Boltz Pfizer deal puts open-source drug AI on a business test

What happened

Boltz, the MIT-rooted team behind a family of open biomolecular AI models, has become a company. It incorporated as Boltz PBC, a public benefit corporation, and announced three things together: a $28 million seed round, a multi-year collaboration with Pfizer, and a preview of a hosted platform that ships with two initial discovery agents.1 Gabriele Corso is CEO, with co-founders Jeremy Wohlwend and Saro Passaro. The group's technical lineage runs through MIT CSAIL, including work tied to Regina Barzilay's group.1

Amplify Partners, Andreessen Horowitz and Zetta Venture Partners led the round.12 Preqin's investor list also includes Factorial Capital and Obvious.3 The timing is slightly inconsistent across accounts. One review places the incorporation, seed and Pfizer announcement in January 2026.2 Preqin's profile dates the seed to December 30, 2025 in one passage and to January 8, 2026 in another.3 The most likely explanation is a close in late December and a public announcement in early January, but that reading is an inference.

Pfizer is not the only pharmaceutical partner. One account mentions a later deal with Takeda.2

The models behind the company

The company's credibility comes from the models. Boltz began as an open reproduction of AlphaFold 3-style "co-folding," which predicts how proteins and other molecules fit together. Boltz-1 arrived in November 2024 under the MIT license.2 Boltz-2 followed with a June 2025 preprint and was released with Recursion.12 It added binding affinity prediction, an estimate of how strongly a molecule binds its target. Structure prediction alone does not provide that, and it matters a great deal for drug design.2 BoltzGen, described in a November 2025 preprint, extended the work to all-atom de novo binder design.2

The licensing is permissive. Weights, inference code, training code and generated designs are all MIT-licensed and can be used commercially without permission.2 Coverage of the Pfizer partnership presents it as a bet on these open models, Boltz-2 and BoltzGen, for both small-molecule and biologics work.4

Open core, hosted product

This is where the tension sits. Boltz says it wants to build open models and turn them into products that people can use without specialist machine learning or infrastructure skills.1 Boltz Lab, the hosted platform, entered beta in January 2026, and an API came after it.2 Preqin describes revenue as coming mainly from subscriptions and partnerships with pharma companies and research organizations. Its tiered services reportedly include "proprietary models and expert support."3

That phrase deserves attention. The public models are fully open. A business that sells access to proprietary models alongside them follows the familiar open-core pattern, in which free components build reputation and adoption while paid ones carry the revenue. Preqin also says Boltz plans to integrate Pfizer's historical data and deploy AI-driven discovery agents.3 Models fine-tuned on a partner's private data would be very hard to keep fully open, regardless of the company's intentions.

The sources do not say how Boltz will separate the two. The open models may keep pace with the hosted offerings, or the most capable versions may eventually exist only behind the platform. The public benefit corporation structure, and the "guarantees" the company emphasized at launch, read as an attempt to answer that question ahead of time.1 Whether they hold up will depend on choices Boltz makes once revenue pressure grows.

Why it matters

For drug discovery, the main point is that a large pharmaceutical company has publicly backed open-source foundation models instead of building everything internally or licensing closed tools.4 Because the weights are MIT-licensed, Pfizer's competitors can download and run the same base models.2 What Pfizer gets that others do not is presumably the collaboration itself: engineering support, tooling, and models adapted to its own data.

That is the bet Boltz is making. Its argument is that pharma will pay for convenience and customization even when the core technology is free. Many open-source infrastructure companies have tried the same approach, with mixed results. The biotech version has an advantage in proprietary experimental data, which is valuable, hard to obtain, and unlikely to be released openly.

The read

Pfizer and Takeda signing on so early suggests the open-first strategy is working as a way to win customers. The harder test comes later. If the best Boltz models end up trained on partner data and available only through Boltz Lab, the open releases could slowly turn into marketing for the paid product. The PBC charter and public commitments give the community something specific to hold the company to. Users who depend on Boltz's openness should keep comparing what the company publishes with what it sells.