Why control points in the value stack matter
At the beginning of 2026, OpenAI reportedly paid around $100 million to acquire a four-person startup called Torch. That’s roughly $25 million per employee.
On the surface, it sounds absurd – another overheated Silicon Valley acquihire headline. But that framing misses the point entirely. OpenAI wasn’t buying people. And it wasn’t buying a flashy product either.
It was buying a control point.
And that tells us a lot about where category power in tech is heading.
This wasn’t an acquihire. It was a Category bet.
What was actually being acquired?
Torch wasn’t building a chatbot. It wasn’t a consumer-facing health app. It was doing something much more foundational: creating a system that allows AI to view a patient’s entire medical history in one place.
That’s not an application layer.
That’s a data layer.
And in regulated, high-stakes industries like healthcare, data layers are where categories consolidate.
Yes, OpenAI could have built this internally. But not quickly – and not cleanly. Healthcare data is fragmented, deeply contextual, politically complex, and wrapped in layers of compliance and regulation that shape what’s even possible.
Rebuilding that from scratch would take years of technical work and institutional negotiation.
Torch had already done the hard part.
Context beats content – especially in healthcare
This acquisition also lines up neatly with OpenAI’s move into ChatGPT Health. According to OpenAI’s own reporting, one in four users submits a health-related prompt every week. That’s over 40 million people engaging with AI around health daily.
That statistic matters because it proves something critical – the behaviour is already there.
People already trust AI to help interpret symptoms, lab results, and medical information. The next step isn’t better conversation – it’s agency.
Healthcare AI doesn’t become transformative by answering isolated questions. It becomes transformative when it can reason over your entire medical history. That’s context over content. And context requires ownership of the underlying data.
Where category concentration kicks In
This is where category power starts to concentrate.
Healthcare AI is moving from “helpful chatbot” to “trusted agent.” And whoever owns the agent must own the history. That means owning the data substrate beneath it.
Which brings us back to why OpenAI paid what it paid.
With one deal, they bought three things at once:
- Technology – a medical data engine that would have taken years to replicate
- Timing – a head start as healthcare AI shifts from conversational tools to agent-based systems
- Expertise – real-world healthcare experience, scars and all
This isn’t an isolated move either. A pattern Is emerging
Look at OpenAI’s recent acquisitions:
- $1.1bn for Statsig – experimentation and telemetry
- $6.5bn for io – hardware and AI infrastructure
- $100n for Torch – healthcare data infrastructure
Consider all three and the strategy becomes obvious.
OpenAI isn’t just building models anymore. Foundation models are rapidly becoming table stakes.
The real competitive advantage now lies in vertical control – owning the infrastructure, the expertise, and the workflows where AI actually gets used.
In other words: the value stack.
From fragmentation to control
This is classic category concentration behaviour.
Early in a category’s life, everything is fragmented. Lots of startups. Lots of surface-level differentiation. Features everywhere.
But once usage explodes, power shifts downward – to whoever controls the deepest layers of the stack.
That’s why the key questions have changed.
It’s no longer:
It’s now:
- Who owns the agent?
- Who owns the history?
- Who owns the workflow people actually trust?
Healthcare just makes this shift more visible, faster. The stakes are higher. Trust, compliance, and continuity matter more than novelty. Once an AI system is embedded in clinical decision-making, you don’t casually swap it out.
That’s vendor nirvana.
The real deal
This deal was never about a four-person startup.
It was about OpenAI planting a flag and saying: We intend to own the healthcare AI category end to end.
And that’s the broader lesson for every tech company watching this unfold.
Categories are no longer won by features. They’re won by control points – data, agents, and workflows.
If you don’t own one of those, you’re probably a plug-in, not a platform.
Or an acquisition target.
And unless you’re sitting on a genuinely category-defining choke point, you shouldn’t expect anyone to pay $25 million per employee for you.
But if you do you do, you are at the centre of the new reality of category power.