Dr Kola Tytler (MBBS MBA MRCGP) | 15 July 2026 | 11 min read
OpenEvidence raised $250 million in a Series D round in January 2026, taking its valuation to $12 billion in a round co-led by Thrive Capital and DST Global, roughly a year after its first outside funding at a $1 billion valuation. Neko Health's latest round values it at approximately $7 billion. One company organises and retrieves medical knowledge for practising clinicians. The other generates and interprets new physical health data about individual patients. Neither is a direct competitor to the other, and that is precisely what makes the comparison instructive: investors are financing entirely different control points across medicine simultaneously, and the resulting market is best understood as a stack of layers rather than a single race.
Layer one: medical knowledge
This layer answers the question "what does the evidence say." OpenEvidence, trained specifically on medical journals and clinical data rather than the open internet, has built what its chief executive describes as the most widely used AI platform among US physicians, reporting over 40 percent of US doctors as users and monthly consultation volumes that reached around 20 million by January 2026. iatroX occupies an adjacent position in the UK context, offering source-grounded retrieval from NICE, CKS, SmPC and SIGN guidance for clinical reference and education. This layer's core asset is coverage and trust in the underlying evidence base, not proprietary patient data.
Layer two: the clinical encounter
This layer answers "what happened in this consultation, and how is it documented." Heidi, Tandem and TORTUS occupy this space, converting ambient audio into structured notes, letters and, increasingly, coded diagnoses and suggested next steps. All three have been visibly moving beyond pure transcription towards broader clinical-assistant functionality over the past year, a trend reflected in the medical device reclassification each is now navigating.
Layer three: diagnostic interpretation
This layer answers "what does this scan, signal or lab result mean." AI-assisted radiology, dermatological lesion analysis, cardiovascular signal processing and laboratory interpretation sit here. It is often embedded inside a larger product rather than sold as a standalone tool, which makes it easy to underestimate how much competitive value is concentrated in this layer specifically.
Layer four: data-generating healthcare infrastructure
This layer answers "what new clinical data should be collected, and how." Neko, Prenuvo, Function Health with Ezra, and Q Bio all sit here, embedding AI directly inside clinics, scanners and diagnostic pathways rather than applying it to data that already exists. This is the most capital-intensive layer of the four, requiring hardware, physical premises, clinical staff and regulatory clearance rather than a software deployment, which is a large part of why Neko's $700 million round looks so different in scale from a typical software Series C.
Layer five: longitudinal intervention
This layer answers "is the person's risk actually changing over time." Wearables, repeated biomarker testing, structured preventive programmes and personalised recommendation engines sit here, and it is the layer where the industry currently has the least mature evidence, since it requires years of consistent data collection per person to say anything reliable about whether an intervention worked.
Why the stack framing matters more than a single ranking
Clinical AI is not converging towards one universal application that does everything, and treating Neko, OpenEvidence, Heidi and a diagnostic imaging algorithm as competitors in the same race obscures more than it explains. Different companies are building durable positions in different parts of the pathway, and the more useful platforms may end up being the ones that connect cleanly between layers, from a new measurement, to its clinical interpretation, to the evidence and guidance a clinician needs to act on it, rather than the ones that attempt to own every layer themselves. It is also worth noting that more testing at layer four tends to generate more, not fewer, questions that layer one and layer three exist to answer: a body scan that flags an incidental finding creates immediate demand for exactly the kind of evidence retrieval and clinical reasoning support that knowledge-layer platforms provide.
Where this leaves clinical knowledge platforms
Neko generates the data. Diagnostic interpretation layers flag what might be abnormal within it. Knowledge platforms such as OpenEvidence and iatroX help the clinician who then has to explain that finding work out what the current evidence and guidance actually say about it. Ambient assistants document the resulting consultation. None of these layers is made redundant by the others becoming more capable; if anything, growth at the data-generation end of the stack increases the volume of interpretation and evidence-retrieval work further up it.
Conclusion
The future of medical AI looks less like a single dominant product and more like an interconnected stack, in which the most defensible position is not necessarily owning the most layers, but being genuinely indispensable within one of them, and interoperable with the rest.