Big Tech’s $1.8 Billion Bet to Teach AI How Life Works

Did you know Big Tech and the US government just bet $1.8 billion on teaching AI how life works?

That’s right! On October 7, Biohub announced a major expansion of its Virtual Biology Initiative. Biohub is the science group backed by Mark Zuckerberg and Priscilla Chan’s Chan Zuckerberg Initiative. Its goal is bold. It wants to build open biological datasets big enough to train AI. Those models could then predict how cells behave, and eventually how to treat disease.

So who is paying? Google DeepMind, Isomorphic Labs and Meta are adding $300 million combined. Meanwhile, the US Department of Energy is committing more than $500 million over five years. Biohub itself committed $500 million in April. Finally, the NIH brings more than $500 million of existing datasets. Altogether, that adds up to roughly $1.8 billion. In fact, Biohub’s own headline says “nearly $2 billion.”

Where the $1.8 billion comes from

Diagram: funding behind Biohub’s Virtual Biology Initiative, by CVisiona (https://cvisiona.com)

However, the fine print matters. As eWeek points out, the total mixes new money with existing resources. For example, the NIH share is mostly data that already exists. NVIDIA and Renaissance Philanthropy also joined. Instead of cash, they bring computing power and fundraising help.

Several research groups are taking part too. These include the Allen Institute, the Broad Institute and the Wellcome Sanger Institute. The Human Cell Atlas and the Human Protein Atlas have also signed on.

You see, reading isn’t understanding

Today’s AI models have read millions of science papers. But reading about a cell is not the same as knowing what happens when you change it. Papers describe results. However, they rarely include the raw, consistent data a model needs to learn cause and effect.

That is why this project focuses on experiments. According to Reuters, the data will include screens that show how cells react to changes. It will also include maps of gene activity and images of billions of cells in living tissue.

Scale is the challenge. Biohub’s head of science, Alex Rives, says today’s datasets cover hundreds of millions of cells. Accurate models, by contrast, will need billions, and eventually trillions.

We need to capture the language of biology, we need to capture the language of the cell.

Alex Rives, Head of Science, Biohub, via Reuters

There is a famous precedent. AlphaFold could predict protein shapes because scientists had spent decades collecting real protein structures in a public database. Similarly, this project wants to build the same kind of training library for whole cells.

We will not solve this challenge without open, experimental biological data at an unprecedented scale.

Pushmeet Kohli, VP of AI for Science, Google DeepMind

Why the US government is in

The Department of Energy calls its role part of the Genesis Mission. That is the federal push to use AI for science. As a result, its national labs bring supercomputers, genome centers, advanced microscopes and robotic labs. In other words, AI for biology is becoming an infrastructure story. We explained that trend in why AI is becoming an infrastructure problem.

Biohub’s bigger bet

This is not Biohub’s first AI move. Last November, the AI lab EvolutionaryScale joined Biohub. Its co-founder, Alex Rives, became head of science. At the same time, Biohub said it plans to grow to 10,000 GPUs by 2028.

Since then, it has released free models on its virtual cells platform. It has also set a grand challenge: one unified AI model of the cell. Priscilla Chan summed up the ambition to Reuters: “Biology has been just sort of a clever discovery-based science until this point.” The new data aims to make it predictive.

But here’s the twist

The goal is not another medical chatbot. Instead, it is basic data that could help AI predict biology before anyone runs the experiment. Rives expects the first dataset in about a year. He hopes for accurate models within five years. Still, those are goals, not promises.

And “open” comes with an asterisk. According to Axios, as reported by The Next Web, commercial funders get about a year of early access. That applies to data they help create. Government-funded work, on the other hand, has no such limit. Rives explained why: “We have to have some incentive for commercial players.” This tension will sound familiar if you follow the open-weight AI debate.

The open questions

Nobody knows yet whether cell models improve steadily as data grows. After all, language models did, but biology may not. Also, much of this data still has to come from wet labs, which is slow and expensive. The real test, as eWeek notes, is simple. Can the models predict experiments they never saw in training?

Meanwhile, the race is on. Reuters notes that Anthropic has added a biology wet lab. The OpenAI Foundation, for its part, runs a $125 million-plus grant program for biological and medical datasets.

For founders, open science data could become a new platform to build on. It looks a lot like the opportunities we explored in open-source AI infrastructure.

Think of it as a new training library for biology. Because the next big AI breakthrough might not be writing better emails. It might be understanding life itself.

Sources & Credits

• Text credit: Biohub, “International, cross-sector collaboration commits nearly $2 billion…” (October 7, 2026)
• Text credit: US Department of Energy, Office of Science (October 2026)
• News reporting source: Reuters via Investing.com (October 7, 2026)
• News reporting source: The Next Web (October 2026)
• News reporting source: eWeek (October 9, 2026)
• Background: Biohub Virtual Biology Initiative launch (April 29, 2026)
• Image: Biohub. Diagram: CVisiona

Should AI training data for biology be fully open, or is a head start for funders fair? Tell us in the comments.