Top 5 Healthcare AI Moves in North Carolina Right Now

North Carolina's healthcare AI story is not a chatbot story. It's a story about who gets to test the tools, who gets to see the data, and how much evidence exists before a model touches a patient. The same week in September made that point five times over, and all five moves are verifiable.

This list ranks verified plays by dollar size or clinical specificity, freshness of the primary announcement, and how clearly the sources say what is live versus still in build or evaluation. University awards and health-system deployments both qualify. We swapped a thinner Novant patient-page item for UNC Health's Evidently enterprise deployment, which has stronger live proof. No invented performance metrics. The rankings are editorial judgments from this evidence pool, not a paid list.

1. UNC and Emory rare-disease AI data resource (ARPA-H RAPID) — up to ~$35 million

UNC and Emory are co-leading a multi-year build under ARPA-H's RAPID program. Prime is UNC. ARPA-H lists up to $34.9 million; UNC says "up to $35 million." Principal investigator is Melissa Haendel; Emory co-lead is Richard Moffitt. The aim is a large-scale dataset spanning roughly 2,700 rare diseases to train models for diagnostic settings that lack rich local data. Identifiers are stripped and access is tiered. "World's largest" is marketing, so we attribute it and don't assert it.

This is a build phase. It is not a live clinical product.

2. Duke DCRI SYMPHONY (ARPA-H ADVOCATE) — up to $15.5 million

Duke leads the SYMPHONY consortium to evaluate agentic AI for heart-failure care. Principal investigator is Manesh Patel. Partners include the American Heart Association, MedStar's human-factors center, multiple health systems, AWS, the Health AI Partnership, and NCDHHS. Patel said Sept. 10 that AI is "already reshaping medicine with great promise but little proof." The work is evaluation and implementation, not an FDA-authorized product today, and Duke's check is not the program's total envelope of $62.7 million over four years. We covered the consortium in a separate report.

3. Kitware and UNC stroke robotics testbed (ARPA-H AIR / TAIR) — up to $17.5 million

Kitware is building an open-source virtual testing platform for robot-assisted endovascular stroke algorithms, with UNC as the clinical partner. Kitware's PI is Rachel Clipp; UNC clinical leads are Yueh Lee and James Ho, working out of the Carrboro Kitware office. The goal is simulation and access research toward future catheter navigation. Lee's "imagine a robot..." language is future tense on purpose. There is no treating robot in North Carolina hospitals today, and the award is for the platform that might one day inform one.

4. UNC Lineberger metastatic breast cancer prognostic model

Emily Ray, MD, MPH, and team describe, in JCO Oncology Practice and the UNC Lineberger newsroom on Sept. 11, a regression model estimating 30- and 90-day mortality risk from routine EHR signals: labs, vitals, subtype, medications. The stated purpose is to prompt hard conversations and palliative involvement, not autonomous treatment. Implementation studies are planned, and the model is not described as live EHR decision support. There is no dollar figure, and we did not independently lock a DOI or AUROC this pass, so we're not inventing metrics. This one is on the list for clinical specificity, not size.

5. UNC Health × Evidently (enterprise clinical data intelligence)

In May 2026, UNC Health selected Evidently for Triangle hospitals and clinics after a 12-week pilot. Evidently is a clinical data intelligence platform, and this is the strongest live-enterprise evidence in the current pool. That's why it outranks the thinner Novant package: Novant Health's Aubrey patient page returned a 404 as of Sept. 12, and PatientGPT still says it "will introduce" its offering. Real announcements, thinner live proof right now.

Honorable mention: Novant Aubrey, which ran a June peri-operative pilot with Qventus, and PatientGPT with K Health. Both are real and both are earlier in the evidence cycle.

What this means for the state

The same-week Triangle cluster shows North Carolina's healthcare AI advantage: evidence infrastructure, access engineering, and data. The discipline of saying what is live versus what is being built is what keeps that authority real.

What to watch

• The RAPID build timeline. When the dataset opens for researcher access, and what tiered access actually gates.
• SYMPHONY's first evaluation. Shadow mode versus IDE, and which health systems enroll first.
• The Lineberger implementation studies. Whether the model crosses into live EHR decision support, and with what guardrails.
• Evidently's rollout. Triangle hospitals first, then what? The 12-week pilot becoming a system standard is the live proof to track.
• The Novant reset. Aubrey's page coming back and PatientGPT actually introducing would change this list next quarter.

Sources: ARPA-H award pages (RAPID; ADVOCATE/SYMPHONY 4704; AIR/TAIR) · UNC.edu and UNC Genetics · DCRI (Sept. 10, 2026) · UNC Health newsroom · Kitware · NC AI News Investigator packs (Sept. 12 and Sept. 15, 2026). "Up to" award ceilings are not cash already spent. Live-versus-planned status follows the primary sources cited. Not medical advice.