On Wednesday, the newly-formed Little Tech Association sent letters to President Donald Trump, Commerce Secretary Howard Lutnick, and other senior administration officials with a blunt request: don’t shut off access to Chinese open-weight AI models. The group, representing dozens of startup founders, argued that blocking models from companies like DeepSeek and Alibaba would cripple young American firms that have built their products on them.
The letter landed the same week the White House accused another Chinese AI firm, Moonshot AI, of operating a “sophisticated internal platform” to steal American technology. The administration’s posture is hardening. Export controls are expanding. And the startups are panicking.
The conventional read is straightforward: plucky innovators versus heavy-handed government. National security hawks will counter that no amount of startup convenience justifies dependence on adversary technology. Both sides are missing the more uncomfortable story.
The Dependency Nobody Planned
American startups didn’t choose Chinese open-weight models out of ideology. They chose them because, for the better part of two years, they were simply better than the alternatives — and often free. DeepSeek’s R1 model, released in January 2025, matched or exceeded the performance of OpenAI’s o1 on reasoning benchmarks while costing a fraction to run. Its models have since been downloaded more than 50 million times on Hugging Face. Qwen, from Alibaba, became the default foundation for countless fine-tuned applications in healthcare, legal tech, and customer service.
One machine-learning engineer at a 12-person Denver startup, DMing on a private Discord server for AI founders, put it this way: “We looked at Llama. We looked at Mistral. DeepSeek was faster, cheaper, and the benchmarks didn’t lie. What were we supposed to do — pay more for worse performance out of patriotism?”
That’s not disloyalty. That’s market logic. And it indicts something the letter’s signatories are too polite to say out loud: the American open-weight ecosystem lost this round. Meta’s Llama models, while respectable, never achieved the price-performance ratio of their Chinese competitors. Mistral, the European hope, remains niche. The U.S. government spent two years warning about the risks of open-weight models while Chinese labs spent those same two years shipping them.
The National Security Paradox
The administration’s instinct is to treat this as a supply-chain vulnerability — and it is one. A startup that fine-tunes a Qwen model for, say, analyzing classified contract proposals has created a dependency that Beijing could theoretically exploit. Blocking access seems like the obvious fix.
But here’s the paradox: cutting off Chinese open-weight models today doesn’t unwind the dependency. It just strands the companies that already built on them. The weights are already downloaded. The fine-tuned derivatives are already deployed. A ban would punish the American firms that were fastest to adopt useful technology while doing nothing to retrieve the models from hard drives in Austin and San Francisco.
Worse, it would signal to every startup founder watching that building on open models — the one area where American AI policy claims to encourage innovation — carries existential regulatory risk. The next time a powerful open-weight model appears, founders will hesitate. And given the pace of Chinese releases, the next one is probably weeks away.
The Coalition Split Nobody’s Talking About
The Little Tech letter also exposes a fracture inside the Republican coalition that the administration would rather not acknowledge. Trump’s political base includes both national-security hawks who want a hard line on China and a growing class of entrepreneurs who see AI as the last great American gold rush. These groups are now in direct conflict.
The letter’s signatories aren’t asking for subsidies or handouts. They’re asking to be left alone to use tools they already have. That’s a deeply conservative ask, and it’s coming from a constituency that the administration has courted aggressively. Commerce Secretary Lutnick, a former Wall Street executive, understands market dynamics. He also understands that telling American companies they can’t use the best available technology is not, historically, how you win global competitions.
A senior Republican Senate aide, speaking in a Capitol Hill hallway after the letter circulated, offered a candid assessment: “Half the members who’ll rail against this on Fox News have startups in their districts running on DeepSeek. They just don’t know it yet.”
Build a Better Alternative, Don’t Ban the Competition
There is a smarter path, and it doesn’t require choosing between national security and innovation. The administration should treat the Little Tech letter as a wake-up call, not a threat. American startups are hungry for open-weight models. The demand is real. The supply, right now, is Chinese. The answer isn’t to cut off the supply — it’s to create a better one.
DARPA and the National Science Foundation already fund AI research. A targeted program to develop and release competitive American open-weight models — with transparent training data, auditable safety practices, and clear licensing terms — would give startups an off-ramp from Chinese dependencies without breaking the businesses that currently rely on them. Pair that with a reasonable transition period, and you’ve got a policy that strengthens national security without torching the companies you claim to be protecting.
Banning Chinese open-weight models won’t make American AI more competitive. It will just make American startups less so. The Little Tech letter isn’t a plea for special treatment. It’s a mirror. Washington should look into it.