OneMedNet signs $11.5M+ agreement to supply regulatory-grade real-world imaging data for foundational AI model training
OneMedNet (Nasdaq: ONMD) announced a $11.5M+ agreement to supply regulatory-grade, de-identified DICOM imaging data from its iRWD network (1,400+ sites at April 2025, later 2,130+) for foundation AI model training. The buyer is not named. OneMedNet does the de-identification and curation up front, takes a one-off fee, and does not share in the trained model's downstream value.
Where the rake sits
OneMedNet is paid $11.5M+ for curation, de-identification and regulatory-grade formatting — work it performs and prices. It holds no claim on what the corpus becomes: the buyer's training creates the durable asset and keeps it permanently. Whether $11.5M is the right price for that input is not determinable from the disclosed terms, and that pricing question, not a recoverable share, is what the deal turns on.
What happened
- OneMedNet (Nasdaq: ONMD) announced an agreement worth $11.5M+ to supply regulatory-grade real-world imaging data for foundational AI model training.
- The buyer is not publicly named in the announcement.
- OneMedNet's iRWD platform sources de-identified imaging and clinical records from a federated network of 1,400+ provider sites (April 2025 disclosure), later expanded to 2,130+ sites.
- Disclosed corpus scale at announcement: 121M clinical exams from 31M unique patients; subsequent disclosures cite 90M patient journeys and 270M studies by August 2026.
- The stated buyer purpose is foundational AI model training — a downstream artefact whose commercial value is not shared back to OneMedNet under the announced fee.
- OneMedNet's contribution includes de-identification, cohort-spec matching and provenance-tracing curation on top of raw DICOM archives.
Who is involved
US regulatory-grade imaging real-world-data provider; its iRWD platform curates de-identified medical imaging and clinical records federated from 1,400+ (later 2,130+) healthcare provider sites for life-science and AI buyers.
Nasdaq: ONMD; iRWD network of 121M clinical exams from 31M unique patients across 1,400+ sites as of April 2025; expanded to 2,130 sites and 90M patient journeys / 270M studies by Aug 2026.
The reading
The deliverable is OneMedNet's existing iRWD corpus - millions of de-identified images with associated clinical reports - handed over for the buyer to train a foundational AI model; the model-building work sits on the buyer's side, while OneMedNet's work is the curation, de-identification and delivery of the substrate it already assembles for pharma/CRO/AI customers.
The payer's decision is which weights to ship in a foundational medical-imaging model; the substrate feeds it because iRWD offers multimodal, de-identified imaging linked to clinical reports at a scale (90M+ patient journeys, 270M studies across 2,300+ sites per the 31 Aug 2026 release) that is hard to assemble elsewhere at regulatory-grade quality.
De-identified images and clinical reports cross to the counterparty under the $11.5M+ agreement, with delivery starting immediately; what does not cross is the iRWD sourcing network itself - the 2,300+ provider relationships, the ingestion/de-identification pipeline and the platform through which data is delivered.
Not determinable from the release: the counterparty, licence scope, exclusivity and any downstream-use royalties are undisclosed, so whether $11.5M+ is fair against the model the buyer will build cannot be assessed from public facts; the deal does fit the pattern where a buyer-enriched one-off fee leaves foundation-model upside with the buyer.
Why it matters
A regulatory-grade, cohort-matched imaging corpus is exactly the substrate a foundation-model developer cannot assemble by scraping — which is why OneMedNet gets paid — but a one-off $11.5M fee against a training run whose output may power years of downstream product is the classic shape of under-capture. Is this really under-capture, or as a defensible bet given OneMedNet's stage and the buyer's willingness to sign at that number?
The argument this deal tests: The identifier on loan
Related deals
- Google wins $10M bankruptcy auction for Spirit Airlines' enterprise data and code to train AI models2026-08-14
- Chegg pivots to licensing decade of proprietary STEM academic content and expert network to AI labs2026-05-13
- CuriosityStream Q2 2026 press release discloses $14.1M licensing revenue driven by AI training partnerships2026-08-12
- Tempus expands multi-year Merck collaboration giving Merck access to de-identified multimodal data and Lens platform2026-03-03
Sources
- globenewswire.com — primary
- nasdaq.com — party background
- sec.gov — party background
Announced 2026-07-28