Boltz, the MIT spinout behind open-source Boltz-1 and Boltz-2, raised a $110M Series A led by Bessemer as six top-20 pharma groups pay for its AI models.
Key Takeaways
- Boltz raised a $110M Series A led by Bessemer Venture Partners, announced November 18, 2026. Valuation was not disclosed.
- Six of the world's 20 largest pharma companies are partners, and molecular token usage is up 50-fold since June.
- The round follows a $28M seed in January 2026, bringing total funding to $138M.
Lead
Boltz, a Cambridge, Massachusetts public benefit corporation that builds AI models for designing proteins and small molecules, announced a $110M Series A on November 18, 2026. Bessemer Venture Partners led the round, with Andrew Hedin and Byron Deeter involved from the firm. The company did not disclose its valuation.
Existing backers Andreessen Horowitz, Amplify Partners, Zetta Venture Partners and Obvious Ventures returned. New investors include Greycroft, F-Prime Capital, Hummingbird and Benchling. Boltz says the money will go toward compute, data, meeting partner demand and hiring.
What Does Boltz Actually Sell?
Boltz sells metered access to molecular design models through Boltz Lab, a platform it launched in January 2026. Customers pay in "molecular tokens," a usage-based pricing scheme introduced on June 16, 2026, alongside two proprietary models: BoltzMol-1 for small-molecule hit discovery and BoltzProt-1 for protein design.
The company says BoltzMol-1 found hits on six of ten targets, testing between 28 and 51 compounds per target. BoltzProt-1, it says, nearly tripled the hit rate of BoltzGen, the earlier open-source protein design tool. These are company-reported results, and the targets were not named in the announcement.
How Did an Open-Source Project Become a Pharma Vendor?
Boltz started as research from Regina Barzilay's lab at MIT. Gabriele Corso (now CEO), Jeremy Wohlwend and Saro Passaro released Boltz-1 in November 2024, an open-source model that matched the accuracy of Google DeepMind's AlphaFold 3 on structure prediction. Boltz-2 followed with binding affinity prediction, a task closer to what drug discovery teams need.
The models drew more than 5 million downloads and over 100,000 scientists. The group incorporated as a public benefit corporation on January 8, 2026, and raised the $28M seed at the same time. It now employs about 40 people.
Free models create adoption but not revenue. The June pivot to proprietary models and token pricing is the step that turned that user base into a business, and the 50-fold rise in token usage between June and November is the figure Bessemer is underwriting. The company did not publish revenue, so the absolute scale of that growth is unclear.
Who Are the Pharma Partners?
Six of the world's top 20 drugmakers work with Boltz, including Pfizer, GSK, Takeda and Bristol Myers Squibb. Pfizer's multi-year collaboration, announced with the seed, has Boltz building exclusive models on Pfizer's data for structure prediction, small-molecule affinity and biologics design.
That arrangement shows where the commercial value sits. Public models are trained on public data. Big pharma holds proprietary assay results and structures that can sharpen a model for its own programs, and a vendor that can train on them privately has something a free download cannot offer.
Why Does This Round Matter for the Sector?
A $110M Series A is large for a 40-person team, and it comes roughly ten months after a $28M seed. The step-up implies investors see usage and partner traction, not only research credentials, as the basis for the price.
The harder question is durability. Boltz competes with well-funded structure and design efforts at large technology groups and other AI drug discovery startups, while its own open-source releases give rivals a free starting point. Pharma buyers can also build in-house. The company's defense is speed of model improvement and the data partnerships, both of which need heavy compute spending.
There is also a gap between model hit rates and drug outcomes. Finding binders in the lab is an early step, and none of the announced results involve a candidate in clinical trials. Revenue from compute-metered tooling can arrive well before any therapeutic result, which suits a platform vendor, but it also caps the upside if tokens become a commodity.
What Comes Next for Boltz?
Expect the next twelve months to center on compute buildup, additional proprietary model releases and more pharma data deals. A mix of open and paid models remains the central tension. If the free models stay close to the paid ones in quality, token revenue will be pressured; if they fall far behind, the scientist community that built Boltz's reputation may drift elsewhere.
Investors will likely watch token volume converted to disclosed revenue, plus whether any partner program built with Boltz models produces a candidate that advances toward the clinic.
Outlook
Boltz has moved from a free research release to a paid platform with six top-20 pharma partners and $138M in total funding. Its 50-fold usage growth since June is the main evidence for the model, though revenue and valuation remain undisclosed. The next test is whether pharma spending on molecular tokens keeps growing as open-source alternatives improve.



