Last Thursday, Anthropic acquired Coefficient Bio, a stealth biotech AI startup based in New York, for just over $400 million in stock.

The company had fewer than 10 employees and was eight months old. It had no publicly disclosed product, no revenue, and no conventional traction metrics of any kind.

Dimension, the VC firm that reportedly held roughly half the company, banked a 38,513% paper IRR on the investment.

Those headline numbers will get debated endlessly. Is it too much? Are we in an AI bubble? Is this an acqui-hire gone wrong?

I think those are the wrong questions.

The right one is, in my view, is: what did Anthropic need so badly that they paid $400 million to get it delivered overnight?

Coefficient Bio was co-founded by Samuel Stanton and Nathan Frey. Both came from Prescient Design - Genentech’s computational drug discovery unit. Nearly the entire team came from the same group.

These aren't first-time founders who built an AI wrapper and got lucky. Frey served as Group Leader and Principal Scientist at Prescient Design, has over 20 publications in leading scientific and machine learning journals, and won a prestigious paper award in 2024 for his work on using generative AI to discover new proteins. He was recently named a 2026 Termeer Fellow, a recognition reserved for emerging biotech leaders. His work has been cited over 5,600 times by other researchers.

Stanton served as Principal Machine Learning Scientist at Genentech, and his research collaborators are now spread across Anthropic, OpenAI, Isomorphic Labs, and Microsoft Research. In January, he posted on X: “We’re ushering biopharma into the Intelligence Age.”

Now you could be doing the napkin math: Anthropic paid roughly $40 million per person for this team. That’s more per employee than most biotech startups raise in their entire seed or Series A round.

Looking at the deal from that lens, it can sound absurd until you realize that people who can build biological foundation models and deploy them inside real drug discovery programs might be the scarcest talent right now.

Now put the price in Anthropic’s context. $400 million against a recent $380 billion post-money valuation equals roughly 0.1% dilution. Against a revenue trajectory targeting up to $18 billion in 2026, this is a line item.

In a race where speed is everything, this wasn't about acquiring a mature asset. It was one move in a larger, coordinated strategy - one that becomes much clearer when you look at what Anthropic has been building since last October. More on that below.

Here's something that I find beautiful: Biology is a language problem.

DNA, RNA, and proteins are sequences. The same transformer architectures that power Claude and ChatGPT can be trained to “read” and “write” biological sequences.

AlphaFold proved this in 2020 when it solved the protein structure prediction problem - a breakthrough that won its creators the Nobel Prize in Chemistry and has since been used by over 3 million researchers worldwide.

That breakthrough created an entirely new category: computational biology powered by foundation models. AI-enabled workflows are now compressing early discovery timelines by 30–40%. More than 200 AI-designed drug programs are in clinical development. The first AI-discovered drug approval is anticipated in 2026–2027.

This is the wave that Anthropic is positioning itself to ride - not by discovering drugs, but by becoming the intelligence platform that the entire industry builds on.

But they're not the only ones who see the opportunity. Google and OpenAI are each making fundamentally different bets on how to own this intersection.

Understanding those bets is the key to understanding the Coefficient Bio deal.

These three companies are making the biggest bets at the intersection of AI and life sciences right now, but each is pursuing a fundamentally different strategy. Understanding the differences tells you where the opportunities, and the risks, are most likely.

The pipeline play.

Google arguably has the strongest foundation in computational biology. AlphaFold won the Nobel Prize in Chemistry, has been used by over 3 million researchers globally, and predicted the structures of more than 200 million proteins - solving a problem the field had been working on for half a century.

Now they're commercializing that advantage through Isomorphic Labs, a dedicated drug discovery spin-off running 17 active programs in oncology, immunology, and cardiovascular disease.

Their partnerships with Eli Lilly and Novartis carry a combined ~$3 billion in potential milestone payments. Their proprietary discovery engine, IsoDDE, was described in February by a Columbia computational biologist as advancing the field on the scale of an “AlphaFold 4.” And an AI-designed cancer drug is entering clinical trials this year.

What's worth paying attention to here is Google's theory of where the value is. Rather than building tools for the broader biotech ecosystem, they're betting that the biggest opportunity is in discovering drugs directly - using their AI advantage to compete in the pharmaceutical value chain itself. Isomorphic Labs is structured as a drug discovery company, not a platform provider.

The decision to keep IsoDDE proprietary reinforces this direction. AlphaFold's open-source release transformed the field and earned enormous goodwill. The shift to closed-source with IsoDDE suggests Google sees the next phase of AI in biology as a proprietary race rather than an open ecosystem play.

The distribution play.

OpenAI's approach to life sciences starts from a very different position: they already have the audience. Over 230 million people ask health and wellness questions on ChatGPT every week, making health one of the platform's most common use cases.

They're leaning into that. ChatGPT Health launched in January 2026 with the ability to connect medical records, Apple Health data, and wellness apps like MyFitnessPal and Function directly into the chat experience.

On the enterprise side, OpenAI for Healthcare rolled out HIPAA-compliant tools to institutions including Memorial Sloan Kettering, Stanford Medicine, Cedars-Sinai, and HCA Healthcare. And on the research front, they've partnered with Retro Biosciences on longevity and have an existing relationship with Moderna.

What's interesting about OpenAI's approach is the bet they're making on where AI creates the most value in healthcare. They're not going after drug discovery or clinical operations - they're going after the layer where patients interact with the healthcare system and where clinicians manage their day-to-day workflows.

Given the scale of people already using ChatGPT for health questions, that’s a reasonable bet.

The platform play.

Anthropic's strategy is distinct from both Google and OpenAI. They're not trying to discover drugs. They're not trying to be the consumer's health assistant. They're positioning Claude as the intelligence layer that pharma, biotech, health systems, and startups all build on.

The infrastructure they've put in place is already substantial. Claude for Life Sciences launched in October 2025 with connectors to PubMed, Benchling, and scientific databases. Claude for Healthcare followed at JPM26 in January with HIPAA-ready infrastructure, clinical trial connectors to Medidata and ClinicalTrials.gov, and integrations spanning the CMS Coverage Database to ICD-10 codes.

On the enterprise side, Sanofi has most of their employees using Claude daily through an internal app called Concierge. Banner Health, a 33-hospital system with over 55,000 staff, is processing clinical notes through it. Novo Nordisk, Genmab, and Komodo Health are also active partners.

The Coefficient Bio acquisition is the next layer of this strategy. It brings in a team that can help Claude move beyond processing text about biology to genuinely reasoning about molecular structures, protein interactions, and drug candidates at a structural level. That's the gap between being a useful tool for scientists and becoming the platform they can't work without.

When you look at Anthropic’s moves in sequence, the Coefficient Bio deal doesn’t look impulsive. It looks like the final piece of a strategy that’s been building for some time:

October 2025 - Claude for Life Sciences.

Connectors to PubMed, Benchling, scientific databases. Focus on preclinical R&D: literature reviews, hypothesis generation, protocol writing, bioinformatics.

January 2026 - Claude for Healthcare (JPM26).

HIPAA-ready infrastructure. Connectors to Medidata, ClinicalTrials.gov, ICD-10, CMS Coverage Database. Prior authorization. Claims appeals. Consumer integrations with HealthEx, Function Health, Apple Health.

January 2026 - Aggressive hiring. Open roles for Research Scientists in Biology & Life Sciences, Product Partnerships Lead for Life Sciences & Healthcare, and Product Public Policy for Health & Life Sciences (requiring “strong relationships with FDA, CMS, HHS”).

February 2026 - $30 billion Series G. $380 billion post-money valuation. War chest fully loaded.

April 2026 - Coefficient Bio. Bringing computational biology DNA in-house. (The pun is unavoidable 🙂)

Here’s what makes this acquisition different from a typical talent grab.

Anthropic had the platform, they had the connectors and they had the enterprise relationships. They had HIPAA infrastructure and a partner ecosystem spanning Deloitte, Accenture, KPMG, AWS, and Google Cloud.

What they didn’t have was anyone who had actually built biological foundation models and deployed them inside a real drug discovery program.

Claude could process text about biology. It could summarize papers, generate protocols, and assist with bioinformatics.

But it couldn’t reason about biology the way it reasons about code. It couldn’t look at a protein sequence and understand what that structure implies about function, binding affinity, or drug potential.

Frey’s team knows exactly what training data, evaluation benchmarks, and fine-tuning strategies would close that gap. They built this capability at Genentech.

They know what biology-specific model development actually looks like - not in theory, but from deploying it inside one of the world’s largest drug discovery operations.

You can’t hire that expertise one person at a time from LinkedIn. Either the team exists or it doesn’t. This one existed. And it was available for 0.1% of Anthropic’s equity.

Eric Kauderer-Abrams, who leads Anthropic’s healthcare and life sciences division, has been clear about the ambition. He’s said Anthropic wants a meaningful percentage of all life science work in the world to run on Claude - the same way coding does today.

This is where I’d like to go a bit off-record and share what I’ve been thinking about since last year when Anthropic started make its presence felt in life sciences.

To start, I asked myself a simple question: if I was in charge of Anthropic’s strategy in life sciences, what would “great” look like for me in 2-3 years? What would be our north star?

When you map Anthropic’s moves over the past 18 months against what they’re hiring for, who they’re partnering with, and now what they’ve acquired, a picture starts to form.

It’s not confirmed strategy, but I think the signals are consistent enough that it’s worth laying out what I believe they may be building toward.

This is already underway. Claude for Life Sciences connects to PubMed, Benchling, and scientific databases. Scientists can do literature reviews, generate hypotheses, analyze genomic data, and draft protocols. Sanofi has most of their employees using Claude daily through an internal app. Banner Health has 55,000+ staff processing clinical notes through it.

With the Coefficient Bio team, this phase shifts. Instead of Claude being a smart text processor that’s useful for biology, the goal is Claude that genuinely reasons about molecular structures, protein interactions, and biological pathways.

This is the difference between a tool that can summarize a paper about a protein and one that can reason about the protein itself.

The connectors to Medidata and ClinicalTrials.gov are the early wedge. The Product Public Policy hire (requiring FDA, CMS, and ONC relationships) signals where this is going: Claude embedded in the clinical trial design and regulatory submission process.

If Claude can help design a trial protocol, monitor enrollment and safety signals, and draft regulatory responses that actually pass FDA review - that’s a fundamental shift in how the development pipeline operates. And it’s the kind of workflow that, once embedded, becomes very difficult to replace.

This is the endgame I think Anthropic is building toward: being the platform that pharma, biotech, health systems, and startups all run on. The AWS of life sciences intelligence.

If that happens, the economics become very attractive. Enterprise contracts with pharma companies are high-value, multi-year, and sticky. Anthropic’s revenue trajectory is already pointing toward $18 billion in 2026. Even a 10–15% share from healthcare and life sciences would represent billions annually by 2028–2029.

I've spent most of this newsletter laying out why this deal makes strategic sense. But no analysis is complete without stress-testing the thesis.

So let's ground ourselves - here's what could go sideways.

Acqui-hires fail all the time.

A 10-person biology team moving inside a 1,500+ person AI company is a culture and integration risk. The best researchers leave when they lose autonomy. Anthropic’s ability to keep this team operating with startup-level speed inside a scaling organization will determine whether this acquisition creates value or just created headlines.

Biology models need biology data.

Training specialized models that genuinely understand molecular biology requires massive, high-quality biological datasets that Anthropic doesn’t currently own. Pharma companies are notoriously protective of proprietary data. Anthropic will likely need additional partnerships or acquisitions to solve this.

Google has a deeper moat in biology.

AlphaFold has a 5-year head start and 3.3 million users. Isomorphic Labs has 17 drug programs and $3 billion in pharma deals. Anthropic is playing catch-up on the pure biology side. Their advantage is platform and enterprise trust, but that’s a bet on distribution over depth.

The clinical proof is still coming.

Phase III results for AI-designed drugs are expected in 2026–2027. If they succeed, this entire space accelerates. If they don’t, expect a valuation correction that would make the Coefficient Bio price tag look very different.

None of these risks invalidate the strategic logic. But they’re worth keeping an eye on.

This deal doesn’t just validate a category; it reveals how big AI companies are approaching life sciences.

Anthropic had the platform, the partnerships, the infrastructure. What they couldn’t build was biological intuition. The fact that they went outside to acquire it tells you something important: this expertise doesn’t emerge from scaling general-purpose models. It comes from people who’ve spent years at the bench and in the codebase.

If that describes your team, your value proposition just got validated at the highest possible level.

Diagnostics. Clinical trial optimization. Regulatory technology. Lab data infrastructure.

If any of these describe your company, understand that the big AI platforms are building the intelligence layer, not the vertical applications. They need companies like yours to complete the ecosystem.

Google is going deep into drug discovery. OpenAI is going wide into consumer health. Anthropic is building enterprise infrastructure. But a platform without specialized applications on top of it is just a hammer without nails.

Someone still needs to build the tools that make AI useful inside a specific clinical trial, a specific regulatory submission, a specific diagnostic workflow. If you’re building in this space, you may now have a partner, a customer, or a potential acquirer.

If you're actively fundraising, this deal gives you a concrete reference point to weave into your narrative. When Anthropic pays $400 million for a pre-revenue team at the AI-biology intersection, it becomes much easier to make the case for why your approach sits in a similar value zone.

It's also worth looking at who backed this deal. Investors in the Dimension, Lux Capital, and broader AI-biology ecosystem just got a very public signal that this category returns capital.

Funds that have recently closed and are watching this space will be deploying with a sense of urgency - and that's exactly the kind of momentum you want to be fundraising into.

This deal is one data point in a much larger shift.

AI isn't coming to life sciences. It's already here. The question now is which founders, which companies, and which investors will be positioned on the right side of it.

By the way, I genuinely enjoyed doing this teardown.

And the process of pulling it all together inspired an idea: a broader “State of AI in Life Sciences” report - covering the full competitive landscape, where capital is flowing, which founder categories have the strongest tailwinds, and what the next 2–3 years look like for anyone building at this intersection.

Would that be valuable to you? If so, just reply “yes” - if enough of you are interested, I'll start working on it.

See you next Sunday!

- Vadim

PS: I'd love to hear from you on this one. What did you think of the teardown? Anything I missed? What's your take on the deal? And would you want to see more deep dives like this? Hit reply and let me know!

PPS: When you’re ready, here’s how I can help: