Sanofi rents the same AI brain as AstraZeneca — and that's the story
Sanofi becomes the second major pharma in three weeks to license Owkin's K Pro. BMS is now counting the bill on its 30,000-seat Claude rollout. Plus: 'first AI drug in the clinic' unpacked, the AI-funding bifurcation, and the Onion Desk.
Pharma's AI story is maturing from headline to ledger. Sanofi renting Owkin's agents instead of buying a company, Bristol Myers counting the bill on 30,000 Claude seats, and a "first AI drug" that turns out to be AI-formulated rather than AI-discovered all show an industry moving past the land-grab into the harder questions of cost, category, and proof. Meanwhile the week's biggest checks — Pfizer's $10.5B oncology deal, Lilly's $1.9B RNA-editing pact — contained no AI at all. The throughline is a market learning to separate the label from the value.
Lead — Sanofi rents the same AI brain as AstraZeneca
DealsSanofi has become the second top-tier pharma in weeks to license Owkin's "K Pro" agentic-AI platform, signing a multi-year collaboration that includes a five-year license spanning its full pharmaceutical value chain, per AllSci; financial terms were undisclosed and the deal extends a relationship dating to 2021. It lands roughly three weeks after AstraZeneca signed its own three-year K Pro license. Owkin builds the agents and deploys them inside the pharma's own IT environment, so proprietary data never leaves the building. The real story isn't the agents but the procurement model — two of the world's largest drugmakers renting the identical third-party platform within weeks suggests pharma has answered build-vs-buy with a third option, rent, and that Owkin is becoming something like a shared industry standard. Per HLTH, Q1 2026 saw 36 major pharma-AI partnerships, double the 18 a year earlier. The non-obvious risk is convergence: if the same vendor's agents run inside rival firms, the proprietary edge narrows to the data, not the tooling.
Bristol Myers starts counting the AI bill
OpinionHaving handed Anthropic's Claude to more than 30,000 employees, Bristol Myers Squibb's next AI problem is the invoice — per Endpoints News, CEO Chris Boerner says the company is now working to get a handle on the rollout's costs. BMS announced the enterprise-wide deployment on May 20 as a "shared intelligence layer" across research, manufacturing, and commercial, also tapping Claude Code for internal engineering; the cost focus surfaced barely two weeks later. This is the first public crack in the enterprise-AI land grab: consumption-based pricing at 30,000-seat scale turns "deploy everywhere" into a live budgeting problem, and every comparable deal — Merck with Google (up to $1B), Novo Nordisk with OpenAI — will hit the same wall. Watch whether the metric shifts from seats deployed to value per query, the one nobody can yet measure.
The "first AI drug" in the clinic was AI-formulated, not AI-discovered
RegulatoryCRDMO Quotient Sciences says it has dosed the "first AI drug" in a Phase I study at its UK facility following MHRA approval — but per Drug Discovery World, the AI did the formulation, not the discovery. "AI-formulated" means machine learning optimized the drug product — excipients, dosage form, dissolution behavior — rather than the molecule itself, a real and useful application but a narrower one than the "AI-discovered drug" headlines it echoes. As such claims multiply, the gap between AI-discovered, AI-designed, and AI-formulated will separate genuine clinical-validation milestones from marketing, and it's telling that the announcer is a contract developer rather than the drug's owner. Reserve the fanfare for a molecule that owes its existence to AI surviving a pivotal trial.
An investor names biotech's AI haves and have-nots
OpinionSpeaking at Nucleate UK's The Forum, early-stage investor Nirmesh Patel of Amino Collective described a clear bifurcation in biotech fundraising between companies with an AI angle and those without, per BiotechTV. For founders, an "AI angle" is sliding from differentiator to financing prerequisite — which actively incentivizes AI-washing — while for investors it raises the odds of paying an AI premium on companies whose underlying biology isn't de-risked. But in a week dominated by clinical disappointments (ADC Therapeutics, Oculis, Celcuity, Praxis) and a regulatory blowup at Fulcrum, capital fleeing toward the "AI" label looks less like conviction than flight to narrative. The real test isn't the premium AI-angle startups command today — it's whether they keep it after their first clinical miss.
Flourish raises $500M for brain-inspired foundation models
FundingFlourish raised $500M for brain-inspired AI foundation models, per Crunchbase. It's neuro-adjacent rather than pharma, but a half-billion-dollar bet on biology-native architectures is worth tracking for anyone who doubts scaled transformers are the end state.
Pfizer and Innovent sign a $10.5B oncology deal — with no AI
DealsPfizer and Innovent inked a $10.5B oncology deal, per Drug Discovery World — the week's biggest biopharma transaction, with zero AI in it. A useful reminder that the largest checks still chase de-risked assets and China-originated pipelines, not algorithms.
A call for explainable protein language models
PapersA call for explainable protein language models ran in Nature Machine Intelligence (CRG; DOI 10.1038/s42256-026-01232-w; via phys.org), a perspective piece published last month. As pLMs start informing real design decisions, "the model said so" won't survive an FDA credibility assessment — explainability is becoming a regulatory issue, not an academic one.
The denominator: ~460 startups, ~$60B invested
Industry dataA widely cited TechCrunch figure puts the field at roughly 460 AI drug-discovery startups and about $60B invested to date. Keep that denominator handy for every "AI still hasn't produced an approved drug" debate.
The Onion Desk
Following a Pistoia Alliance poll naming trust and regulatory uncertainty as the top obstacles to AI in clinical trials, skeptical respondents were reportedly reassured by an AI-generated executive summary confirming that everything is fine and there is nothing to worry about. Elsewhere, Lilly's communications team allegedly stared at the $1.9B Ascidian RNA-editing term sheet for six hours before discovering you can also accelerate timelines using "money" and "scientists" — a draft press release reading "no large language models were involved in the making of this asset" was ultimately scrapped. And executives hailed a record 36 AI partnerships in Q1 against a record 0 AI drugs as "proof the technology works," a claim they confirmed they would substantiate as soon as someone on the call could explain what the technology does.