Miju LabsThe largest, best-capitalised health-data businesses in existence sit one layer away from this thesis and show no sign of crossing into it.
Truveta — a consortium of major US health systems pooling de-identified EHR — raised $320M in January 2025 at a valuation above $1B from Regeneron, Illumina, Northwell Health, Trinity Health and Advocate Aurora, funding the Truveta Genome Project (GeekWire).
Datavant is the scale player. Its 2021 merger with Ciox Health valued the combination at approximately $7B (Healthcare Dive); by 2025 it was on track for over $1B in revenue, connecting 70,000+ hospitals and clinics and 300+ real-world data partners, and it acquired the RWE platform Aetion — 220 staff, 40+ biopharma customers — in a deal announced May 2025 at an undisclosed price (Fierce).
TriNetX runs a federated EHR research network; Carlyle took a majority stake in 2020 at an undisclosed price (Carlyle). Komodo Health raised a $220M Series E in 2021 at a $3.3B valuation (MobiHealthNews) and took a $200M equity infusion in 2024 alongside a restructuring and a 9% headcount cut (Fierce). HealthVerity agreed to acquire Symphony Health from ICON plc on 5 May 2026, completing that month, price undisclosed (HealthVerity).
Verana Health could not be established at all. Funding, valuation and revenue are absent from every source worth standing behind — Tracxn and PitchBook stubs exist, but no primary figures opened. It is a specialty-society registry business in ophthalmology, neurology and urology, and that is the extent of what can be said honestly.
None of them is entering Layer 3
This is the finding, and it held across every company checked.
Truveta built its own language model on its own corpus. It trained the Truveta Language Model on its EHR data and launched an AI research tool in 2026 (GeekWire). That is a move into AI — and pointedly not a move into expert labelling or third-party evaluation. Truveta builds models on its own data for its own customers. It does not sell clinical judgement, and it does not evaluate anyone else's model.
Datavant bought analytics. Aetion is a real-world evidence platform, not an expert network. The acquisition deepens the records-to-evidence pipeline for pharma; it adds no clinician-judgement capability and was not framed as doing so.
The rest are quiet in the same direction. No RWD broker was found building an expert-labelling or third-party-evaluation product.
Different unit of value: coverage and linkage versus a reasoned determination. Different moat: health-system contracts and de-identification machinery versus task design and adjudication. Different buyer: pharma, payers, providers and CROs versus AI labs and model builders. A company that has spent a decade signing data-use agreements with hospital systems has built precisely nothing that helps it write 48,562 rubric criteria — the distinction drawn in What actually gets sold.
The corollary matters as much as the finding. If records businesses are not coming into judgement, a judgement business is not competing with a $1B-revenue incumbent. It also means the records layer is not a route to the judgement customer: Datavant's 70,000 hospitals are worth nothing as a channel to OpenAI. The two markets touch geographically and nowhere else.
The register
| Company | Raised / event | Valuation | Revenue | Moving into judgement? |
|---|---|---|---|---|
| Datavant | Ciox merger 2021; acquired Aetion 2025 | ~$7B at merger | >$1B (2025) | No — bought analytics |
| Truveta | $320M (Jan 2025) | >$1B | Could not establish | No — built its own LLM on its own corpus |
| Komodo Health | $220M E (2021) + $200M (2024) | $3.3B (2021) | ~$200M ARR [UNVERIFIED] | No |
| TriNetX | Carlyle majority, 2020 | Not disclosed | Could not establish | No |
| Aetion | Acquired by Datavant, price undisclosed | n/a | Not disclosed | n/a |
| HealthVerity | Acquired Symphony Health (May 2026) | Not disclosed | Could not establish | No |
| Verana Health | Could not establish | Could not establish | Could not establish | Unknown |
Adjacent and worth distinguishing: Segmed ($10.4M Series A, Sep 2024) and Gradient Health (~$2.75M seed, 20M+ de-identified patient journeys) sell imaging corpora, not annotation (Segmed, Gradient). Rhino Federated Computing ($15M Series A, May 2025, AlleyCorp) sells the ability to compute on data that cannot move — infrastructure, and a plausible partner rather than a competitor (Rhino). Nightingale Open Science sells nothing at all: outcome-labelled datasets, Globus-gated, no download, non-commercial only (Chicago Booth, Nature Medicine). Its founding argument — that medicine's bottleneck is outcome labels, not images — is the strongest intellectual case in the literature for the judgement thesis, and it comes from a non-profit with no commercial interest in making it.
The one price anchor
Records pricing is enterprise-opaque everywhere. The best public anchor: raw national claims feeds from the largest healthcare data vendors "routinely cost six- to seven-figures annually" (Alpha Sophia).
That range is worth holding next to the judgement side, where the only comparable anchors are Mercor's ~$2,500 per benchmark task and BioStack's unspecified "six-figure contract" (What actually gets sold, BioStack and Sepal). A records subscription and a benchmark build can land at the same annual number for a buyer — from completely different cost structures. Records carry near-zero marginal cost per additional customer; judgement carries a physician's hourly rate every time. That asymmetry is why the records businesses are worth billions and the judgement businesses are worth guessing at, and it is the margin problem GMV is not revenue exists to name.
Truveta's revenue. TriNetX's revenue. HealthVerity's revenue. Verana Health's everything. The prices of Aetion, Symphony Health and the TriNetX majority stake. Only one transaction in or near this space has a public price — Datavant/Ciox at ~$7B — and it is a records business, which leaves no public comparable for what a clinical judgement business is worth on exit.