A Robot Lab Grows Human Tissue to Give AI the Data It Needs to Find Real Drugs

Key points

• Vivodyne, a biotech startup spun out of the University of Pennsylvania in 2021, opened what it calls the world's largest human biological data centre near San Francisco last week.
• The company's robotic labs, called HIVE, can grow 20 types of human tissue and run drug experiments on them without human operators stepping in.
• Vivodyne has raised just under $80 million across two funding rounds, both led by venture firm Khosla Ventures.
• Its lab-grown liver tissue correctly predicted human toxicity results 94% of the time; bone marrow matched outcomes across 20 chemotherapy drugs with 100% accuracy.
• Around 90% of drugs that pass animal testing still fail in human clinical trials, a long-standing problem the startup is trying to solve.

Artificial intelligence was supposed to speed up the hunt for new medicines. So far the results have been modest. A handful of AI-designed drug candidates have entered human testing, but none has reached patients as an approved treatment. Vivodyne, a San Francisco-area startup, argues it knows the reason: the AI models have terrible data to learn from.

What does the robot lab actually do?

The machines run fully autonomous experiments on living human tissue without a human researcher touching them. Each HIVE unit can grow 20 different types of human tissue inside controlled chambers, automatically apply drug compounds, and monitor what happens over time.

That last part matters. Today's AI models in biology mostly learn from static snapshots, as CEO Andrei Georgescu puts it: a photograph of a cell at one moment, not a film of what caused it to reach that state. The HIVE machines track hundreds of thousands of ongoing experiments and record how diseased tissue responds to a stimulus as it happens, building the kind of cause-and-effect record that existing datasets simply don't contain.

"All the training is done on static snapshots of these cells," Georgescu told TechCrunch. "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.'"

For readers following the broader data quality story in physical AI, our 12 August report on why bad training data stalls robotics maps onto exactly the same structural problem here, just inside a body instead of a warehouse.

Why does it matter that the tissue is human?

Animal testing fails to predict human outcomes at a striking rate. About 90% of drugs that work in animals fail when tested in people, a figure that has barely moved in decades. Georgescu's analogy: carmakers run crash tests with confidence their vehicles will pass safety standards, but drugmakers walk into expensive human trials without that same assurance.

Human clinical trials typically cost tens of millions of dollars each. Vivodyne's pitch to large pharmaceutical companies is that testing on human tissue first gives them a clearer read on what will work before committing that spend.

The early accuracy figures are striking on their face. Airway tissue matched real human behaviour 96% of the time. Liver cells predicted toxicity correctly in 94% of comparisons against actual human trial results. The honest caveat is that Vivodyne is reporting these numbers itself; independent replication is what the field will need before pharma procurement teams fully trust the platform.

What happens next?

Vivodyne is already working with multiple major pharmaceutical companies, though it won't name them. Georgescu says his facility is currently running twice the throughput of all animal trials held in the US combined, a claim that, if accurate, represents a real shift in how early drug testing could be done.

The longer goal is to use that experimental data to train AI models that understand cause and effect in human biology. Without it, Georgescu argues, AI drug discovery keeps producing results that work in mice and collapse in people. "Absent human testing," he told TechCrunch, "these models are going to cure cancer in mice."

Common questions

Does this mean AI will cure cancer soon?

Not yet. Even Anthropic CEO Dario Amodei wrote recently that claims AI will cure cancer have become more cliché than credible. Vivodyne's work is an early-stage attempt to build better tools; useful drugs from this approach are still years away from patients.

Is the tissue real human tissue or a simulation?

It is neither a computer simulation nor tissue taken from patients. Vivodyne grows the tissue from human cells in controlled lab conditions, producing structures that behave like real human organs without coming from a living person.

What does this mean for animal testing?

Vivodyne's founders believe high-quality human tissue data could eventually replace much of the animal testing done in early drug development, though that shift would require regulatory bodies to accept the new data as sufficient evidence of safety.