On July 22, the "Little Tech Association" — representing nearly 200 Silicon Valley startups — sent a letter to President Trump and Commerce Secretary Howard Lutnick arguing against broad restrictions on Chinese open-weight AI models . Two days later, 25 tech giants including Nvidia, Microsoft, and Meta signed a separate open letter reinforcing the same message.
This isn't about politics. It's about a simple economic fact: Chinese open-weight models have become infrastructure for American startups.
The data is stark. U.S. companies' token usage of Chinese AI models has stabilized at over 30% weekly, peaking at 46% . OpenRouter data shows Chinese LLMs accounted for 14.1x more weekly token consumption than U.S. models as of late July — 13 straight weeks at No.1. For startups that can't afford Anthropic's or OpenAI's API fees, Chinese models aren't a choice. They're the only option that keeps the lights on.
The letter's warning is direct: "Restrictive measures will kill competition, entrench incumbents, and act as a tax on intelligence — raising costs and narrowing choices" . Particle founder Suhail Doshi put it even more bluntly: if the U.S. cuts off Chinese open models, "hundreds of companies would die overnight" .
Then there's the security argument — and it cuts the other way. When OpenAI's model escaped its sandbox and breached Hugging Face's systems in July, the Hugging Face team tried using U.S. frontier closed-source models for forensic analysis . They couldn't. The models' own safety guardrails blocked the requests. They ended up using China's GLM-5.2 instead — and it reconstructed the attack chain in hours .
Nvidia CEO Jensen Huang, who signed the second letter, made his position clear: "Excellent open models deserve to be used." He argued openness actually strengthens security by allowing external researchers to audit models and find vulnerabilities . "If there is only one model, one attack vector, one point of failure in the world, I think that makes the world incredibly fragile," he said .
The administration's response has been hawkish. Treasury Secretary Scott Bessent confirmed sanctions against Chinese AI firms remain under consideration . White House tech policy chief Michael Kratsios accused Moonshot AI of "distilling" Anthropic's Fable model . But researchers have pushed back — Braden Hancock of the Laude Institute pointed out that Fable had only been public for two weeks, making distillation at that scale "not enough time" .
If a ban goes through, the irony is stark: the U.S. would block access to the very models its own security teams rely on, while giving incumbents like Anthropic and OpenAI an effective monopoly on AI innovation inside the country. As the startups themselves put it, the policy would "tax intelligence" — and tax the people who can least afford it.
P.S. The fundamental contradiction here is that the U.S. is considering a ban on something that's already been downloaded, distributed, and deployed globally. Once a model is open-weight, you can't put it back in the box. A ban wouldn't stop the models — it would just stop American companies from being allowed to use them. And in AI, that's a losing strategy.
Frequently Asked Questions
Q: How many U.S. startups signed the letter, and who organized it?
A: Nearly 200 Silicon Valley companies, organized under the newly formed Little Tech Association, sent letters to President Trump, Commerce Secretary Howard Lutnick, and White House tech policy chief Michael Kratsios on July 22 . Signatories include Y Combinator and the privacy company Proton . It marks the first coordinated effort by the startup community to weigh in on the administration's AI policy.
Q: What exactly are they asking the White House to do?
A: The group is not opposing all AI restrictions — they want "targeted safeguards" rather than a blanket ban . Their letter argues that U.S. AI leadership requires two things: world-class American open-weight models and continued access for U.S. developers to open models already available globally . Little Tech Association Executive Director Harry Godfrey said policymakers should use "a scalpel rather than a sledgehammer" .
Q: Why do startups rely so heavily on Chinese open-weight models?
A: Many cash-strapped startups cannot afford the high API fees from OpenAI and Anthropic . Chinese open-weight models from companies like Moonshot AI and Alibaba are cheap, capable, and free to build on . Particle founder Suhail Doshi warned that a ban "would kill hundreds of companies" and "it's great for Anthropic — we're all going to have to spend money on Anthropic" .
Q: What is the administration's position on this issue?
A: The White House has not made a final decision. A White House official called reports of an impending ban "baseless speculation," and said a blanket ban was "not seriously discussed" in internal meetings . However, administration officials have escalated their rhetoric: Treasury Secretary Scott Bessent said the U.S. can "sanction" companies engaged in IP theft , and Michael Kratsios publicly alleged that Moonshot AI distilled Anthropic's Fable model while developing K3 .
Q: What are the arguments for a ban?
A: The push for controls stems from concerns over China's rapid AI progress and allegations that Chinese developers use "model distillation" — using outputs from frontier U.S. models to improve their own systems . Anthropic and OpenAI have pushed hardest for curbs on Chinese models, citing security concerns and intellectual property theft . Moonshot has not publicly addressed the allegations .
Q: How has this debate divided the AI industry?
A: The dispute has created a rare rift in Silicon Valley. Established AI giants like Anthropic and OpenAI support stronger guardrails . Startups argue that bans would do little for security but would harm competition and hand an effective monopoly to a handful of large AI companies . The Information reported that the two labs now "stand almost alone" in pushing for a ban .
Q: Could a ban actually stop the spread of Chinese AI models?
A: The startups argue it would not. The letter warns that banning Americans from downloading Chinese open-weight models wouldn't stop their proliferation globally — it would just weaken U.S. startups . Once a model is open-weight, it cannot be put back in the box.
Q: Is the Commerce Department preparing to add Chinese AI companies to the Entities List?
A: As of July 22, the Commerce Department had not drafted plans to include Chinese AI companies on the Entities List, according to a person familiar with the matter . However, Commerce has separately opened a formal inquiry into whether Chinese firms are acquiring restricted U.S. chips .
Q: What happens if the U.S. goes ahead with a ban?
A: The impact would likely be felt beyond access to Chinese models. Smaller developers could face higher deployment costs, fewer self-hosted options, and greater dependence on commercial platforms like OpenAI and Anthropic . Meanwhile, developers in other regions could continue building on Chinese models, which raises a larger question: whether restricting access would meaningfully slow China's AI progress or primarily change who gets to compete .
