Photo by Michael Schofield on Unsplash
- Nearly 200 US startups signed an open letter opposing any Trump administration effort to ban or restrict Chinese open-weight AI models.
- The models at the center of the fight are DeepSeek and Alibaba's Qwen family, both free to download and widely used by US developers for fine-tuning.
- Signatories argue a ban would hurt American startups far more than it would slow Chinese AI progress, since the model weights are already public worldwide.
- The letter responds directly to Washington proposals to restrict Chinese-origin AI models on national-security or competitive grounds.
The Evidence
A small AI startup can spin up a working prototype this weekend using a large language model that costs nothing to download, was trained by a Chinese lab, and holds its own against plenty of paid alternatives. That tradeoff sits at the center of a letter that landed in Washington this week. As of July 24, 2026, according to The Standard (Hong Kong), nearly 200 US startups have signed an open letter urging the Trump administration not to ban or restrict access to Chinese open-source, or open-weight, AI models.
The letter doesn't name every restriction under discussion, but the framing is clear: it's a direct response to proposals circulating in Washington that would prohibit or limit the use of Chinese-origin AI models on national-security or competition grounds. The two names that keep coming up are DeepSeek and Alibaba's Qwen family — open-weight releases that US developers have adopted for cost-effective fine-tuning and deployment because, unlike frontier models from OpenAI or Anthropic, they're free to download and modify. DeepSeek's earlier low-cost model releases already rattled tech and equity markets once, intensifying policy scrutiny of Chinese AI well before this letter existed. That history is exactly why nearly 200 founders felt the need to get ahead of the next round of scrutiny rather than wait for it to become policy.
What It Means
The core argument in the letter is a supply-chain one, and it's worth taking seriously on its own terms: once model weights are published, they're mirrored, forked, and redistributed globally within days. A US ban doesn't put the genie back in the bottle — it just means American startups lose legal, supported access to tools their overseas competitors keep using freely. The second-order effect is that restricting Chinese open-weight models doesn't slow China's AI ecosystem so much as it slows the American developers who built products on top of it. That's the moat-compression argument in miniature: when a free, capable substitute exists in the wild, the moat around paid US models compresses, and a policy ban only removes the substitute for the side that follows the law.
Over the next 6 to 18 months, expect this to stay a live fight rather than resolve cleanly. Washington has periodically debated export controls and usage restrictions on Chinese AI and chips without settling on a single durable framework, and a letter from nearly 200 startups — however unified — is unlikely to end the debate outright. More likely: a patchwork outcome, where full bans lose momentum in favor of narrower restrictions (government use, defense-adjacent applications, or sensitive-data handling) while general commercial use of DeepSeek- and Qwen-based tools continues. For anyone tracking this as part of a broader investment portfolio built around AI exposure, the signal to watch isn't whether a ban happens — it's whether the restriction gets scoped narrowly (manageable for startups) or broadly (a real cost shock for anyone using AI investing tools built on open-weight infrastructure).
How to Act on This
The winners and losers here are fairly specific. US startups and independent developers building on free Chinese weights are the clearest beneficiaries of the status quo — a ban raises their costs and pushes them toward paid US alternatives they were trying to avoid. Frontier US labs like OpenAI and Anthropic sit in a more complicated spot: a ban on Chinese open-weight models would reduce free competition for their paid products, but a narrow or symbolic restriction changes little for them either way. Policymakers pushing for restrictions on national-security grounds face the weakest technical argument, since the letter's central claim — that the weights are already global and public — is difficult to dispute.
Watch whether any resulting policy targets government and defense use specifically, versus a blanket restriction on commercial deployment — the two have very different implications for startups and for stock market today sentiment around AI infrastructure names.
Startups relying heavily on DeepSeek, Qwen, or any single open-weight model should map out a fallback plan now, before any restriction forces a rushed migration.
Investors weighing AI exposure should treat this as a regulatory-timeline question, not a technology-quality question — the models aren't going away technically, only their legal accessibility is in question.
On balance, the letter's technical argument is stronger than its political odds of fully succeeding — a full ban looks less likely than a narrower, defense-focused carve-out, but startups betting their roadmap on unrestricted access to Chinese open-weight models are taking on real policy risk regardless of how the current debate resolves.
Frequently Asked Questions
Is it currently legal for US companies to use DeepSeek or Qwen AI models?
As of July 24, 2026, no federal ban is in place; the letter from nearly 200 US startups is specifically aimed at heading off proposed restrictions, according to The Standard (Hong Kong).
Why do US startups want to keep using Chinese open-source AI models like DeepSeek?
Because the models are free to download and fine-tune, letting startups build products cheaply without paying for access to comparable proprietary US models.
Could the Trump administration still ban Chinese AI models despite the startup letter?
It's possible — the letter is a response to proposals already under discussion in Washington, and the outcome will likely hinge on whether restrictions get scoped narrowly to national-security uses or applied broadly to commercial use.
Disclaimer: This article is for informational purposes only and does not constitute financial advice. Research based on publicly available sources current as of July 24, 2026.