Boltz Raises $110M Series A for Open-Source Biomolecular AI Models
Bessemer Venture Partners led a $110M Series A in Boltz, the MIT spinout now based in London that builds open source models for biomolecular structure and binding. Existing investors a16z, Amplify, Zetta and Obvious came back, and Greycroft, F-Prime, Hummingbird and Benchling joined. Nvidia invested earlier this year. Boltz raised a $28M seed in early 2026, so the A is roughly four times the seed within the same year. tech.eu: https://tech.eu/2026/11/19/london-based-boltz-raises-110m-series-a/
In plain terms: give Boltz a protein and a molecule, and its models predict the 3D structure of the complex and how tightly they bind, and newer versions generate candidate molecules. When the first Boltz model came out of MIT it matched AlphaFold3 on structure prediction while being fully open, which is why so many academic and startup groups built on it. The company now sells a platform and API on top, and lists Pfizer, GSK, Takeda and BMS as partners.
The bet is the classic open core one, applied to drug discovery. The weights are free; the hosted platform, the affinity tooling and the partnership work are not. AlphaFold3 still comes with usage limits, and Isomorphic just raised $5B to keep its models in house. Boltz is the counter bet: that pharma would rather build on something it can inspect and run itself, and pay for the parts around it.
The money goes to compute, data and hiring; the team is about 40 AI researchers in King's Cross. Benchling on the cap table matters more than it looks: that is the lab notebook a lot of biologists already live in, and the shortest path from a structure prediction to an experiment someone actually runs.
← Back to all articles
In plain terms: give Boltz a protein and a molecule, and its models predict the 3D structure of the complex and how tightly they bind, and newer versions generate candidate molecules. When the first Boltz model came out of MIT it matched AlphaFold3 on structure prediction while being fully open, which is why so many academic and startup groups built on it. The company now sells a platform and API on top, and lists Pfizer, GSK, Takeda and BMS as partners.
The bet is the classic open core one, applied to drug discovery. The weights are free; the hosted platform, the affinity tooling and the partnership work are not. AlphaFold3 still comes with usage limits, and Isomorphic just raised $5B to keep its models in house. Boltz is the counter bet: that pharma would rather build on something it can inspect and run itself, and pay for the parts around it.
The money goes to compute, data and hiring; the team is about 40 AI researchers in King's Cross. Benchling on the cap table matters more than it looks: that is the lab notebook a lot of biologists already live in, and the shortest path from a structure prediction to an experiment someone actually runs.
Comments