Science at the speed of inference

A manifesto for biotech and open science, from Boltz

Once upon a time, at the very beginning of West Coast Venture Capital, biotech and software were nascent twins, full of promise and similarities. But they quickly grew apart. For several decades they lived parallel universe lives, with their own funding models, VC firms, concepts of “product-market fit”, and growth playbooks.

Now they’re suddenly converging again. We’re seeing new kinds of products and companies get built for biology, and new kinds of founders tackle big problems in AI-native ways, that are bringing our worlds together. The Boltz team is one of those stories. Their transformative open source models for helping drug design and biological research is now used by over 100,000 scientists in academia and industry, and today they’re announcing their next chapter: new research, a $28 million seed round, a partnership with Pfizer, and the launch of Boltz Lab. (this will link to your main page)

The Boltz Manifesto is a thrilling call to action for everyone who wants AI to extend into bio and health to cure disease and explore the future. Here it is, in full:

---

The past and the future of drug development

Rational drug design is the idea that we can engineer medicines by understanding, and then deliberately manipulating, how biological “machines” fit together. Many diseases ultimately come down to molecular interactions: the wrong partners binding, the wrong timing, or the right interaction failing entirely. Medicines are molecular instruments designed to counter the origin of the disease, such as a drug that blocks a bacterial enzyme, a protein binder that neutralizes a toxin, or an antibody that recognizes a viral surface.

Designing drugs is a search problem: finding the right molecule that binds the target selectively and with the right developability properties to actually reach and persist at the site of action. In this search, we are bottlenecked by the scale of physical experiments. If we could reliably and efficiently predict the results of these experiments on a computer, we would cut costs, reduce timelines, and, most importantly, find better drugs. This is the ultimate promise of computer-aided drug design.

For decades, computational modeling in drug discovery has been split into two largely separate camps. On one side, physics-based approaches build generalizable models grounded in a few physical principles, requiring relatively little data to fit. On the other side, data-driven approaches learn local patterns from datasets tied to specific targets. Both approaches have advanced science in the global fight against disease, yet each faces critical limitations: physics-based methods often struggle with accuracy and cost, while data-driven methods typically demand specialized datasets and don’t always generalize. As a result, physical experiments have remained the primary bottleneck across every stage of drug development.

AI started to change everything. In November 2022, the “ChatGPT Moment” showed the world that AI was here, now. But nearly two years earlier, in December 2020, the “AlphaFold Moment” had already convinced most biologists of the same thing. As AI researchers at MIT, this was eye-opening. It demonstrated that generalizable data-driven models of molecular biology are possible, that a seemingly intractable biochemical problem could be tackled with a neural network.

Over the past five years, machine learning for structural biology has flourished, pushing beyond single proteins to model complex biomolecular interactions. We believe these models will continue to improve as they have been over the past few years and reach experimental accuracy on increasingly complex tasks and systems. We believe these models will become part of every chemist’s and biologist’s toolbox, relieving them of the experimental bottleneck. We believe in a future where every scientist can iterate at the speed of inference.

But that future depends on a crucial, increasingly uncertain question: will every scientist have access to these models? Frontier teams like DeepMind/Isomorphic have advanced the field multiple times, but access to their latest systems is now largely restricted. That shift began with AlphaFold3 and has continued more broadly: many groups have kept models closed-source, and placed their capabilities behind walled gardens—available only for internal programs, via expensive co-development deals, or at price points that only the largest companies can afford. So how can scientists outside a small handful of organizations get the tools they need?

Boltz origin story

It became clear to us that widely accessible models would be essential if we were going to realize the full potential of this technology—so we built them ourselves. We launched the Boltz project, and over the past year:

• With Boltz-1 and Boltz-1x, we released the first commercially usable open-source structure prediction model approaching AlphaFold3-level accuracy and introduced new techniques to improve structural validity.

• With Boltz-2, we demonstrated that frontier structure prediction models, combined with large-scale supervision, can push binding affinity prediction to unprecedented accuracy.

• With BoltzGen, we launched a protein design model and agent capable of generating binders to arbitrary targets, validated across diverse modalities, targets, and assays.

Today, 100K+ scientists use our models across every top 20 pharma company and thousands of biotechs to accelerate discovery. By releasing these models publicly under a permissive license, we’ve been able to see how people use them in the wild, learn where they’re strong and where they break, and benefit from validation across more programs than even the best-funded lab could run alone.

We started this effort at MIT, but watching people use our models has made one thing clear: to truly serve scientists at every scale, the field needs more than academic code releases. It needs products, and an organization that can build, maintain, and support them over time. That’s why we founded Boltz as a public-benefit corporation (PBC), committed to advancing cutting-edge research and making it accessible to the scientists building a healthier and more sustainable future.

Boltz PBC

We will pursue our mission through three pillars: Research, Product, and Community.

On the Research front, we aim to understand and reprogram biology and disease states from the bottom up. We are building towards a world where AI models can reach—and even surpass—the accuracy of most experimental assays. We began with modeling and designing biomolecular binding interactions, and we are already moving toward more complex, higher-level functional properties and mechanisms.

Our Product vision is to put the most powerful AI models directly into scientists’ hands, embedded within agentic workflows, experimental interfaces, and, in the future, even autonomous labs. Our goal is to build toward a world where a scientist can go from a therapeutic hypothesis to a human-ready molecule without leaving their computer.

Community will remain a core part of the Boltz ecosystem. We believe in open science and are committed to building the best open-source models in the world; Boltz-1, Boltz-2, and BoltzGen were just the beginning. We will continue sharing our research and insights through papers and blogs, collaborating with academic labs and companies on open validation, and pushing on rigorous testing to understand the limits of what’s possible.

The first step toward this mission has been assembling an exceptional team. We’ve been fortunate to welcome outstanding scientists and engineers from leading academic labs and top companies across pharma, biotech, finance, and tech. We’ve also continued to collaborate with researchers at MIT and beyond, alongside industry partners, and to benefit from a vibrant community that contributes code, feedback, bug reports, and thoughtful discussions in our Slack channel.

Supporting this team and mission also requires significant capital and expertise. We’re proud to announce a $28M Seed round led by Amplify, a16z, and Zetta, alongside angels including Clement Delangue (CEO of Hugging Face) and Factorial Capital. These partners bring not only funding, but also deep expertise and a shared commitment to our vision.

How we make money

To sustain Boltz over the long term—and to keep investing in better research, infrastructure, and products—we need a business model that aligns with the value we want to create.

We plan to make money by building and operating software. Boltz is not a therapeutics company, and we will not develop drugs. Competing with our users would be at odds with our mission: we want to build the best tools for the entire ecosystem, and we believe the most impactful discoveries will come from empowering the experts who already have the domain knowledge, the data, and the brilliant ideas.

Our commercial focus is therefore on products that turn frontier models into reliable, scalable capability: state-of-the-art and customized models, hosted compute, agentic workflows, collaborative interfaces, enterprise deployments, and the support and security features teams need to run these systems in production.

We want these tools to be broadly accessible, so we’re launching with a usage-based model: customers pay for what they run and the value they get—whether that’s predictions, design campaigns, or large-scale screening—rather than being locked into one-size-fits-all contracts. We already offer the lowest price per unit of any major provider in the space, and we will work relentlessly to lower the cost of every unit of compute while increasing what users can accomplish with it.

Looking forward

Today we’re launching Boltz Lab along with our first agents for small-molecule discovery and protein design. The platform is built to remove the practical bottlenecks we see everywhere in adoption: affordable compute at scale, robust infrastructure, and interfaces that fit real scientific workflows. And it comes with clear guarantees: you own what you build, your data stays secure, and we do not train on customer data.

This launch is just the beginning. We’ll keep shipping new agents, models, and interfaces—pushing toward workflows where teams can iterate on hypotheses, designs, and experimental feedback in tight loops, and run campaigns that would otherwise require weeks or months of dedicated compute and engineering.

We’re building toward a world where scientists can iterate at the speed of inference—but that future only works if the tools remain broadly accessible. We’re committed to open science, to continuing to release best-in-class open-source models, and to learning in public about where these systems succeed and where they fail. In parallel, we’re committed to building durable products that teams can trust with their best ideas, deploy in real workflows, and rely on over time.

As we move forward—tackling experimental bottlenecks and building the future we believe needs to exist—we know we can’t do it alone. If you’d like to be part of this journey, we’d love to hear from you.

Gabriele Corso, Jeremy Wohlwend, and Saro Passaro

On behalf of the Boltz core team: Gabriele Corso, Jeremy Wohlwend, Saro Passaro, Francesco Capponi, Demitri Nava, Hannes Stark, Noah Getz, Talip Ucar, Luca Cavalleri, Simone Primarosa, Geoffrey Smith, and Luis Sanmiguel.

---

This newsletter is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. Furthermore, this content is not investment advice, nor is it intended for use by any investors or prospective investors in any a16z funds. This newsletter may link to other websites or contain other information obtained from third-party sources - a16z has not independently verified nor makes any representations about the current or enduring accuracy of such information. If this content includes third-party advertisements, a16z has not reviewed such advertisements and does not endorse any advertising content or related companies contained therein. Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by a16z; visit https://a16z.com/investment-list/ for a full list of investments. Other important information can be found at a16z.com/disclosures. You’re receiving this newsletter since you opted in earlier; if you would like to opt out of future newsletters you may unsubscribe immediately.