Today, twenty-five of the most important organizations in American technology—NVIDIA, Microsoft, Meta, IBM, Hugging Face, Andreessen Horowitz, Y Combinator, and others—published “Open Weights and American AI Leadership,” making the case that America’s AI future depends on a strong, open ecosystem. They are right. We, the undersigned startups, add the voice that letter could not carry: the voice of the thousands of small companies that build on open weights every single day, and of the companies that will be founded on them tomorrow.
Every generation of defining American technology companies began as a startup building on an open foundation. The open-source movement of the 1980s and 1990s did not merely lower the cost of software; it made it possible for a few people with conviction and a laptop to build companies that now anchor the American economy. Google was built on Linux. The cloud was built on open source. The pattern is not an accident—it is the mechanism by which America renews its technological leadership. The next OpenAI and the next Anthropic are being founded right now, in apartments and dorm rooms and small offices, and whether they can exist at all depends on whether the foundation beneath them remains open.
For startups, open weights are not an abstraction. We cannot train frontier models from scratch, and we should not have to. Open-weight models let a two-person company download a state-of-the-art model, inspect it, specialize it for one job, and run it on infrastructure it controls—at a cost it can actually afford. They let us match the right model to the right task at the right price, reserve frontier capability for frontier problems, and compete on the strength of our ideas rather than the size of our compute budgets. Open weights are how the AI economy stays open to entry.
Open weights are also how startups keep the promises we make about privacy. Startups are trusted with the most sensitive data individuals and businesses have—their email, their documents, their memory, their customers. With open-weight models, that trust can rest on architecture instead of assurances: data can be processed on infrastructure the customer controls, and it never has to leave. No lock-in, no forced dependence on a single provider, no surrendering the knowledge and value our users build over time. What a person or a business creates with AI should belong to them. Open weights make that promise enforceable.
To policymakers, we echo and extend the call made today: expand startup and researcher access to compute; invest in shared datasets, tools, and evaluation frameworks; protect legitimate, widely used techniques like distillation; and keep the frontier plural by avoiding premature restrictions on open models that would entrench incumbents and push innovation overseas. Regulation written for a handful of giants must not become a wall around them.
To the companies that signed today’s letter: thank you—and hold the line. Keep releasing open weights. Keep the tools, the interconnects, and the platforms open. The surest sign that your letter meant what it said will be the startups that get built on top of it.
The companies that will define the next decade of American AI have not been founded yet. Keep the foundation open, and they will be founded here.