The latest funding data makes the shift toward specialist AI harder to dismiss. Euclid Ventures reviewed 4,395 U.S. and Canadian software and AI financings from 2025 and found that vertical companies accounted for 53% of deals. They received only 30% of total capital because foundation-model and infrastructure companies absorbed several enormous rounds; remove 12 deals above $1 billion and vertical startups took 51% of the remaining money. Investors are increasingly willing to back software that starts with one job rather than every job.
Harvey Made Legal AI a $15.5 Billion Business
Harvey raised $550 million in September 2026 at a $15.5 billion valuation, just six months after a $200 million round valued it at $11 billion. The company says 80% of the Am Law 100 uses its platform, while legal teams have built more than 25,000 custom agents and workflows for contract review, due diligence, and other recurring work. Those figures give investors something a generic assistant struggles to show: repeated use inside an expensive professional process. Harvey does not have to solve every office task to justify its valuation.
PhysicsX Chose Engineers Over Office Workers
London-based PhysicsX took a different specialist route when it raised $300 million in June at a valuation of roughly $2.4 billion. Its software targets industrial engineering problems across aerospace, energy, semiconductors, automotive manufacturing, and materials, where simulation can consume large amounts of engineering time and compute. The product is built around physical systems rather than email, presentation decks, or general search. Narrow can scale.
Betting Has the Same Workflow Problem
Online betting also contains jobs that look simple to a consumer and highly specific behind the screen. An adult opening Melbet (Arabic: ميل بيت) may see a football price move after a red card, an injury update, or a 1-0 goal, while other systems handle account checks, payments, and eventual market settlement. The bookmaker margin remains part of the offered odds, and a fixed bankroll still determines how much a customer chooses to stake. For an AI supplier, mastering one of those operational problems can be more useful than offering a broad model with no knowledge of how a sportsbook actually runs.
Abridge Went Deep Into the Clinical Conversation
Healthcare provides another case where domain depth has produced adoption. Abridge says more than 300 U.S. health systems now use its clinician-intelligence platform, which works inside electronic health-record workflows rather than sitting beside them as a standalone chatbot. In September, the company was selected under the U.S. Department of Veterans Affairs’ five-year ambient-AI enterprise contract, a multi-vendor agreement with a total ceiling of $775.72 million. The software began with clinical documentation and is now extending into decision support and other work tied to the patient record.
UK Capital Is Moving Beyond the Chatbot
British funding data points in the same direction, although it measures AI broadly rather than vertical AI alone. The British Business Bank found that AI companies captured 44% of all equity investment into smaller UK businesses in 2025, the highest share on record. In March 2026, the Bank committed up to £50 million to SuperSeed Fund III, which targets physical AI in manufacturing, energy, construction and autonomous systems. HSBC and Dealroom later reported $6.1 billion of UK startup and scaleup funding in Q3, with health leading deal activity at 62 rounds.
Specialization Does Not Remove the Investment Risk
A narrow market can still be too narrow, and a strong foundation model can erase part of a startup’s product advantage in a single release. Implementation costs also rise quickly in regulated fields: Harvey has to fit legal workflows, Abridge has to integrate with health systems, and PhysicsX has to prove its software against engineering work where errors are expensive. Proof still matters. The valuations investors are awarding in 2026 are going to specialists that can show both domain knowledge and enough room to expand after the first workflow is won.
Harvey moved from an $11 billion valuation in March to $15.5 billion in September; PhysicsX reached roughly $2.4 billion while staying focused on industrial engineering; Abridge has spread across more than 300 health systems. Those companies are specialized, but none is confined to one tiny feature. That distinction is becoming central to the vertical-AI investment case: start with a problem difficult enough to defend, then expand from a position customers already pay for.