OpenAI has acquired Torch, a small health records startup, in a deal that was not publicly priced but is pegged at about $100 million in equity, according to a source cited by The Information. OpenAI confirmed the acquisition, while Torch said its four-person team will move over to OpenAI as part of the transaction—making the deal look less like a traditional buyout and more like a “acqui-hire.”
Torch had been building an application to consolidate a person’s health information in one place so it could be used more effectively with AI. The company’s pitch centered on the reality that medical data is fragmented across dozens of systems: physician notes from office visits, lab results, imaging reports, medication lists, wearable-device metrics, insurer portals, and an expanding universe of consumer wellness and at-home testing services. Torch said it was trying to turn that scattered trail into a unified layer of context—what the team described as “a medical memory for AI, unifying scattered records into a context engine.”
The team’s origins trace back to Forward Health, the now-defunct startup that attempted to reimagine primary care around a membership model and AI-enabled clinical workflows. Torch co-founder Ilya Abyzov wrote on X that the Torch founders met while working at Forward, which was known for its tech-forward clinics and automation ambitions. Forward abruptly shut down in late 2024 despite having raised more than $400 million, a reminder of how difficult it can be to scale healthcare businesses—even those backed by significant venture funding and strong engineering talent.
Torch’s outcome is different. Rather than trying to survive as a standalone healthcare company, the team is now being folded into OpenAI’s new ChatGPT Health initiative, a recently announced service aimed at people using ChatGPT to help interpret information and manage health-related tasks. The fit is straightforward: if ChatGPT is going to be useful for health analysis, it needs better context than a single message thread can provide. Health decisions often hinge on longitudinal patterns—how symptoms evolve, how lab values trend over months, how medications interact, and what a clinician previously ruled out. A system built to aggregate and normalize that history could significantly improve what an AI assistant can do, at least in theory.
At the same time, the idea Torch pursued addresses some of the hardest problems in digital health. Collecting data from many sources is notoriously messy because formats differ, access permissions vary, and patients often have to jump through multiple authentication steps. Even when the data can be obtained, it can arrive incomplete, duplicated, or difficult to interpret without a clinical context. Building a “medical memory” that is both comprehensive and reliable requires not just ingestion, but also cleaning, reconciliation, and careful handling of uncertainty—especially when an AI system might be asked to summarize or reason over what it finds.
There are also obvious privacy sensitivities. Medical data is among the most personal categories of information, and any effort to centralize it raises questions about security, user control, consent, and where the data is processed. OpenAI has not disclosed detailed technical or policy specifics in the text provided here, but the presence of a dedicated team and a clear internal product home in ChatGPT Health suggests the company wants to invest in health-specific infrastructure rather than treating health queries as just another chatbot use case.
For OpenAI, the acquisition signals a pragmatic approach: bring in a tiny, specialized group that has already been thinking deeply about connecting disparate health sources and packaging them into an AI-ready context. For Torch, it offers a path to scale that would have been extremely difficult as a four-person startup navigating healthcare integrations, compliance expectations, and consumer trust.
If ChatGPT Health is meant to evolve beyond answering general wellness questions and into something that helps users organize, interpret, and act on their own data, Torch’s “context engine” concept could become foundational. The big question is how OpenAI will implement it in practice—balancing convenience and intelligence with the safeguards required when the subject matter is people’s real medical lives.
