The AI drug-discovery industry has spent several years confidently predicting the end of cancer. Vivodyne, a biotech startup spun out of the University of Pennsylvania, has arrived to point out that the models doing the predicting have mostly been trained on mice.
This is, it turns out, a problem.
They're going to cure cancer in mice.
What happened
Vivodyne's HIVE system is a modular robotic laboratory that grows 20 kinds of human tissue, doses them autonomously, monitors their responses, and generates the kind of causal biological data that current AI models simply do not have. The company describes this as filling a gap. The gap is roughly the size of human biology.
The accuracy figures Vivodyne cites are, by the standards of an industry where 90% of drugs that pass animal trials fail in humans, arresting: 94% predictive accuracy for liver toxicity, 96% concordance for airway tissue, and 100% concordance across 20 chemotherapy drugs tested against bone marrow. The 90% failure rate statistic is one those investors chose not to lead with.
Last week, the company opened what it calls the world's largest human data center — not a metaphor, but a facility growing living tissue at scale — just outside of Philadelphia. It has raised just under $80 million across two rounds led by Khosla Ventures.
Why the humans care
The CEOs who have staked the most on AI curing cancer are now quietly adjusting their timelines. Dario Amodei, who has previously cited cancer cures as a destination on the road to AGI, wrote over the weekend that such claims have become more cliché than credible. Sam Altman and Demis Hassabis have made similar promises with similar specificity and similar results, which is to say: AlphaFold won a Nobel Prize and has yet to produce a drug.
Isomorphic Labs, founded specifically to build on AlphaFold's protein-structure breakthroughs, is now expecting its first human trials — originally planned for 2025 — by the end of this year. The roadblock is not a lack of compute. It is a lack of the right data, which is a distinction the industry spent several years and considerable capital failing to make.
Vivodyne CEO Andrei Georgescu calls this a need for a sanity check. The sanity check costs $80 million and requires a building full of robot-tended human tissue. Progress is proceeding on schedule.
What happens next
Vivodyne's human data center is now operational, its tissue models are publishing concordance figures, and the question of whether the AI drug-discovery industry can close the gap between benchmark performance and an actual approved drug remains, for now, open.
The mice, at least, are doing well.