Today, we’re pleased to share a special episode of VERTICALS to round out the year.
This one brings together the full Euclid team (Nic & Omar), show co-host Luke Sophinos, and repeat founder / CEO show guest, Todd Saunders. Join us as we do what we always do in December: look ahead, disagree loudly, and—for better or worse—commit our takes to the public record.
Brace yourself for our 2026 predictions. We run through a bunch of categories:
Biggest Vertical AI Winners & Losers
Biggest Old-Guard Winners & Losers
Most Successful IPOs
Top Under-the-Radar Trends
Wildcards
# Vertical Unicorns Minted
Have a wonderful rest of the holiday season and we’ll see you in the new year!
Biggest Vertical AI Winners of 2026
Open Evidence — 2026 is the year we see this PLG-led vertical AI business flip on monetization and continue to crush it. Getting to ~half of U.S. physicians in under two years, with essentially zero revenue, might normally be waved off as consumer-like, with questionable retention and moat. This is one of the rare cases where distribution may be the moat. The second important aspect is that Open Evidence will seek to monetize not with consumers, but with pharma companies. Similar to public vertical software company Doximity, pharma has enormous budgets to reach doctors at the point of intent — and Open Evidence’s product is perfectly positioned to capture that spend. The meta prediction is that Open Evidence becomes the paragon for vertical AI companies that achieve exponential PLG-driven growth and rapidly parlay it into nine-figure revenue, reshaping how investors think about usage-first strategies in vertical AI (Luke).
Abridge — A bit of a contrarian take given Epic’s recent announcement of a competing scribe. But this is one of the few voice AI categories where the complexity of integrations and partnerships creates defensibility that other wedges lack. This prediction isn’t solely about Abridge — though its enterprise footprint makes it particularly well positioned. Healthcare is roughly a quarter of the U.S. economy, which means there is room for many winners. As discussed on the episode, companies like Ambience and Freed also matter, and there are dozens of clinical sub-markets with 10,000+ practitioners that remain largely untouched. In 2026, we expect Abridge and others to successfully expand beyond scribes, while incumbents struggle to innovate against institutional inertia (Nic).
EvenUp — This vertical AI company took legal tech by storm from a different angle: plaintiff law. It scaled rapidly by selling outcomes — auto-drafting time-consuming but revenue-critical demand letters — rather than software. In doing so, it may have earned the right to become the operating system for the category, where others previously failed. That approach accomplished two things at once: it collapsed buyer skepticism (“does this tool work?” becomes “did I get paid?”) and decoupled growth from seat count. The 2026 bet is that this playbook generalizes, and that the next breakout vertical AI companies start to look less like SaaS and more like productized services — ultimately evolving into diversified operating systems aligned with traditional SaaS ambitions (Omar).
GCAI — This prediction reflects a broader reframing of what “legal AI” actually encompasses. Most attention and hype has gone toward selling into Big Law, but the more scalable and repeatable opportunity may sit elsewhere — particularly with in-house general counsel. Inside companies, legal work is less bespoke, more operational, and tightly coupled to core business workflows. Just as importantly, incentives differ: Big Law has little reason to increase efficiency when billing by the hour, while corporate and plaintiff-side legal teams benefit directly from faster, cheaper outcomes. Markets like these can both reduce costs and increase revenue through efficiency, making them ideal targets for vertical AI (Luke).
Biggest Vertical AI Losers of 2026
Salesforce — Their AI strategy continues to flail. While CRM made bold early proclamations around shifting customers over to an agentic, usage-based model, recent earnings calls have revealed anemic uptake. Many see it as branding without coherence. At the same time, pressure is mounting from vertical AI players targeting the legacy CRM from multiple angles. Salesforce will be the classic example of death by a thousand cuts as better-tailored vertical AI platforms offer not only more industry specialization, but also more agility in AI-forward offerings (Omar).
Reddit — Risks deteriorating further into internet slop as bots talk to bots and erode trust. While there’s disagreement here — Reddit remains one of the few places people still find authentic human perspectives — AI-generated content is clearly deleterious to the business model. Interest from OpenAI and Google in Reddit’s data underscores its value. The open question is whether management can protect that core asset at all costs. We’ll see how it plays out (Todd).
ServiceTitan — This one drew some major disagreement from the crew. Todd envisioned a TTAN squeezed from both sides: pricing pressure below, complexity above, and too many focused competitors attacking the middle of the market. They’re well into their inorganic growth phase and may find themselves too “horizontal” for a category (trades) that’s big enough to support multiple, more specialized winners. At the same time, Omar underscored the massive continuing opportunity in trades. Even basic SaaS adoption has a long way to go in this space, and ServiceTitan is certainly the leading brand here. (Todd / Omar)
Old-Guard Losers in 2026
C3.ai — Rebranding into the AI moment can only carry a company so far if fundamentals don’t cooperate. Despite early hype and strong narrative positioning, growth, retention, and customer concentration issues continue to undercut the story. The problem isn’t that the company lacks AI — it’s that “AI” alone doesn’t solve weak product-market fit or narrow use cases. As competitors unbundle and verticalize, the risk is continued pressure leading to unbundling, a distressed outcome, or a forced strategic rethink — a reminder that branding momentum eventually bows to operational reality (Omar).
DocuSign — Drifting toward an inevitable takeout, not because e-signature disappears, but because category defensibility has quietly evaporated. Signing documents has become a feature, not a product: cheap, embedded everywhere, and increasingly “good enough.” Meanwhile, the company’s scale, headcount, and public-market posture look mismatched to what users actually need. The brand still carries weight (“DocuSign” remains a verb, after all) but the moat does not. After more public-market battering, it becomes an obvious acquisition: a valuable name and predictable cash flows, but no credible standalone growth story (Todd).
Blackbaud — A slow-growing, bloated business weighed down by legacy products in a nonprofit vertical that’s quietly being re-served by more modern, workflow-native software. Nothing about this profile benefits from public-market scrutiny: low growth, limited innovation velocity, and customers who value stability over speed. The prediction isn’t dramatic disruption, but inevitability — a private-equity take-private where cost rationalization and incremental modernization make far more sense than pretending it’s a growth story (Luke).
PowerSchool — A more forward-looking and defensive call. PowerSchool sits in an increasingly uncomfortable middle ground: a system of record with real distribution advantages, but one that’s starting to feel exposed as education-focused AI products creep closer to core workflows. While selling into schools remains brutally hard, AI-native tools are no longer content to live at the edges (lesson planning, tutoring, admin assistance). Over time, they push inward — toward scheduling, assessment, reporting, and data flows that PowerSchool historically controlled. The risk isn’t an overnight collapse; Vista ownership buys time, not immunity, and 2026 is when that pressure starts to become visible (Nic).
Old-Guard Winners in 2026
Apple — Some feel Apple’s consumer products have only marginal gains ahead (iPhone) or have failed to launch altogether (Vision). They’ve been comparatively quiet on AI research, and Apple Intelligence and Siri haven’t made the same waves as peers. At the same time, they’re sitting on the world’s largest edge-inference footprint. The prediction is that this becomes a major strategic advantage in 2026, even if it hasn’t shown up yet in obvious product wins (Todd).
Veeva — AI Agents quietly add massive ARR via cross-sell into an entrenched life-sciences customer base. Veeva’s vertical specificity has built a solid, defensible moat despite it’s less-defensible CRM origins. Disciplined execution inside a vertical with high workflow density and real willingness to pay, plus strong apparent management, makes Veeva a likely continued winner in 2026 (Luke).
Autodesk — Its products sit at the absolute center of mission-critical workflows in architecture, engineering, construction, and manufacturing. The question is, will Autodesk catch up with home-grown AI, or they start to see disruption? There are three main reasons we think ASDK will maintain their leverage in 2026: (1) Low ACVs with a massive, global user base makes it hard to compete with them on price; (2) high product complexity that is visual in nature and tough to replicate via NLP interfaces alone; (3) excellent M&A instinct, with a proven history of acquiring relevant emergents in new spaces and incorporating them successfully. (Nic)
Google — Gemini has largely caught up, Waymo is the autonomous-vehicle leader, and they’ve barely scratched the surface of AI integration into the G-Suite. The open question is whether Google can launch net-new AI products that truly break out. This is probably the most consensus take of the group, even if the timing remains uncertain (Nic).
NVIDIA — Amongst many other developments, their massive scale bolstered “AI bubble” talk in 2025. The view here is that NVIDIA remains strong into 2026. The company generates serious cash flow, is arguably not overvalued relative to its position, and sits at the very top of the AI value chain with seriously lock-in. Regardless of which application-layer companies win or fail, NVIDIA remains one of the best-positioned picks-and-shovels plays in AI. They’ve also proven their ability to launch new, in-demand products fueling frontier model battles. This does not mean that they’ll maintain their grip forever. Google’s TorchTPU (supported by Meta) is a serious attempt to chip away at NVIDIA’s CUDA moat1 by making PyTorch run natively on TPUs, which could then take over some share of AI workloads. Realistically, though, impact from such initiatives are likely on a >12 month timeline—and we don’t think data center demand is slowing down in the meantime (Nic / Omar).
2026 IPOs That Will Outperform
Anduril — Potentially the most explosive IPO of the group. Defense spending, autonomy, and geopolitics converge into a story retail can easily understand. If it lists, it trades on both narrative and real revenue, which is a rare combination in modern tech IPOs, even if regulatory reliance is long-term worrying (Luke).
Waymo — The clear market leader in a massive category, and unlikely to live inside Alphabet forever. The core question isn’t technology, but distribution: if non-LiDAR autonomy from Tesla proves comparable, does that change the long-term advantage? Still, once consumers experience Waymo firsthand, demand to own the stock is almost inevitable (Todd).
Canva (if it goes public) — A steady, retail-friendly IPO if it happens. The product is deeply embedded, the brand is ubiquitous, and users don’t churn, even if AI execution has lagged. This one ultimately comes down to management: there is time to catch up on AI, but not unlimited time. Retail will want a credible AI growth narrative (Nic).
Stripe (if it ever goes public) — The ultimate “doesn’t need the market” IPO. If it lists, it’s because it chooses to, not because it has to. The fundamentals are obvious. The only real questions are timing and pricing (Luke).
Under-the-Radar Trends for 2026
AI Meta-Themes: Access, Memory & Inception — While smarter models will remain a focus, this year it will become clear that access is as big of a bottleneck for AI: getting AI-native systems the right clean, persistent, permissioned data. We’ll see real progress from better access layers, whether those take the form of data-aggregators like Parallel, agent data protocols like MCP, and built-in AI-powered tooling to make leveraging your own data easier (not to mention, continued cuts in the time to build integrations).
Second, we will see massive unlocks in AI performance through new approaches to memory. LLMs work because they pay “attention” to a whole conversation—the more tokens it holds in its “context window,” however, the more costly the inference. We’ve all had the experience of ChatGPT making something up before revealing that it actually “no longer has access to” about the document you uploaded 20 back-and-forths ago. Human brains, however, do quite nicely without having every single fact we ever learned at immediate recall. New approaches to memory management and selective recall are in the works: retrieval-native models, virtual context managers, long-context attention, etc. In 2026, we expect such improvements in AI memory will lead to step-function performance jumps, without more training scale.
Finally, AI enables an era of inception software. We use “inception” here in the Christopher Nolan sense, not in the startup stage sense. Historically, customization of products to specific clients was unappetizing to SaaS providers, relegated to either (1) a super-high-end, expensive, consultative service offered to mega-enterprises (Palantir being an exception that embraced it) or (2) farmed out to third-party consultants (think SIs and MSPs specializing in SalesForce or ServiceNow). AI, however, brings the marginal cost of basic development to zero—we’ve already seen it in Lovable, Replit, Framer, et al. Why would startups not enable some of these “creation” features in-platform, by default? From personalized UIs to custom integrations to wholly novel sub-applications, AI “builders” embedded into software platforms could enable customers to self-serve customize (and even innovate). Imagine if enterprises could build on top of best-of-breed, creating fully proprietary elements in-house, without shelling out millions to Accenture etc. We imagine 2026 could be the beginning of the end for homogenous software. (Nic)
The decline of copycat Vertical AI — In 2025, “copilot-for-X,” “scribe-for-X,” “receptionist-for-X” won’t be enough. Specifically, we mean point-solution wedge products that solve a real problem, but lack strong paths to becoming (or tightly integrating with) systems of record, and hence fail to build durable moats. Meanwhile, the difficulty of implementing a basic RAG or voice agent will come down substantially. Finally, point solutions without data control will face the coming (in some cases, ongoing) backlash from incumbent systems of record as they stiff-arm would-be disruptors through blocking of API access, lawsuits, competitive partnerships, and even wholesale product replication (Omar / Luke).
“Deep tech” fragments as a category — Particularly from a VC perspective. There are certain carve-outs here, such as biotech, which has always been distinct from software. It’s often unclear what “deep tech” even includes. In this case, we generally mean“atoms vs. bits” business models that seek differentiation through fundamental IP advances rather than business model or distribution. To some extent, this is a spicy take because there is real regulatory momentum behind certain categories (e.g., defense, though that is probably more rightfully it’s own category). But broadly, we see Deep Tech as a case where venture exuberance created a synthetic category that is heterogeneous and poorly understood: mostly capital-intensive plays with unclear liquidity paths, high regulatory dependency, binary outcomes. Expect it to fragment as certain sub-categories underperform and others remain interesting (Nic).
Headwinds build against roll-ups — The prediction is that roll-ups face increasing structural pressure as more capital crowds the strategy. Roll-ups are, at their core, financial engineering plays: buy fragmented assets, aggregate them, and arbitrage multiples. As venture and crossover capital pours in, asset prices rise, returns compress, and the margin for error disappears. What once worked with discipline and patience starts to break under competition and overcapitalization. The call isn’t that roll-ups stop working entirely — it’s that many fail to produce venture-scale outcomes once capital intensity, execution risk, and slower compounding are properly priced in (Omar).
Retail AI blooms — Retail offers a highly attractive trifecta for AI: hyper—fragmentation, massive scale, and clearly LLM applicability. Our view is that 2026 produces at least one vertical AI company in retail that goes from near-zero to nine-figure revenue shockingly fast. Not because retail is “easy,” but because LLMs unlock workflows that were previously impossible: abstracting supply-demand intent through more seamless voice and text interaction, solving discovery in new ways. Retailers already live in thin-margin, high-volume environments, so when a solution can materially improve conversion, inventory turns, or procurement efficiency, adoption can spread quickly. Sellers are also buyers, with procurement driving revenue and cost directly. And the numbers can get big on the B2B retail side — Shopify, Faire, Doordash, Whatnot, etc. all demonstrate what happens when tech enables net-new channels of buying and selling. Future winners will likely be more specialized and verticalized than these predecessors… a nice segue for our next prediction (Nic).
Sub-vertical explosion — Not “restaurants,” but pizza; not “home services,” but remodeling; not “behavioral health,” but ABA or addiction. The prediction is that founders finally slice markets finely enough to achieve real density, distribution, and product depth. Expect 50+ pre-seed and seed rounds in ultra-narrow categories where incumbents feel “too small” to bother competing — until it’s too late (Luke).
World models enter the mainstream vocabulary — LLMs understand human language, but do they really “understand” how the world works? If you ask a frontier model to solve a college-level engineering math question, it will almost certainly excel. If you ask it to design a building that is proven by engineers to be impossible, it will likely still give you a design—because while it has embeddings for all the constituent concepts and terms, it wasn’t trained to understand the physics on a non-semantic fundamental level… but it was trained to give you a facially credible answer. Some argue that with additional scale, world models (or neuro-symbolic understanding in well-documented contexts, at least) will be an emergent properties of LLMs. The next wave of Vertical AI will likely require domain-specific models / tools that encode symbolic understanding (e.g. science, medicine, engineering, finance). Many are already pushing the frontiers here: Google (Deepmind’s MuZero & Genie), Wayve, NVIDIA (Isaac), a TBA startup from Yann LeCun. By the end of 2026, we predict “world models,” “symbolic AI,” or some equivalent term for the tech will enter the mainstream conversation (Nic).
Religion AI — A massive, fragmented market running on outdated tools, with enormous operational complexity hiding behind the scenes. Donations, payments, programming, community management, and compliance all move real money, yet software penetration remains shockingly low. It’s unsexy, misunderstood, and exactly why the opportunity still exists for SaaS. We haven’t seen Vertical AI emerge visibly here just yet, and we expect that to change (Todd). Some of us also guess that religion itself will get more involved in AI in 2026: either by issuing ecclesiastical guidance around AI interaction or morality; or sanctioning AI that empowers worshippers in some way (Nic). Net-new AI-centered religions? We’ll revisit that one next year.
The AI churn apocalypse — Rapid growth has masked weak gross retention across both vertical and non-vertical AI with PLG motions. In 2026, markets start to notice that you can’t compound revenue if customers aren’t loyal—and options to switch will abound. There will be knife fights on pricing, and some markets will be races to the bottom. Non-mission-critical experimentation budgets will dry up and renewal math becomes unavoidable (Omar).
GovTech hits a breaking point — The group agrees 2026 will be a pivotal time for GovTech, but disagrees about the outcome. Pros: public sector AI will take off not because procurement suddenly gets easier, but because citizens stop tolerating broken digital experiences as they see how LLMs can make interaction seamless in their professional and personal lives. AI raises the minimum acceptable standard for service delivery, creating external pressure governments can’t ignore. Moreover, regulatory momentum for adopting private sector solutions is relatively strong. The winners won’t move fast, but once landed, retention and durability will be unmatched. Generally, we all agree on these value props. But the question is, will they truly come to fruition in 2026? The counterarguments all come down to well-known problems with distribution government procurement is notoriously slow, and even winners in the space like OpenGov have generally succeeded through M&A rather than fast organic growth. For the purposes of this prediction, we are carving out defense tech as a separate category (Todd / Luke).
The resurgence of AI UI — In 2026, startups will reimagine the AI interface. First, we’ve only scratched the surface of voice applications. Today, these use cases are held back by latency imperfections and crippled iPhone-native calling functionality. Second, we’ll begin to abandon the fantasy that the everything should happen in a single text box. We all know well the “dashboard fatigue” of the SaaS era; AI has the power to dynamically generate the right visualizations, screens, and guardrails to guide humans through multi-step AI work (per our “inception” theme above). UI will never go back to menus / button overload; but it will rise from the dead stronger than ever. (Omar)
Hallucination proves surmountable — In 2026, novel “hybrid stacks” will erode the hallucination problem holding back enterprise AI adoption. Not by “fixing” LLMs—which hallucinate by nature of their training and lack of neuro-symbolic “understanding”—but by architecting around them. We will see LLMs used for abstraction, workflow reasoning, and presentation / articulation; but paired with deterministic enterprise search and “structured memory” ontologies for reliable retrieval. Moreover, tooling LLMs can natively reference—and the open ecosystem in which they can do so—will grow. Together, these developers will see more startups cracking mission-critical, low-fault-tolerance use cases (e.g. compliance, finance, healthcare), without relying on intensive RPA-style workflow-building. Perhaps a worry for n8n or UI Path. In any case, trust in AI will be a big theme in 2026 (Nic).
Wildcard Predictions
OpenAI loses narrative control — OpenAI struggles to define itself as a sustainable consumer company while also defending an expensive frontier-model roadmap. At the same time, Anthropic continues to win marketshare with developers through reliable APIs and coding supremacy, becoming the default “enterprise model.” Google, meanwhile, advances on two orthogonal fronts: deep integration across G-Suite workflows and continued leadership in fundamental AI research. In 2026, “best model” no longer maps cleanly to “most valuable AI company,” and exuberance splinters (not fades) as the market realizes it may not be so winner-takes-all. (Nic)
Small AI market correction, not a burst — Valuations take medicine, capital gets more selective, but real usage and adoption stay firmly intact. This looks like digestion, not a bubble popping (Omar).
Continued concentration in SF and NYC — The Bay emerges from 2026 with an even greater #1 lead in terms of startup funding and company creation; AI talent density and growing liquidity, paired with SF civic rejuvenation, compounds its advantage. New York City remains the durable #2 and fastest grower by number of investments, with its unmatched ability to attract ambitious young talent. For better or worse, US startup momentum continues to consolidate around these two hubs in 2026. (Nic)
HubSpot acquires Day.ai — A clean narrative reset via a former CPO and a credible AI-native CRM wedge. Less about M&A scale, more about restoring product leadership and internal velocity (Todd).
Federal crypto launch; Bitcoin breaks $150K — Regulatory legitimization arrives via a government-backed crypto or crypto-adjacent rails, unlocking broader institutional participation. Bitcoin responds by reaching new heights (Nic).
Florida sees another migration boom — Property-tax reform math tilts incentives just enough to restart inflows, especially from high-tax states. Tourism and consumption taxes quietly pick up the slack (Luke).
An AI-powered game goes truly viral — And not just a Character AI-like avatar. In the lineage of Infinite Craft but deeper and social. Persistent world / NPCs that remember, creating wholly unique moments people want to clip and share. (Nic)
The Numbers
How many new vertical AI unicorns will be minted in 2026?
Omar: 100
Todd: 101 (intentionally petty)
Luke: 65–70
Nic: 72
Happy New Year!
If there’s any coherent theme to distill from these 40-ish predictions, it’s that 2026 will be a time of sorting. The gap widens between products that demo well and AI that compounds inside real businesses. Model fundamentals move forward, but it’s not all about scale. Distribution, workflow depth, data access, and retention reassert themselves as the limiting factors to AI platform growth, just as they did in every prior software cycle. Some incumbents prove more resilient than expected; others discover that momentum was not a moat. A small number of Vertical AI companies break out by doing the hard, unglamorous work required to earn trust at scale. We’ll revisit these calls next year and see how we did.
Thanks for reading—we look forward to writing & reading with you in 2026!
— Nic & Omar
CUDA is NVIDIA’s proprietary software layer that allows devs to optimize app-layer products to run on NVIDIA hardware. With much of the current AI stack built on CUDA (PyTorch, TensorFlow, CUDA-optimized kernels, cuDNN, cuBLAS, etc.) switching costs are today quite high.
