Health & Longevity AI
Medicine is where the deployment gap is measured in lives. Regulators are clearing AI faster than health systems can absorb it, and the Gulf is building a longevity economy on top of both. We work on what it takes to move AI from clearance to care.
The sovereign stake
A health system is the largest, most sensitive AI deployment surface a state runs, and demographics leave no choice about running it. The Gulf has understood this earliest and is moving hardest: Saudi life expectancy has risen from 74 in 2016 to 78.8 in 2024 under Vision 2030's health transformation, with 80 the stated target, and Dubai is standing up a dedicated regulator for longevity and advanced healthcare. Design the institutions well and health data becomes the one training asset a state uniquely holds, and clinical regulation starts pulling in trials, talent and capital. Design them badly and a ministry ends up answerable for clinical decisions it cannot reconstruct, made by models it cannot inspect.
The board-level stake
Adoption has crossed the line from experiment to operations: roughly three-quarters of US health systems now use or are deploying an AI platform (Fierce Healthcare, 2026), and clinical documentation is the fastest-moving category. The trap is that the technology is rarely the barrier. Deployments stall on workflow fit, legacy systems and data quality, and a cleared device that clinicians route around is a cost, not a capability. Liability compounds it: in Europe a clinical AI now answers to medical-device law and the AI Act at once, and everywhere the question after an adverse event is the same. What did the model do, and who signed off? Boards that fund integration, monitoring and evidence, not pilots, will own this market.
Clearance is sprinting and deployment is crawling. Regulators had authorised over 1,350 AI-enabled medical devices in the United States by early 2026, more than double the count of 2022, and market forecasts run from tens of billions today towards half a trillion dollars by the early 2030s. Yet the field's own literature keeps returning the same verdict: the model is rarely what fails. Deployments die on workflow fit, legacy infrastructure and data quality, the unglamorous system around the algorithm. This is the founding example of our thesis. When we say the hardest problems in AI are no longer in the lab, the hospital is what we mean.
Clinical AI is Complex AI in its purest form. It is multi-objective by construction: cost against care, throughput against safety, the patient in front of the clinician against the population the budget serves. It is high-stakes by definition. And it carries the strictest auditability duty in the field, because a clinical decision is one a named human must answer for, to a regulator, to a family, and sometimes to a coroner.
Continuously learning models break the old assumption that certification is an event. The US FDA's predetermined change control plans are the first serious attempt to license change itself; the discipline that must follow is re-validation as a lifecycle: monitoring in production, drift thresholds that trigger review, and a clear answer to who re-certifies, how often, against what.
Longevity is where health AI meets sovereign ambition, and the Gulf is converting it from wellness into infrastructure. Dubai is creating a dedicated authority to regulate longevity and advanced healthcare; the UAE's longevity market is projected at roughly $32 billion by 2026 on a widely cited market estimate; Riyadh's Hevolution Foundation has become one of the world's largest funders of ageing research; and AI-first drug discovery has planted labs in Abu Dhabi.
The discipline this needs is the same one we apply to sovereign compute: separate capability from theatre. Geroscience is real science with long horizons and unforgiving endpoints. A sovereign longevity strategy earns its name when it can evaluate claims, run trials and kill what does not replicate, not when it announces a clinic.
Our advice is consistent across both halves of this frontier. Sequence health AI like a deployment, not a moonshot. Data governance and consent come first, because health data is the one input a nation uniquely holds and the fastest thing to lose public licence over. Workflow comes before models; monitoring comes before autonomy; and every system that touches a care pathway needs evidence that survives an adverse-event inquiry. In medicine the demo was never the product. The discharge summary is.
- How should adaptive clinical models be certified, re-validated and, when necessary, withdrawn once they are learning in production?
- When AI contributes to a clinical decision, how is accountability apportioned between clinician, hospital, vendor and model, and what evidence must exist for each?
- What does a national health-data regime look like that can train sovereign models without breaking consent, residency or public trust?
- Which clinical workflows should earn autonomy first, and against what threshold of evidence?
- How does a sovereign longevity strategy separate geroscience from wellness theatre before capital is committed?
- What must a health ministry be able to inspect before a foundation model is allowed anywhere near a care pathway?
Fellows for this frontier are being appointed.
People who have built, governed or operated real systems in this domain. The first five fellows are named, and the rest of the founding cohort follows in September.
Have you done this at scale in Health & Longevity AI? We want to hear from you.
Put yourself forward