Boltz closed a $110M Series A to sell frontier molecular intelligence as a reagent, a16z investors Jorge Conde and Zak Doric said.
I think they are betting that usage beats ownership in drug discovery. Weekly molecular token consumption rose roughly 50-fold in just the last four months after BoltzMol-1 and BoltzProt-1, the firm said.
That is a real signal.
By the time a16z co-led the seed round last year, Boltz open source models had already reached every top 20 pharmaceutical company, the investors said. Commercial models are now deployed at Pfizer, GSK, Takeda, BMS and other leading companies, with founders Gabriele Corso, Jeremy Wohlwend and Saro Passaro adding Boltz Lab for biologists and chemists plus an API that works inside existing workflows and agents such as Claude Code and Codex.
The model is molecular tokens, units of molecular information processed per task. A quick prediction costs little, the firm said, while a full design campaign to generate and refine molecules against multiple objectives can consume far more, much as labs already budget for sequencing reads or assay reagents.
Data still decides quality. Boltz PBC partnered with Excelra to train BoltzMol-1 ADMET models on curated GOSTAR data for solubility, lipophilicity and permeability screening, according to a November 10 announcement. That tells you where near term differentiation comes from, not just architecture but clean pharmacology data that teaches models which compounds to reject.
The fair pushback is that billions have gone to proprietary AI pipelines for a reason. A captive program can keep data, feedback and upside in one place, and frontier labs are chasing the same prize with far larger compute budgets.
That risk does not kill the claim. Proprietary pipelines need one big win, while Boltz can grow if thousands of teams run millions of smaller predictions and screens, and early pharma adoption plus token growth suggest scientists will pay to explore computationally before they commit to synthesis. If you are a founder here, sell compute that earns its next experiment.





