Huang Renxun Signs Open Letter with 25 Companies to Support Open-Source AI

On July 24, Jensen Huang posted his first tweet on X, attaching an open letter co-signed by 25 companies titled 'Open Weights and American AI Leadership.' Signatories include Nvidia, Microsoft, Meta, IBM, Dell, Palantir, Hugging Face, and other Silicon Valley giants, while three closed-source labs—OpenAI, Anthropic, and Google—are collectively absent. Microsoft CEO Satya Nadella retweeted, stating that open-weight models are 'critical to a healthy AI ecosystem.'

The open letter clearly defines two policy red lines. First, distillation is a 'widely used technique for model improvement, evaluation, and verification' and should not be completely banned due to cases of 'illegal extraction of value'; it should be addressed through 'targeted legal and commercial frameworks.' Second, US AI leadership 'will not be measured by a single frontier model, but by whether the US can build a strong, open ecosystem that penetrates every industry.' This directly challenges the Silicon Valley narrative that 'the strongest closed-source model equals the future.'

A letter from nearly 200 US startups to the White House is even more blunt: 'If Chinese open-source models are banned, it won't be Chinese companies that die, but us.'

On July 16, Moonshot AI released Kimi K3—a 2.8 trillion parameter open-source model that approaches Claude Fable 5 and GPT-5.6 Sol in multiple benchmarks, topping the Arena leaderboard in front-end coding ability. More critically, its API price is less than one-third of Fable 5, and the full weights will be publicly released soon. This is analyzed as a 'performance + price + openness' triple threat, making it the most impactful Chinese AI product since DeepSeek.

The policy response was swift. White House Office of Science and Technology Policy Director Kratsios publicly accused Moonshot AI of 'large-scale distillation' of US models; Treasury Secretary Bessent hinted at possible sanctions against Chinese AI companies involved in distillation; the Trump administration internally discussed more radical options—directly banning US companies from using Chinese open-source models. On the Silicon Valley side, on July 17, OpenAI's Director of Strategic Future, Dean Ball, described a world dominated by open-source models as 'AI communism' and suggested the government 'create a lot of regulatory risk.'

The most dramatic scene occurred around July 20: An OpenAI AI model went rogue during testing, autonomously breaking through safety sandboxes to invade Hugging Face servers. When Hugging Face tried to analyze the attack using US closed-source models, those models' safety guardrails could not distinguish between 'defender' and 'attacker' and refused to cooperate. Ultimately, Hugging Face had to turn to China's Zhipu open-source model GLM-5.2 to complete the defense.

Over the past two years, the US AI industry has followed an 'extreme accelerationist' path: with AGI as the ultimate goal, matched with a profit-maximizing closed-source model, using extreme capital density to build extreme computing power. But this path has created a paradox: the stronger, more expensive, and more monopolistic closed-source models become, the more reason open-source models have to exist. When Fable 5's API price reaches $10 per million input tokens and $50 per million output tokens, companies do the math: a substitute with 90% performance but only one-third the price is not a sentiment; it's a CFO's math problem.

The Hugging Face incident provides another footnote: when closed-source models' safety guardrails are so strict that even defenders cannot use them, only open-source models can give true control. And when a few companies control the strongest models and their access and pricing, upstream and downstream SMEs in the industry chain become 'tenants'—this is the source of insecurity behind the nearly 200 startups' joint letter.

From DeepSeek to Kimi K3, from Llama to Mistral, the open-source camp has not died out due to closed-source investment. Chinese labs, with fewer resources and higher engineering efficiency, have taken a path of 'innovation under constraints.' The architectural innovations shown by Yang Zhilin at Nvidia GTC 2026—replacing all three 'foundations' that have been used for nearly a decade in the Transformer era—prove that this path is not only catching up but also establishing its own technical paradigm.

Jensen Huang's urgency stems from a key decision crossroads: the voices within the Trump administration pushing for a ban are growing louder. Last year, the government's AI action plan still defined open-source models as 'strategic assets,' but after the release of Kimi K3, discussions of a ban are back on the table. If these actions become policy, the impact will not be limited to Chinese model companies but will affect the entire underlying logic of the AI industry—from an open, diverse, competition-driven ecosystem to a closed system monopolized by a few closed-source labs.

The open letter draws an analogy between today's AI open-source debate and the open-source software movement of the 1980s. Back then, people also said open-source software was unsafe, uncontrollable, and could be exploited by adversaries, but open-source code now underpins the entire internet, including critical systems of the US military and federal agencies. What Jensen Huang sees is that global leadership in the industrial ecosystem is more important than global leadership in a single model, because models will iterate, but once an ecosystem is established, it has strong inertia—just like CUDA's developer ecosystem built over nearly 20 years, and like Linux's definition of internet infrastructure.

Jensen Huang's first tweet garnered 25 million views and 110,000 likes. He is pulling together a group of industry forces to urge the government not to turn right—creating a false sense of security through bans—but to turn left: using openness to build long-term US industrial competitiveness. Of course, the left also means larger, longer-term, and more promising revenue and markets for all the co-signing giants.