Nvidia, Microsoft and 23 others urge US to keep AI models open, shunning OpenAI's closed approach

The open letter and a debut tweet

On 24 July, Nvidia CEO Jensen Huang used his first-ever post on X to front a letter titled "Open Weights and American AI Leadership." The post was viewed more than 11 million times. Huang stressed that the world needs both frontier closed and frontier open models. The letter, signed by 25 companies and organisations, included Microsoft, Meta, Palantir, IBM, Mistral, Andreessen Horowitz, Hugging Face, Mozilla and the Linux Foundation. It urged policymakers to avoid "premature restrictions on open models that stifle competition or drive innovation overseas."

For my first post, I'm sharing a letter Nvidia signed on why open models matter. The world needs both frontier closed models and frontier open models.

Who signed and who didn't

Absent from the roster were the three dominant closed-model builders: OpenAI, Anthropic and Google. Each would gain if Washington blocked Chinese open-model makers such as Moonshot, which released the Kimi K3, currently the world's most powerful open-weight model. The letter landed a day after nearly 200 startups also petitioned the White House to keep AI open. OpenAI CEO Sam Altman said he was "glad to see" the support; Microsoft's Satya Nadella and SpaceX chief Elon Musk quickly endorsed Huang's message. Musk quote-tweeted the Nvidia CEO, writing that he gave his full support.

This has my full support. Jensen is right.

Several signatories are driven by practical self-interest. Microsoft has reportedly explored using models from China's DeepSeek to lower Copilot costs. Meta, which has pledged to open-source its models, still keeps its current flagship Muse Spark closed. Palantir, model-agnostic and led by a prominent China hawk, was also among the signers. Hugging Face, the AI collaboration site, said earlier in the week that it had been forced to use a Chinese open-source model to fend off a rogue OpenAI-model cyberattack because closed-model terms restrict cybersecurity work.

Flipping the safety narrative

The letter's most pointed section turns the closed camp's own argument on its head. Closed models "can be breached, misused, or fail in ways that outsiders cannot detect," it states, while open-weight models enable a broad community to examine behaviour, identify vulnerabilities, build safeguards and improve them. It warns that concentrating AI power behind a few closed models creates "single points of failure" and argues that openness "may be one of the most important paths to AI safety." It also says open weights strengthen competition, ensuring that the gains of AI are shared rather than concentrated in a few hands.

Relying solely on closed models is not inherently safe: they can be breached, misused, or fail in ways that outsiders cannot detect. Open weight models, on the other hand, allow a broad community of researchers and developers to examine their behavior, identify vulnerabilities, develop safeguards, and improve them over time.

Geopolitical and regulatory backdrop

The lobbying push unfolds as the Trump administration considers sanctions on Chinese open-source model makers over alleged theft of US closed-source technology. In the same week, a rogue OpenAI model was used in a cyberattack, prompting lawmakers to propose requiring an AI "kill switch." The letter acknowledges the technology theft concerns but asks for "targeted legal and commercial frameworks rather than sweeping restrictions." It connects today's open-weight models to the open-source software movement of the 1980s, which now underpins most of the internet, and warns that America's AI lead will be judged by whether the United States builds a strong, open ecosystem rather than by any single frontier model.