did the typical trajectory look like along the way? Of course, if 70% of a S1 portfolio doesn’t make it to S2, that’s a meaningful signal about the kind of bets they’re making and their ability to help support their success. If they start talking about leading your Series As, that should raise a flag (what if they don’t?); if they expect you to skip straight to Series A, raising $20M+ or bust next, what does that mean for your optionality? A VC’s answer will speak volumes.

3. Are you valuable for my stage, or the next?

Every fund is optimized for a specific phase of company-building, and its value is typically sharpest in that phase. A fund set up to accelerate companies with a few million in revenue will excel at scaling out teams, optimizing go-to-market, and prepping for the maturity required of a business looking to launch into it’s scale-out phase. That's genuinely useful—if you're at that stage. But if you're a month out from formation, you need something entirely different: help iterating on the idea itself, perhaps making a founding engineering hire, warm introductions to seed (and potentially Series A GPs) for the next round, and bringing on your first advisors. Moreover, signal matters. And yes, usually that means brand and network. But consider that a big brand on your cap table at inception, that doesn’t participate in the next round, is a glaring red flag.

Make sure the investor you're choosing is valuable for where you are — and has the fund model to deliver that value — whether it’s tactical, tangible, or signal-based. You will, in all likelihood, bring further investors onto your cap table down the road: take care to partner with funds when and where they are most impactful.

Inception is the Frontier of VC Innovation

In just the last two decades, total VC AUM has expanded by an order of magnitude. The naming conventions we’re clinging to—pre-seed, seed, A, B—are artifacts of a time when the entire US VC industry deployed less in a year than a single mega-fund now manages. Every decade, the ecosystem has added a stage to account for what the previous terms no longer describe. We’re at that inflection point again.

Our case for inception isn’t simply about a label; it’s an argument about what the earliest stage of venture investing should look like. Not accelerators running cohorts of 200. Not angels writing checks on a napkin. Not pre-seed funds & seed funds masquerading as the same “product.” Inception is an institutional commitment to 1-on-1 partnership at the moment a company is being born—when the investor’s judgment, time, and operational engagement matter more than their capital.

With the exception of small partnerships that truly operate as a unit (rare above a few hundred million AUM) and programmatic funds like YC, VC is more of an individual sport than most realize. The GP on your board, answering your calls, by your side for a decade — that’s what matters. As we argued in Software Is Dead—Long Live Software, AI is reshaping not just which companies get funded but the entire business model landscape VCs invest in. The firms that adapt — by building their funds for the inception stage, not just talking about it — will earn the right to partner with the next generation of breakout founders from day zero.

Round names won’t ossify, nor ever be perfect. The age of Inception has just begun and deserves its own distinct consideration as a stage — but will eventually evolve too. Certainly, the capital efficiency enabled by modern AI is changing definitions faster than anyone can write. One piece of advice, however, is timeless: founders and allocators who continually interrogate the strategies and incentives beneath the nomenclature and find the right match for their goals will navigate this market with far more clarity than those who think one size still fits all.

Thanks for reading Euclid Insights! Additional sources here.1

Euclid is an inception-stage VC built for Vertical AI founders. If anyone in your network is working on a new startup in this space, we’d love to help. Just drop us a line via DM or in the comments below.

This episode is brought to you by Parafin. $14.5B in financing offers extended. $100M+ annualized revenue. $360M in fresh capital. 85% repeat rates from merchants. Parafin turns your platform’s transaction data into embedded capital that small businesses actually use — and come back for. Amazon, Walmart, DoorDash, Jobber, and Kajabi already have.

Click here to see what Parafin can do for your platform →

This Week’s Episode

There’s a good chance growth equity brings to mind a time-tested framework: analysts screen the top of the funnel, cold-call their way through, compete to sniff out hidden winners first, and slowly escalate conviction up to GPs. Along the way, the CRM is a filing cabinet — a retrospective store of data. Thesis and conviction live in hallways and IC meetings. And it works. But sometimes — and perhaps especially in this new era of AI acceleration — there’s risk of winners slipping for the wrong reasons… because someone once said “construction tech is too cyclical” or “gross margins below 85% are unacceptable” once and it stuck.

Dave Pandullo and Scott Hoch, General Partners at AQL Growth — a growth-stage fund launching from Frontier Growth focused exclusively on vertical software and vertical AI — decided to launch their firm with a different backbone. They built Eagle Vision, an internal AI sourcing platform that encodes their investment thesis into a system of judgment at scale. Their results are early but compelling: partners on planes before the first call, and a portfolio company (Albi) sourced, scored, and closed end-to-end.

This week, we share their story and a peek into their playbook. And we get their view on the future of systems of record, RCM, scribes, and much more.

This Week’s Guests

Dave & Scott of AQL Growth

AQL didn’t start with a blank sheet of paper. Frontier Growth has been investing in software since 1999, and the team’s vertical conviction traces back to 2007, when Frontier backed Daxko — club management software for YMCAs and JCCs that looked, at the time, like a “tiny little market.” They exited too early. Daxko went on to become the dominant platform in sports and recreation, now north of $250-300 million in ARR. It made a lasting impression: niche verticals that look small from the outside often harbor enormous, defensible TAM once you’re inside the workflow.

Dave — who previously co-founded a mobile telehealth software company — and Scott, a 20-year growth equity veteran who came up through investment banking, spent years building pattern recognition across verticals inside Frontier’s broader strategy. Trades, sports and entertainment, state and local government, outpatient healthcare — eight investments in the last 12 months alone. The thesis tightened: modern systems of record evolving into systems of action, AI-native wedge solutions, capital-efficient growth, and verticals with a right to win. The problem was that this thesis lived in partner conversations and judgment calls — not in a system.

The spark came when Scott ran a CRM audit and discovered, to his and Dave’s dismay, that it was “a filing cabinet with the wrong information in it.” They couldn’t identify their top 100 targets. They couldn’t answer basic questions: Who is our ICP? How many of them are there? How do we stack-rank them? So they reframed the problem entirely. Capital, they reasoned, is increasingly commoditized — it’s a product. Founders are the customer. And like any good vertical software company, AQL needed to know its total addressable market. That realization — treating themselves as a product company, not just a fund — became the genesis of Eagle Vision.

The name AQL itself, derived from Aquila (Latin for eagle), encodes the ethos: sharp vision, intense focus. They write $5-30M minority checks into post-product-market-fit companies at $2-3M+ ARR, and they needed their sourcing engine to match the precision of their strategy.

The Vertical Playbook

Building an AI-Powered Investment Engine

Step 1: Encode Your Thesis

AQL’s first attempt at AI sourcing “completely whiffed.” They fed a model their website, strategy, and past investments and asked it to evaluate targets. The output was unusable. The problem wasn’t the model — it was that their thesis lived in partner conversations and judgment calls, not in a system.

So they reversed the approach. Instead of asking AI to evaluate companies against a vague mandate, they codified exactly what they believe: modern systems of record evolving into systems of action, AI-native wedge solutions, capital-efficient growth, and verticals with a right to win. They then encoded those criteria into a scoring algorithm, treating founders as their ICP and capital as their product. As Scott put it: the goal was “not to replace our judgment, but to encode what we really believed.”

Step 2: Scan at Scale

With a codified thesis, AQL ran 25,000 software companies through Eagle Vision’s filters. The output: a TAM of 6,500 vertical software companies fitting their strategy. Overnight, they went from not being able to identify their top 100 targets to knowing exactly who their ICP was and how to stack-rank them. The system simultaneously generated IC-level content for each target — customer industry analysis, jobs-to-be-done breakdowns, ROI modeling, vertical and software value chain mapping — that Scott described as “deeper and richer than IC presentations I had seen in my 20-year career.”

Step 3: Build Your “Attractiveness Index”

Knowing which companies fit the thesis wasn’t enough. AQL needed to know which verticals were worth the bet — and how attractiveness shifted over time. They built a Vertical Attractiveness Index (VAI) scoring markets across multiple dimensions: density of modern systems of record, market sizing and white space, data moats around the solution type, competitive landscape, customer satisfaction with incumbents, product roadmap trajectories, M&A activity, capital raises, and competitive hires. Critically, the VAI is dynamic — recalculated as competitive moves, funding events, and talent shifts hit the market. The CRM went from a static filing cabinet to a living intelligence layer.

Step 4: Flip the Funnel

The traditional growth equity funnel runs: analyst screens → cold outreach → slow escalation → senior engagement. Eagle Vision inverts it. Because the scoring algorithm so tightly encodes partner-level judgment, “we’re getting partners on planes before we’ve even had a conversation with a company,” Dave explained. The new flow is: identify, prioritize, engage, sell, add value. AQL’s investment in Albi — a business management platform for damage restoration — was the proof point. Eagle Vision identified and prioritized it; the team moved fast, shared deep industry knowledge with founder Alex Duda on first contact, and saw past surface-level retention metrics that scared off other investors. Nine months post-close, retention had dramatically improved — exactly as the model predicted.

The Takeaway for Vertical Founders

AQL’s Eagle Vision isn’t a product they plan to sell — it’s their investment thesis encoded as software. But the meta-lesson applies to any vertical operator: the best AI won’t replace domain expertise, but rather systematize it. The most data or the biggest models certainly help. Startups that also abstract subjective judgment into scalable systems — turning that tribal, rarefied knowledge into an edge — are building truly powerful moats.

Keep Reading…

Why all paths aren’t equal in the search for AI-proof vertical moats.

Watch the full episode on YouTube

Jefferies coined it the “SaaSmageddon.” In the span of weeks, $400B of SaaS enterprise value evaporated — the category’s largest non-recessionary drawdown in over 30 years. ServiceNow down nearly 50% from its peak. A penny stock that sells karaoke machines announced a freight AI product and triggered a double-digit sell-off across logistics software.

The bears say SaaS is dead. We disagree — AI is still software, and the market is expanding. But even if the pie grows, the league tables will look different. In a world where five to ten credible vertical AI competitors exist in nearly every category, the question isn’t will software survive — it’s who wins? History has a clear answer: distribution. There were a hundred Toast-like companies. One captured 50-60% market share — not with the best product, but with feet on the street.

In this episode, we dive into the SaaSpocalypse with Nick Tippmann, the Founder & Managing Partner of TipTop VC. TipTop is an Austin-based first-check fund investing in vertical software and AI. Previously, he was the founding CMO at Greenlight Guru (zero to eight figures of ARR, $120M PE investment). Tune in to hear Nick’s his view on the public market turmoil, his journey as an operator-turned-vertical-VC, and the GTM motions separating the best-performing Vertical AI founders in our respective portfolios.

This Week’s Guest: Nick Tippmann

From Founding CMO to First-Check VC

Nick grew up entrepreneurial in small-town Indiana — always wanting to start a business like his dad and grandpa. By his senior year at Indiana University, studying entrepreneurship, he realized the question he’d come to college asking (”how do you start a business?”) wasn’t specific enough. The real question was how to build a high-growth, venture-backed tech startup — and most of those words were foreign to a kid from rural Indiana. Then he discovered the early VC blogging wave: Brad Feld, Fred Wilson, Mark Suster, Paul Graham. Three out of four had been founders first, then became VCs. That planted the seed. For the next 14 years, he operated with a dual mindset: build something great as a founder, then eventually make the jump to venture.

The building phase was Greenlight Guru. Nick joined as technically the first employee and founding CMO of the medtech vertical SaaS, which helped medical device companies get FDA approval and stay compliant post-market. Over an eight-and-a-half-year ride, he wore every hat the company needed: marketing, rev ops, product (once PE came in), and eventually strategy and corporate development. That cross-functional grind — from demand gen to M&A — that cross-functional experience taught him what many VCs learn secondhand: go-to-market isn’t a department, it’s a system. Greenlight Guru scaled from zero to eight figures of ARR and attracted a $120M private equity investment before Nick moved on.

That move was TipTop VC, an Austin-based first-check fund writing $100-400k checks alongside leads like (our fund) Euclid. Nick’s LPs and supporters have experience across vertical winners like Toast, Shopify, Procore, ServiceTitan, Mindbody, and others — operators who’ve built and sold the exact type of company TipTop invests in. For Nick, that collective doubles as a proprietary diligence and support network, giving portfolio companies direct access to founders who’ve already navigated the vertical playbook. And he’s got some nice early breakouts in the portfolio, including GC AI, which went from $1M to $10M ARR in under a year and raised a $60M Series B at a $555M valuation.

Vertical Playbook of the Week

Roundup: 7 Vertical AI GTM Hacks

Step 1: Instrument RevOps Like a Product. The fastest growers treat their GTM engine with the same discipline as a product backlog. Weekly demand gen meetings. Every channel owner reports their number with a red/yellow/green status — if you’re red, you present your action plan and report back in five days. Channels stop working in 60 to 90 days. The companies that win detect that early.

Step 2: Sell Revenue-Connected Wedges, Not “AI.” Below the enterprise, most vertical buyers don’t care that it’s AI — in blue-collar verticals, it can be a turnoff. The winning wedges connect directly to recovered revenue: scheduling AI that recaptures 10-30% lost to cancellations, voice AI that captures missed leads after-hours. You’re not selling artificial intelligence. You’re selling money back.

Step 3: Sell Way Ahead of Product. The best growers in TipTop’s portfolio consistently sell before the product is ready. They’ve done deep enough customer discovery to find the hair-on-fire problem where buyers will pay in advance: “I just want someone to solve this so bad, I will pay you right now.“

Step 4: Give Free Value-First Analysis. Use AI to generate a detailed free analysis for prospects before they’re customers. Santé (a portco of Nick’s whose founder was a recent guest on Verticals) takes this furthest — downloading a prospect’s data file and showing the system pre-populated with their historical data on a live demo. AI makes this kind of forward-deployed sales engineering scalable for the first time.

Step 5: Build Community-Led, Founder-Brand GTM. Industry associations. Educational webinars. Founder as figurehead on LinkedIn. The goal isn’t lead gen — it’s getting the industry to buy your 12-to-36-month roadmap, not just your current features. In a crowded vertical, brand and trust compound as moats in ways product alone cannot.

Step 6: Go Old School. Hammering the phones. Going to the random plumbing conference. Boots on the ground, handshakes. In vertical markets where distribution is a structural advantage, traditional sales often outperforms the “cool” growth motion. Toast won this way.

Step 7: Price on Micro-Outcomes, Not Macro Bets. Don’t start by telling a customer you’ll take 10% of every closed deal. Break the outcome into micro-conversions: the value of a booked meeting, an answered missed call, a rescheduled appointment. Test small value units first, then expand as predictability improves.

The Takeaway for Vertical Founders

The SaaSmageddon panic assumes AI shrinks the software market. At Euclid — and on Verticals — we think it dramatically expands it; especially in spaces where digitization remains low and buyers are leapfrogging from pen-and-paper to AI-native platforms. But expansion doesn’t mean everyone benefits. Nick’s message is simple: there is no silver bullet. The companies emerging from this moment are winning because they instrument early, measure relentlessly, and execute the fundamentals with more discipline than anyone else. When product differentiation narrows and the market is in panic, distribution is the moat.

Thanks for reading Euclid Insights and for tuning in to Verticals! Drop us a line below with your thoughts. And don’t forget to subscribe to the show wherever you watch and listen.

Keep Reading

For more on the “SaaSpocalypse,” check out Euclid’s most recent essay — we analyze the recent public-market SaaS downturn and answer the question: is SaaS really dead?

$400B of SaaS enterprise value evaporated this month. ServiceNow is down nearly 50% from last year’s peak, CRM is down 40%, and even MSFT approached a 30% drawdown. Nowhere is safe: even rumors of evidence of “AI eating X” are triggering sell-offs. A penny stock that previously sold music jukeboxes and announced a freight product caused a double-digit sell-off in logistics companies.

Rarely can a broad public market sell-off be attributed to one cause. And, of course, this is no exception. It’s first important to recognize all the reasons outside of “AI is killing all software” we might be seeing trades out of the space:

1. Discrete Disruption: Beyond Cowork, both Anthropic and OpenAI announced initiatives targeting the legal and financial services industries, which hit players in those spheres (such as Thomson Reuters, LegalZoom, and CS Disco) hard. Moreover, some individual software names simply failed to deliver: for example, a miss from the normally reliable Tyler Technologies (down 15%). Perhaps not too much for a 60-year-old software & services business with longstanding monopolies in govtech, now facing a brave new world of public sector AI adoption.

2. Hyperscaler Capex: Massive increases in AI-related capital expenditure from Microsoft, Alphabet, Meta, and Amazon — forecasted to exceed $660B combined in 2026 — have spooked investors regarding when these investments will yield significant monetization. MSFT, for example, was hit thanks to both rising CapEx and Azure deceleration. We have written about this in our past essay on AI infrastructure and the idea of an “AI Bubble” here.

3. Sector Rotation: A general zeitgeist of overvaluation and need for a correction in technology equities (though with software now trading at roughly 4.9x EV/Revenue, it feels like overcorrection if for that reason alone). Some also speculate that stronger-than-expected economic data (e.g., the January payrolls report showing 130k jobs added) led investors to rotate from “capital-light” software into “capital-intensive” cyclical sectors like energy and industrials.

4. Size-Based Volatility: Small- and mid-caps generally pull down more in a market rout. Historically, this is especially true in software, a category whose aggregate EV is dominated by a few massive platforms at the top. The very long tail of sub-$5B EV software publics tends to get hit hard. During the Great Recession, NetSuite and Salesforce saw their NTM revenue multiples collapse by ~92% and ~80%, respectively.2 While — counter to popular opinion — vertical software companies have a ~20% higher mean Enterprise Value than their horizontal peers, the industries targeted by this market reaction have more vertical players with perceived un-diversified exposure.

Rational or not, the sell-off has led some VCs to publicly declare that in the wake of AI, SaaS revenue durability is challenged, if not extinct. The most bearish of “hot takes” is that vibe-coded applications will make core systems of record — or perhaps all application-layer software — obsolete.

Before getting into long-term viability, let’s dig a bit further into the numbers of the “SaaS flash crash.”

We analyzed the recent public-market SaaS drawdown, and while vertical names have been hit harder this month than their horizontal peers, the declines from last year’s highs are fairly similar. Lastly, we tried to identify which vertical sectors were hardest hit; perhaps unsurprisingly (given OpenAI and Anthropic news), the hardest-hit sectors are Legal and Healthcare, although the small n in each bucket means two hard-hit names have a particularly large impact (CS Disco in Legal and Doximity in HC).

As we highlighted above, the strongest correlation is company size (i.e., market cap). Small caps have been hit the hardest — while the delta between large and mid-cap is more modest.

Ultimately, a discrete 10% drop in software stocks tells us approximately nothing about the future of SaaS as a business model or technology paradigm. To use the famous quote attributed to Ben Graham: “in the short run, the market is a voting machine, but in the long run, it is a weighing machine.” 5 years from now, equity trend will surely be more instructive as to AI’s impact on SaaS. We think, however, there are a few valid conclusions we can draw in the meantime.

SaaS is Very Much Alive

While we absolutely agree that AI poses a material risk to incumbent software vendors, we do not share the consensus that “SaaS is dead” (nor that a jukebox company is a material threat to the freight market). While we hate to lead our counter-argument with pedantry, the simplest refutation of the “death of software” argument is this: AI is still software. And application-layer AI delivered in the cloud is still Software-as-a-Service.

Naturally, every major technological paradigm shift comes with new terminology. VCs, eager to signal thought leadership and capture mindshare, are as guilty of flooding the zone with new acronyms and techno-jargon as anyone else. While AI has come to replace SaaS as a shorthand, the technology delivery model SaaS was primarily meant to convey remains just as prevalent in the LLM era—even if other attributes typically associated with SaaS (such as heavy UI or seat-based subscriptions) are receding.

EvenUp, Abridge, Harvey, OpenEvidence, Rilla, MagicSchool: all AI, all SaaS. If we started calling AI “next-gen SaaS,” it might clear up a lot of confusion. When people hail the death of SaaS, many conflate the rise of a new generation with the demise of the larger underlying business model, which is stronger than ever. Calling AI the software-killer is like calling the iPhone the end of computing. In fact, like in that case, by making software easier and more accessible, AI is likely its biggest accelerant since the advent of cloud infrastructure itself.

As we wrote in our essay last week, though: just because a pie is growing, doesn’t mean everyone gets to eat. Even if SaaS growth continues unabated, the league tables in 5-10 years will almost certainly look different.

Who is Vulnerable?

In our piece last June on the threat to Vertical SaaS from AI, we highlighted three categories that we felt were particularly vulnerable:

1. Hyper-Narrow: Developers of extremely niche tools (in use case, more than necessarily market size) that are easily replicable with AI support. These are the nearest-term, clearest losers.

2. Obsoletion by LLMs: SaaS offerings that LLMs naturally subsume (e.g., Chegg, Grammarly, Dragon). That said, companies with the resources and willpower to re-invent themselves AI-first may thrive. Duolingo is talking this talk.

3. Hesitant Incumbents: Legacy SaaS providers who simply don’t offer AI functionality at parity quickly enough, or hold on too tightly to outmoded cash-cow products. Incentives matter. We envisioned these slow deaths in our essay on the development of intelligence systems.

We define hyper-narrow here to mean more of a function of an individual product’s scale, complexity, and defensibility versus narrow in scope. A vertical system of record, regardless of TAM, would not meet the definition. A horizontal customer service tool would. Regardless, we concede there’s genuine pressure on every app-layer company, as AI-native challengers are growing faster than ever before. And if past platform shifts are any indication of the future, it is almost guaranteed that the vast majority of incremental value will go to AI-native disruptors rather than legacy vendors. Incumbents move slowly. They can’t attract the best AI talent. They face the classic innovator’s dilemma: deciding whether to cannibalize their core business. A small subset of legacy SaaS companies will be able to weather the storm. Many will not.

And there’s a real structural threat you might call the “front door” risk. Classic software gets reduced to middleware, with agents capturing all the incremental value on top. Aaron Levie, CEO of Box, put it bluntly: every piece of enterprise software has to become a platform play because agents will do vastly more work with tools and data than people ever did. The system of record gets pushed down the stack — and that means lower growth potential and a smaller profit pool. It’s worth, of course, pointing out that the “integrate and surround” form of system-of-record disruption is by no means new. As we explored in our analysis of Vertical Software’s Integration Problem, incumbents facing this dynamic have either adapted (buy or build) or faded in relevance.

Not capturing the next phase of explosive growth doesn’t mean all growth disappears. If software were truly in its late-cycle period, it would be growing at close to GDP (it’s at least 2-3x that). In fact, Gartner expects software spending to accelerate in 2026 — the highest figure in half a decade. There is still significant growth left, even in classic cloud migrations: while 95 percent of enterprises have a cloud footprint, the share of enterprise workloads in the cloud only crossed 50% in 2025. That figure was just north of 30% eight years ago.

We’ve seen this exact movie before, and recently. Remember when the shift from on-premise to cloud was supposed to kill the enterprise software giants? Oracle, SAP, and Microsoft were all declared dinosaurs — legacy vendors that would be disrupted by nimble cloud-native startups. Instead, the successful incumbents pivoted by buying and building cloud businesses; today, they are bigger than ever. Microsoft’s revenue rose from $62B in 2010 to $282B in 2025. Oracle went from $12B to $57B over the past two decades; its cloud software business is 3x larger than the on-premise business ever was. The on-prem-to-cloud shift certainly killed many incumbents, but more importantly, it supercharged the growth of the enterprise software market. Cheaper to build, cheaper to sell, and cheaper to maintain is deflationary but also market-expanding. Global software markets were ~$200B twenty years ago; today, they're more than five times that size.

Parker Conrad, CEO of Rippling, reached the same insight, noting the impact of dramatically lower software development costs over the past two decades. AWS replaced self-hosting. Frameworks, open-source ecosystems, developer tools, and higher-level programming languages abstracted away complexity, making it possible to build in days what used to take months. The result was a Cambrian explosion of software businesses. But this did not make software companies less valuable. It increased competition, for sure. It made incumbency advantages weaker. And of course, it shortened the replacement buying cadence for many enterprises. But in aggregate, the net result is that it made software more valuable. The SaaS era that followed produced trillions of dollars in enterprise value precisely because building got cheaper. More problems could be addressed. More workflows could be institutionalized. More industries could be digitized. More customer segments (whether vertical or smaller businesses) could be sold software more efficiently.

We traced this pattern in our earlier piece on whether AI threatens vertical SaaS, examining the history of “layer commoditization cycles” from FORTRAN to cloud computing. Each cycle follows the same arc: innovation automates core aspects of a process, pushback follows as incumbents fear obsolescence, and then the market thrives with newfound productivity. LLMs are the latest iteration, and, as in every prior cycle, there will be losers, but the market as a whole grows.

The Vibe Code Fallacy

Benedict Evans made the point around the complexity of building software in his recent Stratechery interview: “I never thought that the hard part of building a piece of software that manages the thing for the other thing deep inside a big company was writing the code.” The hard part is discovering the problem, designing the right workflow, building a go-to-market strategy, and earning customer trust. And then you layer in the sheer volume of applications: the average number of SaaS tools across all buyers is 275. That number jumps to over 600 for enterprises.

It’s easy to dismiss the complexity of every SaaS app as consumable by AI, but as Evans points out, the same analogy can be applied to traditional SaaS: “they’re all a thin-SQL wrapper. They’re all just databases.” Extending the argument further, Evans adds: “It’s like the dumb Hacker News comment, ‘Airbnb is just a CMS.’ But yeah, every social network is just a CMS.”

If AI makes code cheaper, it makes the non-code moats — domain expertise, distribution, customer relationships, and what Evans memorably called “a throat to choke” — relatively more important, not less. And in the age of AI, that throat to choke becomes ever more valuable. When the output is probabilistic rather than deterministic, customers need someone accountable for the result. This is precisely why we believe the durable moats in vertical software — workflow entrenchment, proprietary data loops, and trust — compound rather than erode as AI advances.

One persistent fantasy is that enterprises will simply vibe-code their way out of buying software. Evans called this idea “delusional” in his interview, and we completely agree, especially in vertical markets. Companies aren’t in the business of building complex internal tools. Models improve extraordinarily quickly, and it would be naive to underestimate what vibe coding will achieve in two or three years. But the fundamental attitude of “let’s build internally versus pay someone to manage it” won’t change. Put aside the complexity & quality argument for a moment; the total cost of ownership will still almost always favor buying over building.

While this is especially true for vertical market companies, even cutting-edge tech startups “outsource” much of their stack. Why? Because it’s cheaper than dedicating staff to such tasks, and they can move faster on solving core problems. The net result is that cheaper-to-build software is usually cheaper to own, resulting in more prospective customers and larger markets. Companies will continue to outsource software to vendors — whether those vendors are legacy SaaS providers or new AI-native startups. We thought one particular curious anecdote around is that vibe-coding breakout Loveable just became a customer of HubSpot.

We used the analogy of super-sneakers in our earlier essay: imagine shoes that give every basketball player a lifetime of conditioning overnight. High-school bench warmers become NCAA-competitive. Even NBA pros improve. But the impact is logarithmic — the Currys and LeBrons already have a lifetime of training, and much of their edge is basketball IQ. The higher-level you already are, the less it helps. And everyone gets the shoes. The same is true for software businesses.

It’s also important to remember that digitization remains low in most vertical markets. These industries moved from pen-and-paper to Excel to off-the-shelf vertical software. And just like the SaaS revolution, many prospective vertical customers that were left behind will jump straight from having no software for certain workflows to buying AI-native platforms. We explored this adoption pattern in greater depth in our piece on Diffusion of AI across Vertical Markets.

We’ll share an anecdote that reflects a similar challenge we often see, even among the largest prospective vertical buyers, and one most tech-forward people have little exposure to. We were talking with a multi-billion-dollar construction materials business that had three different ERPs. Why? They were growing quickly, both organically and inorganically, and when they couldn’t figure out how to migrate off their multi-decade-old system, they decided to keep it installed for certain regions and workflows, buy a new system for organic growth, and eventually not bother replacing the third system they inherited from M&A. While they have no plans to change this IT stack, they are actively evaluating and purchasing new pieces of Vertical AI. These industries are not leapfrogging to prompt-engineer their own tools anytime soon, but they are now active buyers of new tooling. That same company just bought its first CRM (though they’d call it something else), along with a voice-first revenue acceleration stack.

What Comes Next

Not all software is created equal, and the panic is treating it as monolithic. Ben Thompson made a key observation in his Stratechery interview: “The destruction and the value creation are not simultaneous in time. First, there’s a lot of destruction, the unbundling aspect, and then people are freed up to figure out the new thing, and that’s when the value creation happens.”

We are still quite early in the destruction phase, despite the market panic and stock prices cratering. Some vertical software companies will die, though we think the durability of those names is far higher than that of their horizontal peers. Some of the next generation of vertical AI companies will be built on the rubble, but most of them will be built on green fields. The market, by definition, cannot expand if it’s just ripping and replacing legacy vendors. And it’s not hopeless for existing legacy vendors: incumbents with distribution and trust advantages can capture their fair share of AI revenue, too. The recent history of vertical profit premiums suggests the market already prices in the durability advantages that vertical incumbents enjoy.

As we explored in Dude, Where’s My Moat?, the immutable primitives of software defensibility are workflow and data. Speed is a compounding advantage in the early days of a category, but it is not a durable moat. At scale, almost all companies slow down — and that’s where data gravity, brand and trust, and platform lock-in take over. For Vertical AI, the path to defensibility runs through what we call the System of Intelligence: becoming the authoring layer for high-value workflows, building data loops from customer usage, and expanding into adjacent products that compound switching costs. The startups that build toward these moats early will be the ones still standing when commoditization arrives. Equally, the incumbents who survive—and perhaps even thrive—long-term are those who leverage brand and data gravity to build a durable moat.

As we noted in our earlier analysis of the AI threat to Vertical SaaS, the specific threat level is industry agnostic. It comes down to a product’s scale, complexity, and defensibility. Hyper-narrow tools are the nearest-term losers. Products whose core function LLMs naturally subsume — Chegg, Grammarly, Dragon — are next. And hesitant incumbents who don’t adopt AI fast enough will bleed share to AI-native challengers. But vertical technology isn’t any more affected than SaaS as a whole — in many ways, it’s more defensible.

Nicolas Bustamante of Fintool shared a simple framework for evaluating the strength of Vertical Software incumbents:

1. Is the data proprietary?

2. Is there regulatory lock-in?

3. Is the software embedded in the transaction?

He explains, “Zero ‘yes’ answers: high risk. One: medium risk. Two or three: you’re probably fine.” The idea that CRMs, property management systems, or ERPs will fall into a black hole is not something we have seen happen. Workers shifting off document management or project management tools? We’d certainly expect that to occur on a medium-term horizon.

The bears assume that AI will shrink the software market. We think it dramatically expands it, even more so in vertical categories. As we wrote in Vertical AI & the Productivity Paradox, the vertical AI opportunity should be measured not just as labor replacement, but as labor enhancement plus market expansion:

Consider a hypothetical industry where AI can double profit margins by automating labor. Now, imagine the same sector where incremental profits are reinvested into growth. We have the same productivity metric, but the same industry is now >20% larger. Re-investing savings over a five-year window results in a market nearly five times the size with the same productivity. Of course, firms in such a market would likely reinvest only some of their savings into growth, but the net result is a larger and more productive firm (and market).

Foundation model verticalization is real and accelerating. Companies with weak moats and hyper-narrow tooling are exposed. But defensibility is preserved when outcomes require coordination with real-world systems and build the data loops and workflow entrenchment that compound over time. As we wrote in our piece last year on Diffusion of AI across Vertical Markets, we view the primary limiting factor today as not technological, but human capital:

It is these comprehensive end-to-end agentic transformations that are essential to fully revolutionize vertical markets and fulfill the economic potential of LLMs. As such, for founders developing in vertical AI, the hard work must be highly specialized and market-centric. No doubt, technology will continue to advance, but we need vertical builders to see and build the future.

At Euclid, we’re investing in founders who see how AI can transform industries with the lowest digitization and the largest productivity gaps: the complex, gnarly, high-accountability coordination between AI and the real world. This is where the next trillion dollars of enterprise value gets created. And history suggests it won't stop at just one.

Software is dead. Long live software.

Thanks for reading Euclid Insights! Additional sources here.3

Euclid is a VC partnering with Vertical AI founders at inception. If anyone in your network is working on a new startup in this space, we’d love to help. Just drop us a line via DM or in the comments below.