This is the kind of project that sounds unquestionably good—until you notice who’s holding the keys.
A nonprofit called Biohub is teaming up with the US government and some of the biggest AI players—Google, Meta, and a few research groups tied to them—to pour about $1.8 billion into what they’re calling the Virtual Biology Initiative. The headline promise is simple: build big, open biology datasets that show how cells respond under different conditions, then use AI to speed up research. Public reporting also says there’s a $300 million joint investment coming specifically from Meta, Google DeepMind, and Isomorphic Labs.
On paper, this is hard to hate. Biology is messy. Experiments are expensive. Data is scattered, locked up, or collected in ways that don’t line up. If you can create shared datasets that lots of researchers can use, you lower the cost of discovery for everyone. You make it easier for a small lab to compete. You reduce the odds that ten teams run the same experiment in parallel because nobody can see what already exists.
But I don’t trust the “open data” story when the same companies funding the work have a direct business reason to shape what “open” means.
The real prize here isn’t just knowledge. It’s the dataset itself. Whoever helps define what gets measured, how it gets labeled, and what “good quality” looks like quietly sets the rules for the next decade of AI-in-biology. You don’t need to “own” the data to benefit from it. You just need to be best positioned to use it—meaning compute, talent, and the ability to train huge models on it fast.
That’s where the tension lives. A nonprofit plus government agencies like the Department of Energy and the National Institutes of Health gives this a public-interest glow. The tech funding gives it muscle. Together, they can do something universities and small labs can’t: build a massive, standardized, reusable data engine.
And that’s exactly why we should be careful.
Imagine you’re a grad student in a mid-size university lab. If this initiative works, you might be able to pull down datasets that would have taken your lab years and a fortune to generate. That’s amazing. You could test ideas faster. You could focus your time on better questions instead of just collecting raw measurements.
Now imagine you’re a startup trying to develop a therapy or diagnostic tool. Open datasets could cut your early research costs. But it could also raise the bar so high that only the biggest players can keep up. If the “open” dataset becomes the default training ground, then the companies with the biggest AI infrastructure will still be the ones who can squeeze the most value out of it. “Open” doesn’t automatically mean “fair.”
There’s also a quieter risk: the dataset becomes the map, and everyone stops exploring off-map.
If this initiative chooses certain cell types, certain measurement tools, certain conditions, it will naturally steer what researchers study because people follow what’s easiest to access, easiest to publish, easiest to compare. That can speed up science, sure. It can also narrow it. Biology is full of surprises that show up because someone looked in a weird place with a weird method. Big shared datasets can accidentally turn “weird” into “unfunded.”
And let’s talk about incentives, because they don’t disappear just because a project sounds noble.
Google and Meta aren’t putting money into this because they woke up and decided to be patrons of basic science. They want better models, better tools, better platforms, and eventually better products. That doesn’t mean they’re villains. It means their goals will always compete with the messy, slow, sometimes-unprofitable work of science. When tradeoffs come—like what data can be shared openly, what gets restricted, what gets prioritized—who do you think wins the argument in the room?
The best version of this is still worth wanting. Shared biology datasets could help researchers understand disease faster. They could make it easier to validate results. They could reduce waste. They could even help train the next generation of scientists on real data instead of toy examples. If you care about progress, you should want more public-good infrastructure like this.
But the worst version is also easy to picture.
Open in name, but practically gated by compute costs. Open, but shaped by the needs of a handful of companies. Open, but with terms that quietly tilt advantage toward the funders. Open, but creating a single “standard” pipeline that becomes a choke point for what counts as legitimate research.
And because this involves government agencies, there’s a second layer: legitimacy. Once public institutions lend their credibility, it becomes much harder to question the direction of travel without sounding anti-science. That’s convenient for everyone who wants to move fast and avoid hard debates about control.
What I genuinely don’t know is how the governance will work in practice. Who decides what gets collected? Who decides what “open” actually means? Who gets to audit the dataset quality, the labeling choices, the gaps? If it’s truly built to serve the broad public, those answers should be boring and transparent. If they’re fuzzy, that’s the tell.
So yes, fund this. Build the datasets. Move biology forward. But don’t pretend this is just charity with microscopes. This is power—power to set the agenda, power to define standards, power to turn biology into a data problem that certain players are uniquely built to solve.
If this becomes the foundation for the next era of biology, who should get to decide what the foundation looks like?