Joshua Miller and Ouwen Huang are applying their FarmShots computer-vision playbook to a harder bottleneck: usable healthcare data.
By RuntimeWire Staff · Published
Primary source: U.S. Securities and Exchange Commission
Why it matters #
Medical AI developers need diverse clinical data before they can train or validate useful models. Gradient's single-investor round backs a founder team expanding from imaging into the broader patient record.
Joshua Miller and Ouwen Huang's Gradient Health raised $3 million from a single investor, according to an SEC Form D record dated August 25, 2026. SEC acceptance activity was recorded on August 24.
The Durham, North Carolina-based medical data provider reported a fully sold $3 million equity offering, with no capital remaining. Gradient listed August 13 as the date of first sale and reported one investor. The filing classifies the offering as equity but does not disclose more specific security terms, the investor's identity or Gradient's valuation.
That single $3 million check backs an eight-year-old company in the expensive middle layer of medical AI: obtaining healthcare data from providers, removing identifying information, organizing it into usable cohorts and delivering it to developers building or validating models.
Miller and Huang arrived at that problem through agriculture. The pair previously worked together at FarmShots, a computer-vision business that analyzed satellite imagery to identify crop disease and pest damage. Syngenta acquired FarmShots in 2018, giving the founders an exit before they applied a similar image-analysis background to medicine.
The shift preserved the founders' core insight: machine-learning systems need large, well-organized image libraries before they can reliably recognize a tumor or a damaged crop. Healthcare added privacy controls, fragmented hospital systems and the need to represent different patient populations, scanners and clinical settings.
From imaging library to data infrastructure
Gradient sells access to Atlas, a platform where medical AI teams can search de-identified datasets by disease, modality, demographics and other metadata. Gradient says Atlas covers CT, MRI, X-ray, ultrasound, mammography, ophthalmology and other imaging types, with machine-learning-ready exports available in as little as 48 hours.
Those are Gradient's service claims rather than independently measured delivery guarantees. The product nonetheless targets a concrete problem for medical AI developers: public datasets can be too small, narrow or poorly documented for commercial model training and validation, while negotiating individual agreements with health systems can take months.
Gradient's role is to aggregate supply from hospitals and imaging providers, then sell usable datasets to AI developers, medical-device businesses and life-sciences groups. Gradient also offers data partners a share of revenue when their de-identified records are used, turning dormant clinical archives into a recurring commercial asset for providers.
The financing follows a product expansion announced in June 2026. Gradient said Atlas was moving beyond radiology images and reports to support electronic health records, pathology, laboratory results and ECG data. Miller's argument was straightforward: a scan rarely contains the full clinical story, and models need the context physicians use when making decisions.
Gradient said in June that Atlas offered access to over 20 million imaging studies. That figure is self-reported. Gradient also said the additional data types would be introduced progressively, making the announcement a product roadmap as well as an expansion of the available platform.
The wider scope matters commercially. Imaging datasets can support narrow diagnostic models, while linked records, laboratory results and pathology can serve foundation-model teams trying to understand patient histories and disease progression. Multimodal data also gives Gradient more ways to define highly specific cohorts, which can make each dataset more useful and potentially more valuable.
A concentrated new round
Gradient previously announced a $2.5 million seed round led by VentureSouth, with participation from Wavemaker 360, Bessemer's Scout Fund, Charlotte Angel Network and Twenty5Twenty. A June 2023 Form D reported $2,418,697 sold from a proposed $2.75 million offering.
Across the 2023 and 2026 SEC notices, Gradient reported at least $5,418,697 in securities sold. That total excludes any financing that was not captured in those filings. Gradient has also acquired DataAppraisal, a healthcare data business focused on sourcing anonymized records from health systems, though Gradient did not publish financial terms for that 2024 transaction.
The new round looks different from a conventional syndicate. Gradient reported one investor and set the minimum outside investment at the full $3 million offering size. The filing therefore documents a concentrated bet rather than a collection of smaller checks. The buyer's identity will determine whether the capital carries an additional strategic relationship with a health system, technology provider or existing financial backer, but the filing alone does not establish one.
The timing aligns with Miller and Huang's push to make Gradient a broader healthcare data supplier. Gradient expanded Atlas beyond imaging into multimodal records in June 2026. The $3 million sale gives Gradient additional capital as that product scope grows, although the filing does not allocate the proceeds to specific hires, contracts or development work.
Medical AI's data layer gets crowded
Gradient is competing in a category where data access is being bundled with model development, validation and deployment tooling. Segmed raised $10.4 million in 2024 to expand its medical imaging data network. HOPPR raised $31.5 million in 2025 for infrastructure that combines curated data with foundation-model development and validation. Flywheel sells imaging data management and research workflows to health systems, life-sciences groups and medical-device developers.
Gradient's bet is that developers will value a searchable supply network and fast cohort delivery without requiring Gradient to own the entire model-development workflow. Expanding Atlas beyond imaging raises the size of that opportunity and the operational burden. Each additional data type brings different formats, de-identification requirements and relationships with healthcare providers.
Miller and Huang have spent their careers turning difficult image collections into machine-readable products. The latest financing gives them another $3 million to prove that the approach can become durable infrastructure for medical AI, where access to the right patient data often decides which models can be built at all.