Jensen Huang used his first X post on July 24 to back open-weight AI, but the real story is the split it exposed: Nvidia, Microsoft and Meta want Washington to protect open models while OpenAI and Anthropic want tighter guardrails around the most powerful systems.
Jensen Huang has been running Nvidia for more than three decades. He had never posted on X. Then on Friday, July 24, he used his first message not to announce a chip or an earnings milestone, but to share an industry letter defending open-weight AI. His point was blunt: open models strengthen safety, cybersecurity, innovation and national sovereignty. That was the debut.
As Business Insider reported, the letter, titled Open Weights and American AI Leadership, was signed by 25 organizations, including Nvidia, Microsoft, Meta, Palantir, IBM, Dell, Mistral, Hugging Face, Andreessen Horowitz, Y Combinator, CrowdStrike, Mozilla and the Linux Foundation. Satya Nadella posted the same argument on X the same morning. The letter tells policymakers that broad restrictions on model weights would weaken American AI leadership instead of protecting it.
You should care about the phrase open weights because it isn't policy decoration. It describes whether a model's trained parameters can be downloaded, modified and run outside the original developer's servers. Meta's Llama models made that approach mainstream in the United States. Mistral helped prove that serious open-weight models could come from outside Silicon Valley too. If you're building a product that fine-tunes a model for customer support, code generation or internal search, this fight decides whether your infrastructure remains something you can control or something you rent by the query.
The Missing Signatures
The letter's signatories matter. The absences matter too. OpenAI, Anthropic, Google and Amazon were not on the list, according to multiple reports on the letter. But the original draft overstated the silence around OpenAI. Sam Altman did respond publicly on X, saying he wanted the United States to win in both open source and proprietary models. That's different from signing the letter. It's still not silence.
Anthropic's position is easier to read because Dario Amodei has already made the safety case in public. In prior testimony and public remarks, he has warned that once powerful model weights are released, the developer loses the ability to monitor usage, revoke access or reliably update safeguards. That's a real concern. You don't have to dismiss it to notice the commercial pressure sitting beside it.
Anthropic confidentially submitted a draft S-1 to the SEC on June 1, according to the company's own announcement. Reuters reported the same filing and noted that Anthropic had recently raised money at a $965 billion post-money valuation. The IPO picture for OpenAI is murkier. OpenAI is also being watched as a likely public-market candidate, but the timing is less settled than the original article claimed. Morningstar reported on July 17 that OpenAI executives appeared to be pointing toward 2027, while reports still point to possible trillion-dollar public valuations for both frontier labs. So don't write this as two companies days away from IPO paperwork. One has filed. The other is still a moving target.
Frankly, the incentive is not subtle. If Washington makes it harder for companies to deploy Llama, Mistral, Qwen or DeepSeek-style open-weight models, closed-model APIs look safer to procurement teams by default. OpenAI and Anthropic can argue that from a safety posture. They can also benefit from it. Both things can be true.
What A Restriction Would Mean
If you're building on open-weight infrastructure, this is not an abstract Washington quarrel. A broad restriction could create compliance risk for companies running local or fine-tuned models, export headaches for international teams, and new licensing pressure on hosting providers that nobody has thought through yet. The letter's signatories argue that illegal distillation or intellectual property theft should be handled through targeted legal and commercial remedies, not by treating open-weight AI itself as the threat. That is the stronger argument.
The timing explains why the issue has become urgent. Axios reported this week that the Trump administration has been weighing how to respond to Chinese open-weight models, including Moonshot AI's Kimi K3. Axios also reported that OpenAI and Anthropic have been warning Washington about the risks posed by powerful Chinese open models. That puts them opposite the companies signing Friday's letter, even if everyone claims to support American AI leadership.
The open camp is not pure either. Nvidia sells more chips when more people train and run their own models. Meta benefits when Llama keeps pressure on closed competitors. Y Combinator has every reason to protect cheap infrastructure for its startups. Self-interest is normal in technology policy. The question is whether the policy still works for everyone else.
On that test, a broad clampdown fails. China's DeepSeek, Qwen and Kimi models don't disappear because US firms stop releasing weights. Developers will route around blocked tools, foreign labs will keep publishing, and American startups will be left asking why their own government made the cheaper path harder to use. The better answer is narrow enforcement against theft, clear rules for government use and open American models strong enough to actually compete.
What Huang's first X post confirmed is this: the fight over AI architecture is now fully public, and the sides have stopped pretending otherwise. One camp wants open weights treated as national infrastructure. The other wants the most capable models kept behind controlled doors - and it happens to profit from that arrangement. If you're building with AI, this isn't a philosophical split. It's your cost structure, your bargaining power with the largest labs in the world, and ultimately who sets the terms of your deployment.
Also read: HCLTech bets $1.48 billion on AI infrastructure in Odisha with Sarvam as its model partner; Anthropic upgrades Claude voice mode to run Opus and Sonnet while automating Gmail, Calendar, and Slack; Midjourney buys Co-Star to seed a consumer app empire it never needed to build before
