In 2025, Insilico Medicine's ISM001-055 delivered positive Phase IIa clinical results — the first drug ever designed by artificial intelligence from the ground up to reach this milestone. No AI-discovered compound has yet won full regulatory approval, but the FDA has authorised over 1,250 AI-enabled medical devices, growing 350% in five years. The global AI in healthcare market is projected to reach $50–56 billion in 2026, up from $39 billion the year before. This is not the AI hype cycle. The FDA cleared a new device every 31 hours in March 2026. This is medicine's quiet revolution, arriving one regulatory clearance at a time.
Not investment advice. Data sourced from Blott Healthcare AI 2026 Report, IntuitionLabs FDA AI Medical Device Tracker March 2026, JAMA Network Open Systematic Review November 2025, Grand View Research AI in Healthcare Market Report 2026, The Imaging Wire FDA AI Authorisations December 2025, Innolitics 2025 Year in Review. All figures current as of June 2026.
For decades, "AI in drug discovery" meant software helping human scientists search faster through known chemical libraries — a meaningful but incremental acceleration of a process that still fundamentally relied on human chemists making the final design decisions. In 2025, that changed in a way the industry had been anticipating but had not yet witnessed. Insilico Medicine's ISM001-055 — a drug candidate whose molecular structure was generated end-to-end by the company's AI platform, not merely screened or optimised by it — delivered positive Phase IIa clinical results. This is the first instance of a fully AI-designed compound clearing a genuine human clinical efficacy hurdle, not merely passing a safety check or completing preclinical modelling.
The honest caveat matters as much as the milestone: no AI-discovered compound has yet achieved full regulatory approval anywhere in the world. Phase IIa is an early, encouraging signal — not proof of a marketable medicine. But the gap between "AI helps scientists find drugs faster" and "AI designs a drug that works in human patients" has now been closed at least once, and the pharmaceutical industry's own data shows it is being closed at increasing frequency: approximately 80% of professionals in pharmaceutical and life sciences now use AI in drug discovery, and the technology is credited with reducing typical discovery timelines from five to six years down toward roughly one year for certain compound classes.
The most underappreciated fact about healthcare AI in 2026 is the sheer regulatory throughput now occurring at the FDA — a pace that itself signals the technology has moved decisively past the experimental phase into routine clinical infrastructure. In March 2026 alone, the FDA cleared 24 AI/ML Software as a Medical Device applications — a new authorisation roughly every 31 hours, every single day of the month. April 2026 brought 27 more. Fifteen clearances landed in just nine days in mid-December 2025.
The growth curve that demonstrates regulatory normalisation, not hype
The median time from submission to FDA clearance in 2025 was 142 days, with a quarter of all devices cleared in under 90 days — demonstrating that, for well-prepared submissions through the established 510(k) pathway, the regulatory system has genuinely adapted to evaluate AI-enabled tools at a pace comparable to traditional medical devices. Aidoc's CARE1 platform received FDA clearance in February 2025 as the first foundation-model-powered clinical AI system to clear regulatory review — a meaningful technical milestone, since foundation models (the same broad architecture underlying large language models) represent a fundamentally more flexible and adaptive class of AI than the narrow, single-task algorithms that dominated the first wave of medical device clearances.
The overwhelmingly dominant clinical field for AI deployment — and for good structural reasons: medical images are data-rich, pattern-detection tasks (spotting a lung nodule, classifying a stroke, flagging a fracture) map naturally onto what deep learning does best, and the imaging workflow already runs through digital systems that AI tools can plug into directly. Lung cancer, stroke, and breast cancer detection remain the highest-impact target areas.
ECG signal analysis, arrhythmia detection, and cardiac imaging interpretation represent the next-largest cluster of AI clearances. Cardiac electrophysiology and interventional cardiology also rank among the highest-value specialties for industry-clinician AI device payments — reflecting both clinical importance and the technical sophistication of cardiac AI tools relative to simpler imaging triage software.
Stroke detection platforms like Viz.ai represent some of the most mature, widely-deployed AI tools in any specialty — automated triage that flags time-critical stroke cases for immediate clinician attention. Neurological surgery received the largest total AI device payment amounts of any specialty in industry-clinician financial relationship data from 2017–2023, reflecting both clinical complexity and commercial value.
Pharmaceutical and biotechnology companies represent the single largest end-use revenue category in the broader AI healthcare market — larger than hospital and diagnostic deployments combined in dollar terms, even though FDA device clearances remain concentrated in imaging. This reflects the enormous capital intensity of drug development, where even marginal AI-driven efficiency gains translate into hundreds of millions of dollars in accelerated time-to-market value.
The single largest holder of radiology AI clearances, built substantially through acquisition — Bay Labs, BK Medical, Caption Health, MIM Software, icometrix, and Spectronic Medical have all been absorbed into GE HealthCare's AI portfolio. This is the clearest evidence in the entire healthcare AI landscape that the dominant players are not venture-backed AI-native startups but established medical device incumbents using their distribution networks, regulatory expertise, and acquisition capital to consolidate the most promising point solutions developed elsewhere.
The second-largest holder of radiology AI clearances, following the same consolidation playbook as GE — most prominently through the acquisition of Varian, the oncology and radiation therapy specialist. The pattern across the top five companies in this space (GE, Siemens, Philips, Canon, United Imaging) is consistent: scale incumbents who already own the imaging hardware are best positioned to bundle AI software directly into the device sale, capturing far more value than a standalone AI software vendor selling into hospitals that already have imaging equipment from someone else.
The most significant AI-native pure-play company in the clearance rankings, and the first company in the world to achieve FDA clearance for a foundation-model-powered clinical AI system (CARE1, February 2025). Aidoc's position as the leading independent AI vendor — rather than an imaging hardware incumbent — makes it the clearest test case for whether AI-native companies can build durable, defensible positions in a market the equipment giants are aggressively consolidating through acquisition.
The most significant single proof point in AI-driven drug discovery to date. ISM001-055's positive Phase IIa results represent the clearest evidence yet that AI-generated molecular structures — not merely AI-assisted screening of existing compound libraries — can produce genuine clinical efficacy in human patients. The company's broader platform approach, spanning target identification through molecule generation, positions it as the most closely watched AI-native pharmaceutical company globally, even as the ultimate test — full regulatory approval — remains unmet by any AI-discovered compound anywhere.
The overwhelming majority of FDA-cleared AI devices lack robust clinical trial evidence — and the pace of approval has decisively outrun the pace of rigorous validation. Fewer than 2% of cleared devices are supported by randomised controlled trial evidence, the gold standard for clinical validation. The FDA's reliance on the established 510(k) pathway — designed originally for incremental hardware modifications, not adaptive software that can change its own behaviour over time — means the regulatory framework is still catching up to the technology it is approving. The shift toward Predetermined Change Control Plans, now included in roughly 10% of 2025 clearances, is the FDA's attempt to build lifecycle oversight suited to AI's adaptive nature, but it remains a minority practice rather than the norm.
Industry-clinician financial relationships around AI medical devices are highly concentrated — and that concentration tracks uncomfortably closely with existing healthcare inequality. AI device payments are concentrated among clinicians affiliated with large, teaching hospitals in urban, high-income areas — precisely the institutions least in need of AI's efficiency gains and most able to afford premium-priced AI tools regardless. Single firms account for over 60% of payments in specialties such as cardiac electrophysiology and neurology, raising genuine questions about whether AI in healthcare is closing or widening the gap between well-resourced and under-resourced care settings — a concern explicitly echoed in dedicated equity-focused research programmes like the Pitt-Leidos partnership building AI tools for underserved cancer screening populations.
No AI-designed drug has yet won full regulatory approval anywhere in the world — and the gap between "promising Phase IIa results" and "approved medicine sold to patients" remains the single largest unresolved question in the entire AI drug discovery thesis. ISM001-055's milestone is genuinely significant, but Phase III trials, the larger and more statistically demanding stage of clinical testing, carry materially higher failure rates than Phase II across the pharmaceutical industry as a whole — AI-discovered compounds have not yet demonstrated they are exempt from this pattern. The bull case for AI drug discovery rests on a reasonable but still unproven assumption: that AI-generated molecules will succeed in Phase III at rates comparable to or better than traditionally discovered compounds, a question that will not be definitively answered for several more years.
Long-horizon thinking on capital, technology, and the forces shaping the next decade of wealth creation. Written from first principles. Not consensus. Not noise.