Ex-Meta researchers raise $40M for biotech AI startup

Meta is all in on commercial chatbots, and that focus just pushed one of the top AI biology teams in the world out the door.

After Meta shut down its protein-folding research unit in April, former team lead Alexander Rives and his entire eight-person group stuck together. They launched a new startup called EvolutionaryScale and have now raised roughly $40 million in a seed round led by Lux Capital, according to Forbes. Prominent AI investors Nat Friedman and Daniel Gross also joined, valuing the company at $200 million.

Scaling protein language models

Proteins are the building blocks of cells, and their 3D shape dictates how they work and which drugs can bind to them. Figuring out that shape used to take years of brutal lab work.

While DeepMind made history in 2020 by releasing AlphaFold to predict 200 million protein structures, Rives' group took a different angle. They applied Transformer architectures—the same backbone behind ChatGPT—directly to biological sequences. By treating proteins like text, their model could predict unknown structures.

Last November, the team published a paper in Science showing their system ran 60 times faster than AlphaFold, even if it was slightly less accurate on average. That speed allowed them to build a database of 700 million possible 3D structures. Per pitch documents, they are raising money to dramatically scale up the size of this biological model to see what new properties emerge.

Big tech shifts, VCs step in

Meta originally built its AI lab for blue-sky exploration, but the generative AI frenzy forced a pivot toward immediate consumer and business products. Much like OpenAI disbanding its robotics team back in 2021, Meta decided biology was too distant from its bottom line to keep funding.

AI for biology is notoriously hard to monetize in the short term. Even established players like Schrödinger trade at sub-$3 billion market caps relying on older molecular modeling tools, and traditional pharma remains skeptical of a ChatGPT-style eureka moment.

Still, venture capital is eager to take the bet. In a market where unicorn foundation model labs like Inflection, Cohere, and Adept are pulling in massive funding rounds, EvolutionaryScale is betting that sheer model scale will crack the code of biological design.

Why it matters

For biopharma researchers and synthetic biologists, EvolutionaryScale offers a fundamentally different tool. Trading a bit of precision for a 60x speed bump lets scientists screen millions of candidate molecules in days instead of months. If their upcoming scaled models become accessible via API or open weights, early-stage drug pipelines would be smart to test them immediately.

However, biology has a brutal reality check that text models don't face: wet lab experiments. In software, a model hallucinating a few words is fine; in drug discovery, being slightly off on a binding site means millions of dollars flushed in downstream clinical trials.

We think Meta's move looks shortsighted on pure scientific ambition, but it clearly illustrates how big tech is narrowing its bets to protect near-term margins. In doing so, it left the field wide open for independent startups and VCs to capture what could be AI's most consequential frontier.