On Friday, July 24, 2026, Jensen Huang — the most consequential figure in artificial intelligence hardware, co-founder and CEO of Nvidia, and until that day a conspicuous absence from social media — made his first-ever post on X. He chose not to announce a new chip. He chose not to celebrate a partnership or a quarterly result. He used his platform debut to share a three-page open letter signed by 25 of the most powerful companies in the American technology industry, warning Washington not to repeat the single most damaging mistake it nearly made to the software industry in the 1980s.
The letter is titled “Open Weights and American AI Leadership.” It is a policy document, a competitive manifesto, and a declaration of industrial war — depending on whose side you are reading it from. And the sides it has created say almost as much about the future of artificial intelligence as anything in the letter itself.
What Sparked It: The Kimi K3 Moment
To understand why this letter landed when it did and with the force it did, you have to start on July 16, 2026, when Moonshot AI — a Chinese artificial intelligence laboratory — released Kimi K3.
Kimi K3 is a 2.8-trillion-parameter mixture-of-experts model, and it arrived like a depth charge into the AI policy debate. It debuted around third place on the Artificial Analysis leaderboard, sitting behind only Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 — and ahead of both on some practical coding evaluations. It is, by the most credible independent benchmarks available, one of the most capable AI models anywhere in the world. Full weights — the downloadable files that anyone can run on their own hardware — were promised for July 27.
Chinese open-weight models had already been gaining serious ground. Before Kimi K3 arrived, Chinese open-weight models already accounted for 46.4% of routed token usage on OpenRouter, the popular multi-model API gateway, compared to 35.7% for US models. Kimi K3 accelerated a trend that was already alarming Washington, and it did so at a moment of extraordinary political sensitivity.
The response from the Trump administration was rapid. The White House began weighing at least five restriction tools: Entity List designations for Chinese AI labs, federal procurement bans, security advisories, liability requirements for US companies using Chinese model outputs, and restrictions on the distillation technique that may have been used to build Kimi K3. White House science and technology director Michael Kratsios accused Moonshot of building Kimi K3 by distilling Anthropic’s Fable 5 — using outputs from Anthropic’s closed frontier model as training data for a smaller, more efficient open model, without authorisation. Treasury Secretary Scott Bessent floated sanctions for distillation he likened to theft, and said US model watermarks had been found in Chinese systems. On July 21, Bessent stated: “open source is not open season on American IP.”
Eight days later, Jensen Huang posted.
What the Letter Actually Says
“Open Weights and American AI Leadership” is a three-page policy document that makes six interlocking arguments for why the United States should not restrict open-weight AI models, and why doing so would damage rather than protect American technological leadership.
The first and most historically grounded argument draws a direct parallel to the open-source software movement of the 1980s and 1990s. The letter argues that open-source software created the shared foundation that powers the modern internet, American technology companies, and critical government systems globally — and that this happened despite fears, at the time, that making software freely copyable and modifiable would undermine American competitiveness. The letter argues open-weight AI models can play the same foundational role in the AI era: accessible infrastructure that enables an entire ecosystem to build on top of, rather than a proprietary stack controlled by a handful of companies.
The second argument is economic. Open weights allow startups, universities, hospitals, factories, and small businesses to deploy advanced AI without training models from scratch or paying frontier API prices for every inference task. Restricting them does not protect American AI leadership; it concentrates AI’s benefits in the small number of companies with the compute budgets to build frontier closed models — and those companies are not the majority of the American economy.
The third argument inverts the conventional wisdom on safety. The letter argues that relying solely on closed models carries no built-in safety guarantee, because closed models can be compromised, misused, or fail without outsiders noticing. Openness may be one of the most important paths to AI safety, because it allows the entire research and security community to inspect, test, and improve models in the open. Hugging Face’s own experience during the week the letter was published illustrated this in real time: Hugging Face’s security team attempted to use Anthropic’s Fable 5 to help contain a cyberattack, but the model’s guardrails prevented it from working in a defensive security context. They turned instead to an open Chinese model, Z.ai’s GLM 5.2, and contained the attack quickly.
The fourth argument addresses distillation directly, and with precision. The letter calls distillation a widely used, legitimate, decades-old development method. It explicitly distinguishes between the technique itself and its potential misuse: policymakers should not conflate legitimate model-development techniques with misappropriation, it says, and unlawful extraction deserves targeted legal and commercial frameworks rather than sweeping restrictions on a foundational training technique. This is a deliberate rebuke to Bessent’s framing while stopping short of defending the specific allegations against Moonshot.
The fifth argument is about sovereignty and geopolitical competition. Open models enable countries — including American allies — to build their own AI capabilities without depending on any single nation’s proprietary systems. A world in which American companies export open AI infrastructure is a world in which American technical standards, American safety frameworks, and American values propagate globally. A world in which Washington bans open weights drives that infrastructure development toward less accountable alternatives.
The sixth argument is about what happens if restrictions are imposed. Policymakers have an important opportunity to act — keeping the frontier plural by avoiding premature restrictions on open models that stifle competition or drive innovation overseas, the letter states. “Our AI leadership will be judged not by one frontier AI model, but by whether the United States builds a strong, open ecosystem that diffuses into every sector.”
Who Signed — and Who Didn’t
The 25 signatories of the original letter form a striking coalition that does not usually find itself on the same side of a policy argument. Nvidia, Microsoft, Meta, Dell Technologies, IBM, Palantir, CrowdStrike, ServiceNow, Hugging Face, Mistral AI, Mozilla, the Linux Foundation, Andreessen Horowitz, Y Combinator, and Perplexity are among them. Jensen Huang chose his first-ever social media post to amplify it. Microsoft CEO Satya Nadella reposted the letter the same day. Elon Musk, whose SpaceX did not formally sign, quote-tweeted Huang with a message of full support: “Jensen is right.” Within hours of the post, the thread was measuring in millions of views.
Then there are the absences. OpenAI, Anthropic, and Google — the three companies whose closed frontier models dominate the current AI landscape — did not sign. Their absence is not an oversight. It is a position.
OpenAI and Anthropic have been the most vocal in Washington about the dangers of Chinese open-weight models. They have made security the heart of their case for restriction. Each has, separately, argued to policymakers that strong open-weight Chinese models represent a risk that requires regulatory response. Their commercial interest in this position is not hidden: if Washington bans or restricts Chinese open-weight models, the competitive pressure those models place on closed frontier API pricing is substantially reduced. The closed-model leaders would gain if Kimi K3 and its successors are declared off-limits.
The story moved quickly. OpenAI’s Sam Altman said publicly that he was “glad to see” the support the letter generated — a pivot from his earlier posture that observers read as a recognition that the political momentum had shifted. In a subsequent development, OpenAI joined an expanded coalition of 35 companies backing the letter’s position, leaving Anthropic as the most prominent holdout in the entire industry debate. The letter’s coalition had effectively grown from 25 to 35 signatories in the space of 24 hours, making Anthropic’s absence the dominant political story of the moment.
The Rhetorical Trap: Xi Jinping Moves First
One of the most consequential and underappreciated dimensions of this story is the rhetorical position it creates. A week before Huang posted, Chinese President Xi Jinping made his debut appearance at the World AI Conference in Shanghai on July 17, delivering a keynote that claimed the same “openness” mantle for China. Xi called for “a symphony of global cooperation” rather than “a solo performance by any single country,” urged the world to “jointly oppose overstretching the national security concept in the field of AI” — a direct jab at US export controls — and pledged cooperation with Africa, Latin America, Asia, and BRICS nations to prevent “new historical injustices.”
Xi is now occupying the same pro-openness, anti-restriction rhetorical space that Huang’s letter claims for American industry — just from the other side of the Pacific. The timing creates a genuine problem for both the letter’s coalition and Washington’s ability to frame this as a simple binary between open American values and restrictive Chinese ones. The messaging war over who owns the word “open” in AI is now fully underway, and it is not being fought only in policy corridors.
The 1980s Parallel: Why the Historical Argument Matters
Huang’s choice to anchor the letter in the software history of the 1980s is not rhetorical decoration — it is the most substantive argument in the document. In the early years of the open-source software movement, there were serious voices in government and industry arguing that freely distributable software represented a threat: to commercial software companies, to national security through potential backdoors, to intellectual property norms. Those arguments were made in good faith by serious people. They did not prevail, and the outcome — the open-source foundation on which the entire modern internet, cloud computing, and the American technology industry’s global dominance is built — demonstrates why they should not have.
The letter argues this is that moment again, and that the technology industry and policymakers are being asked to choose, in real time, whether to restrict a foundational technology before understanding its full implications. The bet Huang and his 24 co-signatories are making is that open-weight AI models are to the AI era what open-source software was to the internet era — the shared infrastructure that enables an entire ecosystem and whose restriction would damage the restrictor far more than the restricted.
The counter-argument — made implicitly by OpenAI and Anthropic’s absence and explicitly by Treasury Secretary Bessent — is that this analogy breaks down because open-weight AI models in the hands of a hostile state are not equivalent to open-source software: they represent a direct capability transfer of frontier intelligence to adversaries who will use it without American rules, American safety frameworks, or American accountability. This is not a frivolous argument. The distillation allegations against Moonshot AI, if proven, would mean that a Chinese laboratory used American closed-model outputs to build an open-weight frontier model that is now freely downloadable by anyone in the world, including adversaries of the United States.
Researchers have cast doubt on the specific Kimi K3 distillation allegations, noting that Fable — the Anthropic model Moonshot is accused of copying — was not publicly available long enough for Kimi K3 to have been trained on its outputs. But the underlying policy question the allegations surface — whether distillation can be used to extract proprietary capability from closed models, and whether such extraction can be meaningfully detected or prevented — is entirely real and entirely unresolved.
What Washington Is Actually Weighing
As of July 25, 2026, no executive order or legislation has passed. Open-weight model deployments remain legal. The watch items are specific: Entity List designations for Moonshot AI that would create enterprise procurement risk; any executive action that imposes liability on US companies using Chinese model outputs; and restrictions on distillation as a training technique that could affect a broad range of legitimate AI development.
White House AI adviser David Sacks framed the letter as a rejection of what he called “incessant machinations to kneecap the open model ecosystem” — not a demand that all software be open source. His framing signals where the administration’s current lean sits, even as Bessent’s harder rhetoric on distillation-as-theft points in a different direction. The administration is not speaking with one voice, and the next few weeks will determine which voice wins.
The Stakes: Who This Really Decides Between
Strip away the policy language and the historical analogies, and what this fight is really about becomes clear. Open-weight AI models are winning on cost and capability. Chinese open-weight models already handle nearly half of routed inference usage on major platforms. If Washington restricts them, the primary beneficiaries are not ordinary Americans or American national security in any concrete operational sense. The primary beneficiaries are the handful of closed-model companies — OpenAI, Anthropic, Google — whose pricing power depends on there being no viable open alternative near the frontier. The letter, at its core, is the rest of the industry lining up against that outcome.
Nvidia sells chips to everyone. Open models running on Nvidia GPUs are as profitable for Nvidia as closed models running on Nvidia GPUs. Meta, Mistral, and Hugging Face build open models and have direct commercial interests in their continued legality. Microsoft sells Azure compute and benefits when more models are deployable on its infrastructure. The signatories’ commercial interests align with openness. The absentees’ commercial interests aligned, until very recently, with restriction.
The 1980s software parallel Huang invoked contains one detail the letter chose not to spell out explicitly: the companies that feared open source the most in the 1980s were the companies selling proprietary software licenses. They were not wrong that open source would undermine their pricing power. They were wrong that this was bad for America.
Sources
- Fortune — Jensen Huang just used his first ever X post to warn the AI industry not to make the mistake that software narrowly avoided in the 1980s: https://fortune.com/2026/07/24/jensen-huang-open-source-letter-nvidia-kimi/
- CNBC — Nvidia, Microsoft, Meta warn against ‘premature restrictions’ of open-weight models: https://www.cnbc.com/2026/07/24/nvidia-microsoft-meta-open-weight-ai-models.html
- The Next Web — Nvidia, Meta and 23 others sign a letter defending open weights — OpenAI, Anthropic and Google did not: https://thenextweb.com/news/open-weights-american-ai-leadership-letter-huang-nvidia-openai-absent
- Tom’s Hardware — Nvidia and 24 other companies sign open-weights letter as Washington weighs Chinese AI model ban: https://www.tomshardware.com/tech-industry/artificial-intelligence/nvidia-and-24-other-companies-sign-open-weights-letter-as-washington-weighs-chinese-ai-model-ban
- HotHardware — NVIDIA CEO Jensen Huang Uses X Debut To Push Open-Weight AI Models: https://hothardware.com/news/nvidia-ceo-jensen-huang-x-debut-push-open-weight-ai-models
- OfficeChai — NVIDIA CEO Jensen Huang Backs Open Models In First Post On X As Part Of Letter Signed By 20 Companies: https://officechai.com/ai/nvidia-ceo-jensen-huang-backs-open-models-in-first-post-on-x/
- CoinSprobe — NVIDIA CEO Jensen Huang Makes His X Debut — First Post Defends Open AI Models: https://coinsprobe.com/nvidia-ceo-jensen-huang-makes-his-x-debut-first-post-defends-open-ai-models/
- Cryptopolitan — Nvidia CEO Jensen Huang backs open-model letter in his first X post: https://www.cryptopolitan.com/nvidia-ceo-huang-backs-open-model/
- MetaTalks AI — Nvidia and Meta urge Washington not to restrict open-weight AI as OpenAI and Anthropic stay off: https://www.metatalks.ai/nvidia-and-meta-urge-washington-not-to-restrict-open-weight-ai-as-openai-and-anthropic-stay-off/
- Corenexis Blog — Open Weights and American AI Leadership: Why Nvidia, Microsoft and Meta Signed the Letter — But OpenAI Didn’t: https://blog.corenexis.com/open-weights-american-ai-leadership
- ByteIota — Big Tech Signed an Open-Weight AI Letter. Check the Guest List: https://byteiota.com/open-weight-ai-coalition-letter-openai-absent/
- BigGo Finance — OpenAI Joins 35 Tech Firms in Open-Weight AI Push, Leaving Anthropic Isolated: https://finance.biggo.com/news/3ebfc206-786e-4acd-8a25-bbf34fa8ab10
- Bankwatch — “Open Weights and American AI Leadership” — A Reading: https://bankwatch.ca/2026/07/25/open-weights-and-american-ai-leadership-a-reading-2/
- ExplainX.ai — Open Weights Letter: NVIDIA, Microsoft, Meta — July 24: https://www.explainx.ai/blog/open-weights-american-ai-leadership-letter-july-2026
- Yahoo Finance — Jensen Huang just used his first ever X post to warn the AI industry not to make the mistake that software narrowly avoided in the 1980s: https://finance.yahoo.com/technology/ai/articles/jensen-huang-just-used-first-175154980.html