Vivodyne bets AI drug discovery lacks human-tissue data, not better models

Vivodyne runs automated 'HIVE' robotic labs that grow human organ tissues, dose them with drug candidates, and sell predictive testing and training data to pharma companies.

AI drug discovery's bottleneck is causal human-biology data, not models; robotic labs dosing 20 human tissue types at scale will generate it and predict trial outcomes.

Just under $80M raised across two rounds led by Khosla Ventures (first: $38M seed, 2023)

Spun out of the University of Pennsylvania in 2021 by bioengineers Andrei Georgescu and Dan Huh. Georgescu's diagnosis: AI models trained on static cell snapshots or mouse data learn states, not causes — 'they're going to cure cancer in mice.'

Launched with a $38M Khosla-led seed in 2023, analyzing 10,000+ tissues at a time. By August 2026 it had raised just under $80M and opened what it calls the world's largest 'human data center' outside San Francisco, claiming twice the throughput of all US animal trials.

Clinical failure is expensive: trials cost tens of millions, so even small gains in preclinical prediction are worth paying for.

Animal and single-cell data is what everyone has; causal human-tissue data at scale is what nobody has.

Owning the robotic lab makes the data advantage compounding and hard to copy in software.

When every competitor trains on the same data, the moat is generating data nobody else has — own the instrument, not the model.

As of August 2026 Vivodyne is scaling its San Francisco-area 'human data center,' serving unnamed major pharma partners and positioning its experiment stream as reinforcement-learning data for future biology models. It has raised just under $80M, all Khosla-led.