Anthropic spent $400M and a Nobel hire on Claude Science. Nobody's locked in.

Here is the part the launch coverage glided past. Anthropic shipped Claude Science on 30 June 2026, wrapped it in a $400 million acquisition and a Nobel laureate, and pointed it at a 10x revolution in life-sciences R&D. Impressive. Now look at the customer list, because that is where the story actually lives — and what it says is that nobody has committed to anyone.

Novo Nordisk is a named partner for Anthropic's science push. It is also a named partner for OpenAI. The Allen Institute has worked with both, too. Read that plainly: the marquee customers everyone cites as proof of momentum are the same customers the rival cites. Big pharma is not picking a winner — it is running parallel evaluations, hedging across every frontier lab, and extracting subsidised access from all of them while committing to none. That is not a knock on Claude Science. It is a warning about what the launch does and doesn't prove. It proves capability. It does not prove lock-in.

The instinct is to score this on assets, and on assets Anthropic looks dominant: it bought Coefficient Bio, it poached John Jumper — the man who led AlphaFold and won a Nobel for it — and it has Amodei's conviction behind the whole programme. But assets are inputs. The output that matters is whether a wet-lab scientist opens Claude Science on Monday and never opens the OpenAI equivalent again. Capability transfers between labs faster than anyone admits; workflow habit does not. The only durable moat in AI-for-science is becoming the environment where research actually happens, so the data exhaust and switching costs compound. Anthropic knows this — it's why it built a workbench, not a chatbot skin. But building the moat and filling it are different jobs, and the filling has barely begun.

On timing, Anthropic is the challenger, not the incumbent. OpenAI shipped GPT-Rosalind, tuned for biological reasoning, back in April 2026 — roughly two months before Claude Science launched. Google DeepMind, meanwhile, is the house Jumper left, with a decade of scientific-computing credibility and AlphaFold's brand. So the real position is this: Anthropic has arguably the best talent and the most coherent strategy, but not the earliest product and not an exclusive customer. That's a genuinely strong opening, and it is only an opening. The lab that wins AI-for-science won't be the one with the best launch post in July — it'll be the one whose workbench a thousand scientists refuse to close by December.

So here is the scorecard that matters, and it is uncomfortable for the bulls. Anthropic has the best hand of talent, the splashiest acquisition and the clearest thesis — and none of that is the same as owning a customer. The thing that would actually build a moat is boring: a scientist opening Claude Science every morning until switching away feels like moving house. That habit takes months to form and is trivially undercut by a rival throwing bigger credits at the same lab. Until Anthropic can point to a pharma partner that stopped evaluating OpenAI and Google in parallel and actually committed, the honest label on this launch is expensive, impressive and unproven. Watch the renewals, not the reveal — that is where AI-for-science will be won or lost.