AI Labs Are Buying Biology Data. What Should a Biotech Keep?
The new money is good news for startups with strong assays. The asset that lasts is the record of how each experiment was run.
By Yaoyu Yang ·
AI labs are going all in on buying biology data. That's good news for bio startups who have data, with a caveat.
This all came into the limelight in early October 2026, when Nikhil Raman, an investment partner at Union Square Ventures, wrote that the large labs started buying bio data six to twelve months ago. In his words, "this spend is starting to ramp very aggressively." A day later, Google DeepMind, Isomorphic Labs and Meta said they would put $300 million into Biohub's Virtual Biology Initiative.
WHY ARE AI LABS BUYING BIOLOGICAL DATA?
AI labs are buying biological data because a biology-based model can only learn from experiments someone has run. Biohub's initiative is built around that gap. Its goal is cell response data "across far more cell types and conditions than have yet been studied," to train predictive models of cells.
Raman's view is that the best startups will use this money to "supercharge their novel assay and work towards independently powerful models and eventually directly produce therapies." We agree. A team with an assay that produces clean, standardized data at scale now has a second customer for it.
WHO IS ALREADY BUILDING A DATA BUSINESS AROUND AN ASSAY?
A-Alpha Bio is one of the teams ahead of the curve. In July 2026 it launched the Atlas Consortium with five founding members, including GSK, Boltz, Cradle and Dyno Therapeutics. Members pay a subscription and design new antibody-antigen datasets each quarter, and A-Alpha runs them on its AlphaSeq platform.
The pooled data goes to every member, about 28 million affinity measurements per member per year. The data is only part of A-Alpha's advantage. The other part is a platform that can keep producing standardized, high-quality measurements, quarter after quarter.
HOW WILL AI DATA BUYING CHANGE BIOTECH?
Pharma companies and AI labs want different data. A pharma company buys data for one program, but an AI lab wants datasets it can reuse across many problems. That means data collected for one buyer rarely suits the other.
The buying will also get more targeted. Patrick Malone, a partner at KdT Ventures, expects the labs to move on from buying datasets that already exist. The next step, he wrote, is "commissioning specific experiments that resolve gaps in a model's understanding," because "a dataset is a one-time sale."
WHAT IS THE RISK FOR A STARTUP SELLING DATA?
Selling data built for a model is a risk unless the company can also reuse it for its own science, and we'd take that risk seriously. We think what sets a team apart is what it keeps after the sale: the record of how it ran its experiments. That record stays the team's even if the labs stop buying.
Startups can get so focused on delivering a dataset that documenting the work becomes an afterthought. Later, when someone needs to reconstruct a result, the team may struggle to recover the protocol, conditions and decisions behind it.
That gap gets expensive when a team wants to file an IND or sell an asset. An IND has to describe how the drug is made and controlled, and how each nonclinical safety study was run. GLP study reports also have to describe the methods used and any changes to the protocol. Meeting those requirements usually means paying consultants. An NIH guide for startups puts regulatory consulting for a biologic program at $100,000 to $1 million, and rebuilding a lost record adds to that bill.
The same record also makes the next sale easier. If the labs start commissioning experiments, as Malone expects, we think the work will go to teams that can show how each dataset was produced and run the next one the same way.
WHAT SHOULD A TEAM DO WITH THE MONEY?
If we were selling data to an AI lab today, we'd put the money back into two things. The first is the assays themselves. The second is infrastructure that captures every experiment, its context and how it was run, in a format that both scientists and AI agents can use.
In practice, that record has a few parts. Each protocol is versioned, and each run records the version it used. Each sample links to the step that made it, and each data file links to the samples it came from. With those links in place, a dataset sold to a lab can still be traced back to the runs behind it.
This is a big part of why we're building Cypher AI. Cypher AI is digital infrastructure for life-science R&D: an ELN, a LIMS, data analysis and a services Marketplace in one platform, with an AI agent that works across them. We think the long-term value is in keeping biological data connected to the experiments, workflows and decisions behind it, as well as in generating more of it.
FREQUENTLY ASKED QUESTIONS
What is Biohub's Virtual Biology Initiative? It is Biohub's program to build the datasets and tools for predictive models of cells. Google DeepMind, Isomorphic Labs and Meta's $300 million is part of about $1.8 billion committed with the US Department of Energy, NIH and Biohub itself.
What kinds of biological data are AI labs buying? Mostly assay outputs, produced at scale in a standard way. Biohub's initiative is generating data on how cells respond to interventions, and A-Alpha's consortium produces antibody-antigen affinity measurements.
How much are AI labs paying for biological data? Few deals disclose a price. Raman has said, as an anecdote, that he has heard of labs spending in the mid seven figures to low eight figures with a single data provider.
HOW CAN YOUR TEAM START?
Choose an assay your team runs often. Record the next few runs against a versioned protocol and link each sample to the step that produced it, so you can trace a result back to the work behind it. Create a free Cypher AI account and book a 15-minute onboarding call from the home screen.
For R&D organizations that need their own cloud, SSO or a forward-deployed engineer working alongside the team, the Enterprise plan covers that. You can book a conversation through the demo link on the site.