AI isn’t close to curing cancer. This startup says it knows what it will take.

Details

Vivodyne, a University of Pennsylvania spinout that has raised nearly $80 million from Khosla Ventures, argues that AI drug discovery is failing not because models are too weak but because they're trained on the wrong data — animal testing results and single-cell studies rather than causal data from living human tissue. Its HIVE platform grows 20 kinds of human tissue in modular robotic labs, autonomously doses them, and monitors outcomes at scale. The company reports concrete concordance numbers: 94% predictive accuracy on human toxicity trials for liver cells, 96% match for airway tissue, and 100% concordance across 20 chemotherapy drugs tested on bone marrow. Georgescu claims HIVE is already achieving twice the throughput of all US animal trials combined — though those figures are self-reported. Context matters here. Dario Amodei, Sam Altman, and Demis Hassabis have all publicly predicted AI will cure diseases, yet the actual pipeline is thin: a handful of AI-designed compounds entered human trials, one reaching Phase III, but none approved. AlphaFold hasn't produced a single approved drug; Isomorphic Labs' first human trials, planned for 2025, are still forthcoming. Vivodyne's thesis is a data infrastructure bet: better human-derived training data will unlock drug discovery in ways that better models alone cannot.