The next two years of clinical AI: our thesis

Consolidation of the base layer

Foundation model providers will further consolidate. Application-layer companies will increasingly build on a small set of clinically fine-tuned bases. The value at the base layer will accrue to a handful of platforms.

Fragmentation at the application layer

The application layer will fragment by specialty, by care setting, and by workflow. This is where most durable companies will be built.

Emergence of clinical operating companies

Hybrid software-and-service companies — those that own both the AI and the clinical delivery — will define the next generation of category leaders. Pure software companies will struggle to maintain differentiation.

The talent bottleneck

The scarce talent is not ML engineers. It is people who understand clinical operations, regulatory strategy, and modern AI simultaneously. Companies that build this talent bench early will define the next decade.

Beyond the model card

A model card tells you what a system was trained on. It does not tell you how it behaves in the third hour of a Monday clinic, when the note is dictated on top of a crying toddler and the medication list is out of date. The most useful artifact we have seen this year is not a model card but a behavior log — a running record of edge cases the model has encountered, the human decision that followed, and whether the outcome was better or worse than the counterfactual.

Founders who publish some version of this log to their enterprise buyers close deals faster. They also get better feedback, because clinicians see themselves in the artifact and respond in kind.

The retrieval layer is the product

In deployment after deployment, the differentiator turned out not to be the base model but the retrieval strategy over the institution's own data. Which notes are indexed, at what granularity, with what recency weighting, and with what filter for hallucinated citations. Teams that treated retrieval as a first-class product surface — with its own eval and its own PM — pulled ahead of teams that treated it as an implementation detail.

Expect the next round of clinical AI companies to look, from the inside, more like search companies than like model companies.

Why the incumbent playbook keeps failing

The instinct of a large healthcare organization is to procure a platform, run a governance committee, and let the technology diffuse through mandate. That playbook worked for imaging PACS in the 2000s. It has failed, visibly and expensively, for the current wave of AI-native tooling. The reason is structural: modern systems have to be tuned to the institution's own data, workflow, and liability posture, and that tuning is a product-engineering exercise, not a procurement exercise.

Founders who understand this shape their commercial motion around a technical champion inside the health system — usually a CMIO, a service-line chief, or a director of quality — and treat every deployment as a co-development contract with clear evaluation gates.

The corollary is that the sales cycle is longer than any founder wants to admit, and the winners raise capital that lets them survive twelve months of pilot without a single dollar of expansion revenue.

The economics underneath the demo

A demo is a snapshot of the best hour of a system's month. Enterprise economics are a function of the worst hour. The teams pricing themselves confidently in 2026 have done three things: they have measured deflection or throughput impact against a matched baseline, they have quantified the reviewer or clinician time returned per shift, and they have translated both into the currency the health system already uses in its own budget cycle.

Where founders lose margin is in the gap between what the model costs to run at peak and what it delivers on the median case. Inference cost curves are helping, but not fast enough to bail out a business built on a demo-day accuracy number. Underwriting a deployment requires a distribution, not a headline.

Expect procurement to get sharper about this in the next two cycles. The health systems that were embarrassed by 2023-era pilots have hired the people who will ask the harder questions in 2026.