In January, a computational chemist at one of the twenty largest drugmakers could download Boltz-2, run it on her company's own servers, and pay nothing. By October, the same employer had signed a contract to run Boltz's newer models inside its own computing environment, on its own Nvidia hardware. Same scientist. Same targets. An invoice where there used to be a download.
That migration is the asset Boltz sold this week. On 18 November the London lab announced a $110M Series A led by Bessemer Venture Partners, with a16z, Amplify, Zetta, Obvious, Greycroft, F-Prime, Hummingbird and Benchling. Six of the world's twenty largest pharma companies now pay for the platform. Paying molecular-token volume is up 50x since June.
We would put more weight on the 50x than on the headline. A funding round is a negotiation between investors. Usage is a negotiation with customers, and it is harder to fake.
The gift was the distribution
Boltz did not begin with a sales problem. It began as a set of open models. The founders came out of MIT's CSAIL and Jameel Clinic; their code matched AlphaFold3 on key tasks, collected more than 5M downloads, and reached roughly 100,000 scientists. Pfizer, GSK, Takeda and Bristol Myers Squibb found the models before Boltz found them.
June changed the meter. Boltz shipped its first proprietary models — BoltzMol-1 for small molecules, BoltzProt-1 for proteins — and began selling molecular tokens, the units of information its models process. Same distribution, new toll booth. In June's experiments, BoltzMol-1 found confirmed hits against six of ten challenging targets while testing 28 to 51 compounds per target. Bessemer says a validated hit can now cost as little as $10,000, against hundreds of thousands by conventional routes.
The giveaway was a customer-acquisition strategy with a nine-figure payoff. That is our read, and it is the part of the story this round is really financing.
A timeline that reads like a pricing experiment
CHRONICLE: Boltz, from free models to metered models
────────────────────────────────────
2025 · 🧪 MIT spinout ships open Boltz-1 / Boltz-2
JAN 2026 · 💰 $28M seed; Boltz Lab opens
JUN 2026 · 🔒 First proprietary models: Mol-1, Prot-1
OCT 2026 · 📄 BMS installs Boltz inside its own compute
NOV 8 · 📈 Paying token volume 50x the June baseline
NOV 18 · 💵 $110M Series A led by Bessemer
Sources: Boltz, Bessemer Venture Partners, European Biotechnology, 2026
Read top to bottom, this looks less like a research roadmap and more like a pricing experiment: give the model away, build dependence, then charge for the version that performs better.
Two ways to sell a molecule
Two days before Boltz's round, Isomorphic Labs raised a $5B Series C co-led by Greenoaks and Sequoia. It is the largest venture round in biotech history, past Altos Labs' $3B in 2022, and it landed six months after a $2.1B Series B. Bloomberg reported talks at a valuation near $50B.
Isomorphic is the mirror image. It keeps its best models close and sells partnerships, not compute. President Max Jaderberg, speaking at Bloomberg Tech in London on 2 November, was blunt about the distinction:
We're not selling access to the models, we're not selling tokens or API access.— Max Jaderberg, President, Isomorphic Labs
On 19 November, Merck joined Lilly, Novartis and J&J on Isomorphic's partner list, for a small-molecule oncology target the companies call strategically important. The structure is an upfront payment plus research, development, regulatory and commercial milestones, with tiered royalties. Isomorphic can co-fund development for a higher royalty.
The deal math explains the attraction. McKinsey counted $29.5B in headline AI deal value among the top 20 drugmakers in 2025, with about 2% paid upfront and 60–90% tied to clinical and commercial milestones. Tokens are cash now. Milestones are cash later, if the biology cooperates.
For a pharma buyer, the two models solve different budget problems. A token contract is operating expense, approved by a computational chemistry group and renewed on usage. A milestone partnership is a strategic bet, signed at the corporate level and defended for years. Boltz is selling the first. Isomorphic is selling the second. Both can work, and they fail in different ways.
What $110M buys, and what it does not
Boltz says the money goes to compute, data and hiring. The team is 40 people, up from a small research group a year ago. The company's own framing is infrastructure: molecular intelligence sold by the dose, the way labs already buy sequencing reads and assay reagents.
What the round does not buy is proof. Boltz discloses token growth, not revenue. The 50x is a usage multiple, measured against an undisclosed base, and usage can rise while a customer's willingness to renew stays unproven. For a company selling picks during a gold rush, that is the number investors should press on.
The competitive pressure runs in both directions. Isomorphic can outspend Boltz on compute several times over, and Chai Discovery and a queue of well-funded protein-design labs are chasing the same chemists. Boltz's defense is the installed base: models already inside Pfizer and BMS workflows are hard to rip out, and every new token is a small switching cost. That moat is real but shallow. It depends on the proprietary models staying measurably better than the free ones the company already gave away.
The wider market supports the wave. Drug discovery startups raised a record $23.2B in 2026, according to Dealroom figures cited by CNBC. AI drug discovery funding rose from $4.1B in 2023 to $8.4B in 2025, per McKinsey. Isomorphic's own programs remain preclinical, and Demis Hassabis's first-trial target has already moved once, from the end of 2025 to the end of 2026.
Biotech has a long memory of platforms that looked inevitable at the financing stage. The science is rarely the failure point. The distance between a good molecule and an approved drug is where most of the money and nearly all of the failure sit. A model that ranks molecules well is a genuine tool. A model that predicts which molecule helps a patient is a clinical claim, and those are settled in trials, not term sheets.
Whether Boltz discloses revenue per token, not only token growth.
Whether a new open model family ships, or the open line stays a funnel.
Whether Isomorphic's first human trial starts before the end of 2026.
The models keep improving on benchmarks. Patients remain the only benchmark that settles the argument.