Meta Platforms Inc. (NASDAQ:META) and two Alphabet Inc. (NASDAQ:GOOGL) units, Google DeepMind and Isomorphic Labs, are jointly putting $300 million into a Mark Zuckerberg-backed effort to build AI models that can predict how living cells behave.
The investment expands Biohub’s $1.8 billion Virtual Biology Initiative. The research nonprofit founded by Zuckerberg and Priscilla Chan wants to create a “universal virtual cell” that could predict how cells respond to drugs, genetic changes and other interventions.
The bottleneck is data. While language models could learn from vast amounts of text that already existed, AI biology depends on experimental data that often has to be created from scratch.
Biohub wants to produce that data at scale, potentially allowing researchers to test ideas in software before committing resources to laboratory work.
Commercial partners will get one year of exclusive access to the data they help generate before it is released publicly, Axios reported.
Big Tech Wants Biological Data
Biohub head of science Alex Rives previously led Meta’s protein-language-model research at Meta’s FAIR, where his team trained AI systems on hundreds of millions of protein sequences.
Biohub expects its first large dataset within about a year and is targeting accurate predictive models within five years, Reuters reported.
The $1.8 billion is not all new funding. It includes Biohub’s earlier $500 million commitment, more than $500 million of planned Department of Energy spending and NIH datasets created through more than $500 million of previous federal investment.
Google Has a Drug-Discovery Bet
Alphabet could put the Biohub data to work through Isomorphic Labs, its AI drug-discovery company led by Demis Hassabis.
Isomorphic raised $2.1 billion in May to expand its AI drug-discovery platform and move potential medicines toward clinical trials. The company has partnerships with Eli Lilly, Novartis and Johnson & Johnson, and Benzinga previously covered its expansion.
Whether simply adding more data produces better single-cell models remains unproven. A June Nature Methods study trained 400 single-cell foundation models across 6,400 experiments and found performance often plateaued before researchers used all available training data, with no clear data-scaling laws like those seen in large language models.
AI Spending Pushes Into Biology
The investment comes as prediction market traders see little chance of the AI boom breaking this year.
Polymarket traders currently put a 6% chance on an AI-industry downturn by Dec. 31, with roughly $2.4 million traded on the contract.
Meta and Alphabet are extending the AI spending race into biology, putting $300 million behind a one-year head start on the data needed to train virtual-cell models.
Image: Shutterstock
Login to comment