Health data licensing programs compared: Segmed, Gradient Health, Dandelion and OneMedNet

Which companies pay healthcare providers for medical data?

Several established companies pay hospitals, health systems and imaging providers for de-identified records, then license them to AI developers, device makers and drug companies. They all pay a share of revenue when the data is used. The table sets out what each one says on its own pages, checked on November 6, 2026. Before you compare offers, see what your data is worth from a few questions about the practice, with no files.

Tempus is listed as a market signal, not as a place to send records: it licenses its own de-identified dataset, and its October 2026 filing is the clearest public price for patient records.

Source: Gradient Health: become a data partner

Segmed: revenue share for imaging providers

Segmed says it "enters into revenue share agreements with healthcare providers in exchange for licensing their data". It de-identifies and structures the data, then makes it available to researchers. It claims SOC 2 Type II attestation, ISO 27001 certification and full HIPAA compliance (Segmed).

Its homepage claims 2,800+ healthcare partner sites, 150,000,000+ imaging studies and data sourced from all 50 US states. It quotes the president of Fox Chase Cancer Center on that center's investment in Segmed (Segmed). In March 2026 Segmed, iMerit and Advocate Health released a free annotated 3D mammography data set (iMerit).

Segmed does not publish its share percentage. Ask for the share, the reporting cadence and whether any exclusivity applies before you sign.

Source: Segmed: for healthcare providers

Gradient Health: how the data partner revenue share works

Gradient pays data partners "a revenue share, every time their data is used by our AI developer customers". It publishes a worked example: a $3M deal for 10M exams, one partner supplying 4M of them (40%), and a 20% "typical revenue share for data partners", so that partner receives $240,000 from the one deal. Future use of the same data earns more (Gradient Health).

The example is Gradient's own illustration, not a guaranteed rate. Its FAQ says the details of each revenue share depend on the type, scale and usability of the data.

Partners often give access to a cloud instance holding EHR, PACS or RIS data. Gradient de-identifies it in its cloud, its ethics board reviews projects, and another check runs before data reaches a customer. Through a November 2025 partnership, imaging providers that use DataFirst can opt in and "generate recurring revenue by contributing de-identified data" (Gradient Health).

Source: Gradient Health: become a data partner

Dandelion Health: a consortium for large health systems

Dandelion runs a national consortium of community-based hospital systems that includes Sharp HealthCare, Texas Health Resources and Sanford Health. Montefiore Einstein, with 10 hospitals and more than 200 outpatient sites, joined in October 2025. The consortium gives life sciences companies and AI developers access to imaging, waveforms, clinical notes and structured records (Montefiore Einstein).

On payment, a 2024 report says the health systems "collect a share of the revenue Dandelion Health receives from leasing of the data pool to clients" (Clinical Research News). Dandelion does not publish the share.

Dandelion works with large health systems. A single clinic or small imaging center is unlikely to join the consortium directly.

Source: Montefiore Einstein: Montefiore Einstein and Dandelion partner (Oct 2025)

OneMedNet: is it legit, and how does it pay providers?

OneMedNet is a Nasdaq-listed company (ONMD) that files reports with the SEC. For health systems it offers a fully managed licensing program with $0 upfront. It says revenue is tied to "data richness and match frequency" and that "a defined share flows back on a recurring cadence with transparent reporting". It claims HITRUST r2 and SOC 2 Type II, HIPAA compliance and full BAA coverage (OneMedNet).

It uses HIPAA Safe Harbor as the baseline and Expert Determination when a use case needs it. Its homepage reports an initial contract worth $11.5M+ (July 2026) to supply millions of de-identified records for training a foundation AI model (OneMedNet). In November 2026 it announced a mid six-figure re-order from a company building radiology foundation models (GlobeNewswire).

These are the company's own statements. It does not publish the provider share or name its partner hospitals, so ask for both in writing.

Source: OneMedNet: for health systems and providers

How much is patient data worth? What the Tempus filing shows

Tempus AI's Form 8-K of October 23, 2026 is the most specific public price for patient records. Tempus will use commercially reasonable efforts to license 500,000 de-identified records of patients with Alzheimer's disease or other specified neurological conditions to Pathos, at 100,000 a year for five years. Pathos pays "$50 million in data license fees of $100 per delivered record", plus $5.0 million for record updates. Separately, Bristol-Myers Squibb will pay $137.5 million in fixed payments for oncology data and services through 2031 (SEC EDGAR).

Read the number with three caveats from the same filing. Tempus and Pathos share executive leadership and equity ownership, so the deal falls under Tempus's related-person policy and a special committee approved it. Tempus may license each record only to BMS, Pathos and certain subcontractors for a period ending 18 months after the later of delivery or October 22, 2027. And Tempus licenses its own dataset; it is not a program that buys records from practices.

Revenue-share programs do not translate into a per-record price. As a rough guide for an owner, Mercor publishes company-level ranges of $80K to $820K for a medical practice or health system with 10+ clinicians or 2,500+ patients, and $80K to $620K for an imaging and diagnostics center of any size, read on its listings page on November 6, 2026.

Source: SEC EDGAR: Tempus AI Form 8-K (Oct 23, 2026)

Can you sell patient data under HIPAA? Safe Harbor vs Expert Determination

A covered entity or business associate may not sell protected health information except with the patient's authorization (45 CFR 164.502). Data that meets the de-identification standard is not individually identifiable health information, so that rule no longer applies to it. HHS names two methods for meeting the standard: Expert Determination and Safe Harbor (HHS).

• Rule text: 45 CFR 164.514(b). HHS guidance: Methods for de-identification of PHI.
• Imaging has identifiers in DICOM headers and in text burned into the pixels. See licensing medical imaging data for the imaging-specific checks.

Source: HHS: Methods for de-identification of PHI

BAAs, patient consent and state laws

A program that receives identifiable records to de-identify them is acting as your business associate. HIPAA lets you disclose protected health information to a business associate only if you obtain "satisfactory assurance" that it will safeguard the information (45 CFR 164.502(e)), which means a written business associate agreement (HHS). OneMedNet, for example, advertises full BAA coverage.

Consent works differently before and after de-identification. Selling identifiable records needs each patient's authorization. Licensing properly de-identified data does not, under HIPAA. Your own notice of privacy practices, patient agreements and payer or EHR contracts may still restrict it, so read them first.

State law adds terms. In California, any contract for the sale or license of de-identified patient information must say that it includes de-identified patient information, that the buyer is prohibited from re-identifying it, and that onward disclosure needs the same restrictions (Civil Code 1798.148). Washington's My Health My Data Act exempts protected health information and data de-identified under the HIPAA standard (RCW 19.373.100). Other states differ, so check with counsel where your patients live. See licensing patient records for the full checklist.

Source: eCFR: 45 CFR 164.502 (sale of PHI, de-identified data, business associates)

Can a small practice sell medical data?

A small practice can license de-identified data, but it has fewer direct routes. Dandelion works with large health systems, and OneMedNet's program is pitched at health systems. Segmed and Gradient take healthcare and imaging providers. Gradient's DataFirst route is meant for "imaging providers of every size".

• What you can do: describe your EHR or PACS, years of history, patient volume and specialties; ask a program how it de-identifies and under which method; sign a BAA before any identifiable record moves; compare share, reporting and exclusivity.
• What you cannot do: sell identifiable records without each patient's authorization; send identifiable records to a program before a business associate agreement is signed; promise data your EHR or payer contracts reserve to someone else.
• Imaging and diagnostics centers of any size appear in published buyer listings. Mercor's listing for medical practices and health systems asks for 10+ clinicians or 2,500+ patients.
• If you are selling the practice, the records may matter to the price. See medical and dental practice valuation.

Source: Gradient Health: DataFirst partnership (Nov 2025)

How to compare health data offers, and next step

Ask every program the same five questions: what share you receive and of what; how often you are paid and what reporting you get; whether the license is exclusive and for how long; which de-identification method is used and who certifies it; and whether you can stop new uses and have data deleted. The program that names the most numbers in writing is usually the easiest to compare.

Start with what you hold, not an export: systems, years, patient and study counts, specialties. See what your data is worth before you speak to any program.

Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes (how Origin is paid). See what your data is worth.