OpenAI has quietly bought itself a new front door into your medical life: Torch, a tiny one‑year‑old health startup that tries to pull all your scattered health data into one “medical memory” and plug it into AI.​

If you’ve ever tried to piece together your own health story—portals at three hospitals, lab PDFs in your email, a Fitbit history somewhere in the cloud—you already understand the problem Torch was built to solve. The app connects to hospitals, labs, wearables, and consumer testing services, then stitches everything into a single, structured view that an AI system can actually reason over. Torch’s founders summed it up bluntly: your data is scattered across “four hospitals, two labs, seven apps, and three web portals,” and AI is useless if it can’t see the full picture.​

That vision is exactly why OpenAI came knocking. Multiple reports peg the deal at around $60 to $100 million in equity—real money for a team that’s reportedly just four people and roughly a year old. Torch was founded in early 2025 and had only just started rolling out its app when it agreed to be acquired, which makes this less a classic scale‑up exit and more a “buy the brain and the blueprint” move. All four co‑founders are heading to OpenAI, and Torch, as a standalone app, is effectively disappearing into the larger machine.​

The timing is not subtle. Just days before the acquisition went public, OpenAI announced ChatGPT Health, a dedicated health experience that lets people link medical records and wellness apps like Apple Health and MyFitnessPal into a protected space inside ChatGPT. Out of the box, that sounds impressive, but without a serious data‑unification layer, it risks becoming just another pretty interface on top of the same old fragmented records. Torch gives OpenAI exactly what it needs here: a backend that was purpose‑built to ingest messy clinical data, lab results, prescriptions, visit summaries, and even recordings or notes from doctor appointments, then clean it up for AI.​

What Torch calls a “medical memory” is basically a context engine for health data. Instead of ChatGPT staring at a single lab report in isolation, it can see years of labs, diagnoses, medications, lifestyle data, and clinician notes as one coherent narrative. That matters when you ask it something simple like “Should I be worried about this cholesterol result?”; the useful answer often depends on trends over time, co‑existing conditions, and what your doctor has already tried. Torch’s tech is designed to surface that context for the model so it doesn’t hallucinate wildly from a single PDF.​

Under the hood, Torch isn’t just sucking in HL7 feeds from hospitals and calling it a day. It also tries to capture the stray but critical pieces that usually vanish: the off‑hand instructions a doctor gives at the end of a visit, the logic behind a treatment plan, or the voice notes patients record as they process bad news. Those fragments rarely live in neat, queryable systems, but they’re often what patients and families cling to when they’re trying to remember what was said and why. By preserving that side of the story alongside the formal record, Torch aims to support continuity of care in a way that typical portals simply don’t.​

On OpenAI’s side, this is part of a broader land‑grab in health. ChatGPT Health is pitched as a way to help people prepare for appointments, understand lab results, and manage lifestyle changes without pretending to replace clinicians. OpenAI has already been courting large health systems—including names like HCA Healthcare—with enterprise products that wrap its models in compliance, audit, and workflow layers. Giving those systems a reliable way to unify data via Torch is a strong pitch: same models, but now grounded in a complete clinical picture instead of whatever the EHR happens to expose in a single view.​

The money, in this context, is almost as interesting as the technology. A sub‑100‑million‑dollar acquisition is small change compared with the multibillion‑dollar dealmaking happening around AI chips and infrastructure, but it sends a clear signal: OpenAI is willing to do focused M&A to own the layers that really matter for verticals like healthcare. Last year, it already bought a hardware‑adjacent startup founded by Jony Ive for over six billion dollars, and it recently brought in a dedicated corporate development lead from Google to keep these kinds of deals flowing. Torch is one of the first obvious “vertical infrastructure” buys in that playbook.​

For the Torch team, joining OpenAI is a scale decision as much as anything. In their public note, the founders basically admit that the mission—making sure “nothing important gets lost in the noise again” when it comes to your health—is bigger than what a tiny startup can realistically do alone. Partnering with one of the most resourced AI companies on the planet means their “medical memory” idea can be shipped not just as a niche consumer app, but as the default health data fabric for millions of ChatGPT users and, eventually, health systems.​

For patients, the upside is intuitive: less paperwork, less login roulette, and a higher chance that the AI you consult about your health actually understands your history. Imagine a single interface where you can ask, in plain language, how your kidney function has trended over the last three years, or whether your current medication list plays nicely with the new drug your specialist just prescribed—and get an answer that is both grounded in your real data and ready to discuss with your doctor. That’s the aspirational version of what OpenAI is trying to build with ChatGPT Health plus Torch.​

The risks, of course, are just as obvious. Healthcare is one of the most regulated, sensitive domains imaginable, and OpenAI already sits in the crosshairs of policymakers worried about data privacy and AI overreach. Unifying medical data at this scale raises tough questions: who exactly gets to see what, how consent is handled, what happens if a model misinterprets a record, and how much responsibility a consumer‑facing chatbot should bear when someone acts on its advice. Regulators and health systems will want detailed answers before they let this anywhere near clinical decision‑making.​

There’s also the competitive backdrop. While OpenAI snaps up Torch, Anthropic is rolling out healthcare tools for Claude, and every major tech giant with a cloud business has some kind of health data strategy. Historically, Big Tech’s forays into health records have been messy—Google Health’s early attempts and Apple’s own slow‑burn Health Records effort are reminders that interoperability can crush even the biggest players. The difference this time is the presence of powerful general‑purpose models: instead of just storing and syncing data, the pitch is that AI can actually help interpret it.​

Torch slots in as a quiet but important piece of that puzzle. If ChatGPT Health is the bedside manner and model intelligence, Torch is the plumbing: the layer that actually makes sure all the right information arrives in the room. It’s not glamorous, but in healthcare, the boring stuff—data formats, interfaces, context—often determines whether a tool is a toy or something clinicians might eventually trust. OpenAI’s bet is that owning that layer, even via a small, early‑stage acquisition, gives it an edge that pure API competitors will struggle to match.​

For now, the official story is simple: Torch is “joining OpenAI,” the standalone app fades into the background, and ChatGPT Health quietly gets a lot smarter about your records. The more interesting question, and the one patients, doctors, and regulators will be watching, is whether this combination can move beyond demos and headlines to actually make care feel less fragmented and more human. If it works, you won’t think about Torch or model architectures at all—you’ll just have one place to ask health questions, and for once, the system will actually remember who you are.​