Most U.S. doctors are quietly using this AI tool. Few patients know about it.

OpenEvidence says roughly 65% of U.S. doctors used its AI tool in April across nearly 27 million clinical encounters, and the company claims about 650,000 U.S. doctors plus 1.2 million international users. The startup has raised $700 million in under a year, reached a $12 billion valuation, and is expanding into notetaking, billing and EHR integration, while some hospitals remain cautious about PHI/privacy and evidence quality. The article is broadly positive on AI adoption in healthcare, though it flags hallucination, dependence and patient-safety concerns.

Analysis

The key market signal is not just AI adoption in medicine, but the migration of clinical workflow from static reference tools to embedded, high-frequency decision software. That shifts the value pool away from generic search and toward vertically integrated, distribution-locked platforms with proprietary content rights and enterprise EHR integration. In practical terms, the winner set is broader than the article suggests: EHR vendors, medical content licensors, cloud/compute providers, and AI infrastructure suppliers benefit from a growing layer of recurring usage that is hard to dislodge once integrated into point-of-care routines.

For DOCS, the most important second-order effect is not a single product launch, but the validation of a freemium physician network with near-zero distribution friction. If OpenEvidence-like tools become the default “first lookup” layer, then network effects compound through habit formation and enterprise rollout, which can compress the sales cycle for adjacent clinical products. The risk, however, is regulatory and reimbursement friction: if health systems tighten PHI rules or require validated clinical evidence for use in care pathways, consumer-style adoption can slow quickly even if individual doctor usage remains high.

NVDA is a quieter beneficiary because the upside is less about direct training workloads and more about inference intensity across an expanding installed base of clinical users. The counterpoint is that many healthcare AI workflows are query-light and content-heavy rather than compute-heavy, so the near-term revenue lift may be modest versus the headline TAM narrative. The more durable revenue driver would come if these tools move from search into ambient documentation, billing, and visit integration, which would materially raise token consumption and enterprise GPU demand over 12-24 months.

The consensus is probably underestimating how quickly incumbent medical information vendors can reprice their product if embedded AI becomes a must-have rather than a nice-to-have. That said, the current enthusiasm likely overstates the certainty of patient-outcome improvement; the real edge is efficiency, not necessarily diagnostic accuracy. The tradeable setup is therefore a “picks and shovels plus platform” story, with the main risk being a policy-driven re-rating if early studies show no measurable clinical benefit or if privacy incidents force enterprise buyers to pause deployments.

Key Decisions for Investors

• Long DOCS on a 6-12 month horizon as a vertical AI workflow winner; pair against a basket of slower-moving healthcare IT incumbents if execution risk emerges, targeting 20-30% upside if physician adoption continues to translate into enterprise monetization.
• Add NVDA on weakness as an indirect beneficiary of healthcare AI inference growth; use a 12-18 month view and consider upside calls rather than outright stock for asymmetric exposure if OpenEvidence-style products expand into documentation and billing.
• Short a basket of legacy medical reference / workflow friction names on rallies if they lack defensible AI distribution, with a 3-6 month catalyst window around enterprise renewal cycles and AI feature launches.
• Watch for a regulatory pullback trade: if any major health system restricts PHI use or a high-profile hallucination event surfaces, buy short-dated downside protection on DOCS and related healthcare AI names because adoption sentiment can de-rate quickly.
• If you want cleaner relative value, long healthcare AI platform exposure versus generic software names that lack licensed content or workflow lock-in; the moat here is domain data, not raw model quality.