The Week Big Tech Became Big Pharma

This week, Anthropic launched Claude Science and announced it's starting its own drug discovery programs. Not just tools for drug discovery — actual drug programs, targeting neglected diseases, with Novo Nordisk and the Allen Institute as reference customers.

That completes the set.

Microsoft has Microsoft Discovery, an agentic R&D platform running on Azure,. Google has Gemini-powered AI co-scientist plus Isomorphic Labs — the DeepMind spinout that's pushing AI-designed drug candidates into clinical trials and OpenAI has GPT-Rosalind, a frontier reasoning model built specifically for biology.

Four companies. Four science-specific models, apps, and programs. Four moves into the exact market, BenchSci and every other techbio company have spent years trying to build a moat in. This isn't a trend anymore. It's a land grab, and it happened in about six months.

Why now, and why all at once

I don't think this is a coincidence; there are four real reasons this is happening right now.

One: science is where the token spend actually is. The most expensive inference you can run isn't customer support or code completion — it's reasoning over a protein structure, a mechanism of action, a multi-step experimental plan. Reasoning tokens already cost 10 to 30x more than standard tokens, and science questions are the hardest, longest, most tool-heavy reasoning tasks that exist. If you're an AI lab trying to prove your model is worth what you're charging for it, biology is the best demonstration you can buy. It's also the domain where we understand the least, which means the ceiling for value creation is enormous.

Two: it's the best answer to a real perception problem. AI has an image problem right now — job displacement, misinformation, energy use, and the sense that the technology mostly benefits the people building it. Curing a neglected disease is the cleanest, least arguable counter-narrative available. You cannot easily attack a company for using its model to pursue diseases that nobody else will fund.

Three: there's a Nobel Prize sitting right there. Demis Hassabis won the 2024 Chemistry Nobel for AlphaFold. That's not a marketing win — that's the single highest form of legitimacy in science, handed to an AI company for solving a fifty-year-old biology problem. Every other AI lab watched that happen and drew the obvious conclusion: this is the fastest path from "AI company" to "institution that matters."

Four: the founders are scientists, not just engineers. Hassabis has a background in neuroscience. Dario Amodei has a PhD in biophysics from Princeton. These aren't CEOs who discovered biology as a market opportunity in a strategy offsite. They came from the field, left it, built AI companies, and are now walking back in with vastly more capital and compute than they had the first time. That's not a pivot. That's a return.

The stranger thing

Here's what should actually give you pause, and it's not the four reasons above. It's that Anthropic isn't just selling Claude Science to drugmakers — it's becoming one. That's a direct conflict of interest with every customer buying the product. You're asking pharma companies to trust your platform with their most sensitive R&D data while you're running your own drug pipeline on the same infrastructure. 

I think I know why Anthropic is doing it anyway, and it isn't strategy. Dario's father,, died from a rare illness in 2006. Four years later, a breakthrough turned that same illness from roughly 50% fatal into 95% curable., I don't think you build a drug discovery program that starts with "neglected diseases" — the ones with no commercial upside — unless there's something more personal underneath it than market share. This is something he spoke about openly. 

That doesn't resolve the conflict of interest. It explains it.

The mirror

Here's the part that actually matters if you run a company in this space, and I'm including myself in this. For years, the pitch from every AI-for-drug-discovery startup — BenchSci included — has been some version of "we understand biology and pharma better than a generalist AI ever could." That pitch just got a lot harder to make. These aren't underfunded generalists anymore. They have frontier models, dedicated scientific reasoning capability, direct pharma partnerships, and in Anthropic's case, they're willing to run their own drug programs to prove it works.

If your company's right to exist depends on "the big labs won't bother building this specifically for biology," that thesis died this week. The honest question every techbio and AI-for-pharma company needs to ask isn't "how do we compete with a general-purpose model" — that fight was always winnable. It's "how do we compete with a frontier lab that has decided science is now core to its mission, has Nobel-level credibility, and doesn't need this market to be profitable to keep investing in it." That's a different and much harder problem, and pretending otherwise is how you get run over.

The companies that survive this won't be the ones with the best positioning slide about proprietary data. They'll be the ones who can look at this list — Anthropic, Microsoft, Google, OpenAI, all in, all at once — and say honestly what they have that none of the four can replicate. If the answer takes more than a sentence, it isn't real yet.