NVIDIA’s Jensen Huang: Open Weights and US Leadership in Artificial Intelligence

Source: NVIDIA Translation: Shan Eoba, Golden Finance

In the 1980s, early open-source software pioneers challenged a prevailing view—that software can only progress when enterprises strictly control the code. This movement helped establish a transparent ecosystem in which developers around the world could study, modify, and improve the software. Today, software developed by the open-source community supports much of the internet, and it is also the foundation for the systems used by the world’s largest technology companies. It also serves as the underlying support for scientific research, cybersecurity, and other critical missions carried out by the U.S. military and federal agencies. Open source not only reduces software costs; it creates a shared foundation of knowledge on which generations of American engineers and entrepreneurs have built their institutional sovereignty.

The United States now faces a similar choice in the field of artificial intelligence. Our AI leadership will not be judged by any single frontier AI model; it will depend on whether the United States builds a strong, open ecosystem and embeds it into every industry. This is crucial for creating opportunities for innovation and prosperity across the nation. It requires expanding AI accessibility, encouraging competition, building a robust application layer, and giving Americans greater control over the technology they rely on. Open-weight models—that is, AI models that anyone can download, review, modify, and run on their own infrastructure—are an important part of this foundation, because they make advanced AI more accessible, adaptable, and widely available.

Open-weight models broaden participation in the AI economy

Startups, established companies, universities, and public institutions can build applications on top of advanced models without having to train a model from scratch, and without paying the steep prices of frontier models for every task. Open-weight models enable every organization to match the right model to the right task at an appropriate cost—keeping frontier-level computing power for truly frontier problems, while running efficient, dedicated models everywhere else. It is this kind of discipline that will keep AI economically sustainable even as it scales to billions of everyday tasks. The way the United States wins in the AI era is by embedding AI into the workflows of factories, hospitals, farms, classrooms, and street storefronts.

Open-weight models strengthen market competition, delivering broad economic dividends

Open-weight models also strengthen competition—because competition is what ensures AI dividends are widely shared, rather than concentrated in the hands of a few. By allowing many organizations to build, adapt, and deploy advanced models, open-weight models create competition not only among model developers, but also across clouds, chips, applications, and services. This competition sparks innovation, lowers costs, and distributes the gains from AI throughout the entire economy.

Open-weight models also give customers greater control

As organizations invest more in AI, they want to ensure they will not be locked into a single vendor, and that they will not lose the knowledge and capabilities they accumulate over time. Open-weight models provide this assurance by allowing organizations to control their own data, evaluate and adapt models to meet their needs, and deploy models wherever their business requires. When organizations create value with AI, open-weight models allow them to own that value through self-improving models, dedicated capabilities, and accumulated knowledge—thereby driving U.S. sovereignty and prosperity.

To be sure, open-weight models do carry real and unique risks. Once the weights are released, they fall outside the control of the original developers, and modified versions are difficult to trace or reverse. But the right way to address this risk is not to ban open-weight models. In a world where network attackers use advanced AI, defenders need models with comparable capabilities so they can detect, simulate, and respond to emerging threats. Open models broaden defensive capabilities, increase transparency, and allow vulnerabilities to be discovered and fixed by many teams.

In fact, openness may be one of the most important paths toward AI safety and security. Relying solely on closed models is not inherently secure: they may be compromised, misused, or fail in ways that outsiders cannot detect. Concentrating advanced AI capabilities behind a small number of closed models amplifies this risk—it creates a small number of single points of failure, weakens competition, and places critical technologies in the hands of a few providers. Open-weight models are the opposite: they allow a broad community of researchers and developers to review model behavior, identify vulnerabilities, develop protective measures, and improve models over time. As open-source software has demonstrated—transparency can be safer than closure—AI security may ultimately depend on empowering more people to test and harden the models that society relies on. It makes rigorous benchmarking and evaluation possible, makes red-team testing possible, and makes protective measures against real and verifiable harms possible, rather than assuming that closed systems are safer by default.

A strong AI ecosystem is not something to be taken for granted. Policymakers have an important window for action. This includes: expanding channels that give startups and researchers access to compute power; investing in shared training assets (datasets, tools, evaluation frameworks); and maintaining diversity at the frontier by avoiding overly early restrictions on open-weight models—restrictions that would stifle competition or drive innovation overseas. These measures must also focus on how a robust application layer can expand the sovereignty-preserving use of AI across the entire economy.

In shaping this ecosystem, policymakers should be careful not to confuse legitimate model development techniques with improper appropriation. Distillation—using the output of one model to help train or improve another model—is a widely used technique for improving, evaluating, and validating models. It continues a long tradition of learning from existing technology, building on it, and improving it—one that has driven innovation since the rise of the open-source software movement. By contrast, the act of illegally extracting value from closed models does raise reasonable concerns. These concerns should be addressed through targeted legal and commercial frameworks, rather than imposing blanket restrictions on technologies that play important roles in AI innovation.

The AI era can be a prosperous era. With the right choices, open-weight AI can expand opportunities, strengthen competition, extend U.S. technological leadership, reduce risk, and ensure that the benefits of this extraordinary technology are broadly shared across the economy. This future is worth building, and the United States should lead in building it.

Co-signing institutions

U.S. Innovators Alliance, Andreessen Horowitz Venture Capital, Arcee AI, Arena, Black Forest Laboratory, Box, Claudestreak, Dell Technologies, Emergence Capital, Hugging Face, IBM, Linux Foundation, Mariana Mining, Meta, Microsoft, Mistral, Mozilla, NVIDIA, Palantir, Perplexity AI, Reflection, Replit, ServiceNow, Telnyx, Y Combinator Startup Accelerator.