AI Clinical Decision Support

AI clinical agents are replacing human administrative and care tiers

The structural shift from human-staffed administrative and care workflows to AI agents is accelerating across the full care continuum. Hippocratic AI — having completed over 115 million clinical patient interactions at a $3.5B valuation — demonstrates that AI agents can operate at population scale for chronic care management, medication checks, and post-discharge follow-ups. Nolla Health's vertically integrated AI doctor with clinician-in-the-loop oversight, and Adaptive Innovations as healthcare's first AI-native home health provider, signal that agentic systems are now colonising clinical delivery itself, not just administrative edges. Amperos Health's AI agents for end-to-end denial and claims management illustrate how revenue cycle — a historically labour-intensive tier — is being fully automated.

Abridge — serving Mayo Clinic, Duke Health, and Johns Hopkins Medicine across 50+ specialties and 28+ languages — has built what investors now describe as a 10-year regulatory moat in ambient clinical documentation. Commure's unified platform processes 40M+ ambient appointments and $25B in annual claims, demonstrating how ambient capture becomes the beachhead for owning the entire clinical and revenue cycle stack. The Microsoft-Nuance acquisition (>$20B) validated the infrastructure value of this layer years ago, and Epic's 325-million-patient EHR network remains the integration target every ambient AI company must plug into.

Why it matters · Ambient documentation is not a feature — it is the data-capture layer that unlocks downstream agentic automation, making early EHR integrations and regulatory clearances near-impossible competitive moats.

The 90-day stage mix reveals a barbell structure: seed (18 deals, $3.2B) and Series C (18 deals, $8.5B) are statistically tied on deal count but separated by capital intensity, while Series B (7 deals, $807M) is notably thin. This suggests the market is simultaneously funding a wave of new AI-native health entrants — Coral ($12.5M seed led by Lightspeed and Z47), Lucis ($20M Series A led by Singular with General Catalyst) — while also making large conviction bets on proven platforms. The near-absence of Series B rounds implies either rapid promotion to C or a culling of mid-stage companies that failed to differentiate.

Why it matters · For operators, the thin Series B layer signals a dangerous 'missing middle' — companies that survive seed must demonstrate enterprise traction fast enough to justify a direct leap to growth-stage capital.

Nvidia leads all investors in this theme with 28 deal participations, co-investing at Series C alongside Sequoia, Lightspeed, JPMorgan, and B Capital in a $2.5B round at a $27.5B valuation, and alongside General Catalyst and Vista Equity in an $800M Series C at an $8.3B valuation. Nvidia's NemoTron model and full-stack ambitions — now extending into healthcare AI infrastructure — mean it is not merely a passive LP but an active platform builder shaping which health AI companies get access to compute and distribution.

Why it matters · Healthcare AI companies that secure Nvidia as a strategic co-investor gain preferential compute access, chip roadmap visibility, and enterprise channel leverage — a compounding structural advantage over pure software peers.

Triomics' OncoLLM platform — integrating directly into Epic for cancer centers — and Iterative Health's AI-driven clinical trial infrastructure across 100+ gastroenterology and hepatology research sites exemplify a durable pattern: specialty AI commands premium valuation because it reduces clinical trial failure rates and unlocks revenue from hard-to-reach patient populations. Flatiron Health's Roche acquisition and Tempus AI's public market presence set the exit comps that keep institutional capital flowing into oncology data and AI platforms.

Why it matters · Specialty AI platforms that own proprietary clinical datasets and trial-network relationships are building defensibility that horizontal EHR vendors and general-purpose LLMs cannot easily replicate.