Health-AI companies as buyers

If the labs are the prestige customer and the small one — the argument next door — then the funded health-AI application companies are the volume. They have physicians in the loop by necessity, they ship into clinical settings, and there are dozens of them with fresh capital. This is where a vendor thesis expects to find its first invoice.

The evidence says they build.

#Hippocratic AI is the only unambiguous large-scale buyer, and it built

Hippocratic's safety page states it plainly: "We hired over 7.7K U.S. licensed clinicians to make 775K test calls" (hippocraticai.com/safety). NVIDIA's customer story describes "clinical review by a network of over 6,000 nurses and 300 physicians" as a mandatory stage in a three-step agent validation process (NVIDIA). This is the largest contracted clinician workforce documented anywhere in health AI, by an order of magnitude over anything else in the record.

Three things follow, and only one of them is encouraging.

It proves the demand exists at genuine scale. A company at $3.5B post-Series C (Fierce) decided that thousands of licensed clinicians reviewing agent behaviour was worth the outlay.

It was assembled in-house. No vendor is named on the safety page and none was found elsewhere. Hippocratic did the recruiting, the credentialing and the payment rails itself, and then launched a Nurse Advisory Council for LLM safety on top (MobiHealthNews).

And the capability is the marketing. Hippocratic sells the clinician network as its differentiator — the reason its agents are safe enough to put in front of patients. A capability a company markets as its moat is the last thing it outsources. Hippocratic is simultaneously the proof that this demand is real and the proof that a well-capitalised buyer builds rather than buys when the capability is the product. It reappears in The physician panels as the most credible pivot threat in the sector, for exactly that reason.

#Nine clinical roles across eight companies, and not one reviewer posting

Eight health-AI applicant-tracking boards were readable live on 30 August 2026: Abridge (42 open roles), Commure (72), Ambience (17), Nabla (16), Elation (17), Suki (11), OpenEvidence (8), Corti (5) — 188 roles in total. Among them: nine clinical or clinical-adjacent roles, of which perhaps six touch model evaluation at all.

Not one of the eight is advertising for a clinical reviewer, physician annotator, clinical rater or equivalent hourly evaluation role in the United States. The only contract clinical role anywhere in the set is Ambience's Clinician Support Specialist (Contract) — a twelve-month customer-support job on a 1pm–9pm ET shift, not evaluation work (Ashby).

What they hire instead is senior clinical staff to own the function permanently. Ambience's own advert is the clearest statement anyone has published: "At the heart of that work is our clinical AI team: physicians and other clinicians who build, evaluate, and refine the AI that generates clinical documentation used by health systems across the country. Clinical AI is a new discipline. There is no playbook for it." It is hiring a manager because the in-house team has grown enough to need one. Abridge runs a department called Clinician Science and staffs it with a Clinician Scientist at $240K–$280K, a Senior Physician Executive at $160K–$260K and a Product Lead, Clinical Decision Support at $240K–$290K (Ashby).

Abridge plainly runs a reviewer pipeline underneath that — its engineering blog describes offline evaluation using "structured feedback provided by expert clinical reviewers" and physicians conducting "blinded head-to-head comparisons" (tech.abridge.com) — and declines to say whether those reviewers are employees, contractors or vendor-supplied. Two readings survive: an unnamed supplier, or off-platform recruitment through advisory boards, health-system partners and physician networks that never touches a public ATS. The second reading is the more troubling one for a vendor, because it means these companies are already doing the sourcing work you would sell them, informally, for free.

#Commure sets the price ceiling for routine review

Commure raised at a $7B post-money valuation in May 2026 (GlobeNewswire via Yahoo). Its only clinical hiring across 72 open roles is two Medical Documentation Specialists in Bengaluru, inside a team called MDS Operations (India) (Ashby).

Say it plainly: that is the price ceiling for routine clinical review. A seven-billion-dollar company doing clinical documentation work is doing it at Indian salaries, offshore, as an operations function. Any pricing model that assumes US board-certified rates for tasks that are legible as documentation review is competing against that number and will lose. The pricing survives only where the artefact demonstrably cannot be produced by an offshore documentation specialist — adjudicated disagreement, specialty-matched reasoning, de-novo case authorship. That constraint is what What actually gets sold is about, and it is the same wage-arbitrage logic that governs Clinical medicine.

#OpenEvidence is the strategic anomaly

A company built entirely on clinician trust, valued at $12B after a $250M Series D in January 2026 (CNBC), has eight open jobs and not one of them is clinical (Ashby).

That is not an oversight; it is the architecture. OpenEvidence surfaces peer-reviewed literature with citations rather than generating clinical judgement. Its ground truth is the published literature, which is already adjudicated, already expert-authored and already free. There is no rubric to write because the source is the rubric.

If that pattern wins, a chunk of this market disappears. You cannot sell clinician judgement to a company whose product is designed around not needing it — and the fastest-appreciating asset in health AI right now is the one that took that bet. It is the single most important architectural risk to the thesis, and it deserves more weight than its eight job postings suggest.

#The register

Sources: Sacra/Abridge, Fierce, Ashby boards, CNBC, Fierce/Hippocratic, Nabla, Aidoc, Tempus IR. Ambience's valuation rests on a low-authority aggregator and should not be quoted onward.

#The volume nobody can sell into

There is one more category of clinical evaluation labour, and it is large, expert, and structurally unpurchasable. Ambience cites "a rigorous head-to-head evaluation of Ambience's clinical AI platform across 550+ clinicians and 230+ specialties" at MultiCare (Ambience). Abridge publishes a methodology guide for health systems evaluating ambient AI (Abridge). In each case the clinician time is contributed by the customer's own staff as part of a procurement decision, and no money changes hands for the judgement itself.

Any market-size figure that ignores this free supply overstates the pot — which is the correction that runs through Pharma, payers and providers and How much money is actually in the buyer pool.