The promise of AI curing cancer has become so common that even AI leaders are calling it cliché. But one biotech startup, Vivodyne, says it has identified the real bottleneck—and built a machine to fix it. The company believes the entire AI drug discovery industry suffers from a critical data problem, not a lack of algorithms or computing power.
AI drug discovery has generated plenty of headlines but few tangible results. A handful of AI-designed drugs have made it to human trials, with one reaching Phase III, but the reality is that today’s AI models lack the right kind of data to truly understand human biology. Nobel-winning AlphaFold advanced our understanding of protein structures, but it hasn’t produced a new drug yet. Isomorphic Labs, built on AlphaFold, pushed its first expected trials from 2025 to the end of this year.
Why Current AI Models Fall Short
Vivodyne CEO Andrei Georgescu puts it bluntly: the industry needs “a sanity check.” Existing models don’t have the data to capture human biology’s complexity. Most training happens on static snapshots of cells or single proteins, not living tissue. “All the training is done on static snapshots of these cells, and the models are not conditioned at all by the how a cell got to that state,” Georgescu explains. “The model learns ‘this is cell state A,’ ‘this is cell state B,’ but never ‘cell state B is the effect of inflaming cell state A.'”
This data gap helps explain why 90% of drugs that pass animal testing still fail in human clinical trials. The pharmaceutical industry spends tens of millions per trial, often with little confidence in the outcome. Even Anthropic CEO Dario Amodei recently noted that claims about AI curing cancer have become more cliché than credible, though he’s made similar claims himself in the past.
Vivodyne’s Solution: The HIVE Lab
Vivodyne’s approach differs from traditional AI drug discovery methods. Spun out of the University of Pennsylvania in 2021, the company has built modular robotic labs called HIVE that can grow 20 types of human tissue, then autonomously dose and monitor them. This generates causal biological data from living tissue—the kind that today’s AI models desperately need.
Human Tissue with High Predictive Accuracy
The company’s tissues closely match real human organ behavior. Their liver cells show 94% predictive accuracy compared to human toxicity trials, airway tissue matches real behavior 96% of the time, and bone marrow achieved 100% concordance when testing 20 different chemotherapy drugs.
Last week, Vivodyne opened what it calls the world’s largest “human data center” just outside San Francisco. The company, which has raised nearly $80 million across two rounds led by Khosla Ventures, says it’s already achieving twice the throughput of all animal trials conducted in the U.S. While Vivodyne won’t name partners publicly, it’s working with multiple major pharmaceutical companies.
Building the Foundation for Better AI
Georgescu sees these autonomous biology labs as essential for generating causal data that can train new AI drug discovery models. He points to recent studies finding no clear data scaling laws when training generative AI on existing cellular data. The HIVE machines track hundreds of thousands of ongoing experiments where diseased tissue is exposed to various stimuli, potentially enabling reinforcement learning that produces models capable of understanding human biology more meaningfully.
The Future of Combination Therapies
Looking ahead, Georgescu believes this approach will be crucial for developing combination therapies that target multiple disease pathways—something most current drugs can’t do. “If we want combination therapies, the space that has to be searched explodes—it can’t be an experimental approach,” he says. “You have to say, ‘I want this effect to happen, so what cause should I invoke?’ Establishing causality in human biology is the basis of all of this.”
A Sanity Check for the Industry
The broader vision is accelerating drug candidates through the pipeline by providing better certainty before expensive clinical trials begin. Georgescu compares it to automotive crash tests—automakers are confident their cars will pass safety requirements before testing, but drugmakers rarely have that same confidence going into human trials.
AI drug discovery may not be close to curing cancer, but Vivodyne’s approach suggests a clearer path forward. By focusing on generating the causal data that current models lack, the company is building what could become the foundation for more effective AI in healthcare—not by replacing human testing, but by making it more successful when it happens.
