The Little Tech Association, a group of nearly 200 Silicon Valley companies, recently wrote to the White House and government officials like Commerce Secretary Howard Lutnick, urging them not to restrict access to Chinese open-weight AI models. 

Days later, major tech companies including Nvidia, Microsoft and Meta sent their own letters, warning that premature regulation of open-source AI models would stifle competition and push innovation overseas.

Why U.S. companies are worried

The letters came as some U.S. lawmakers and national security officials were pushing for tighter controls. According to Axios, the U.S. government explored such restrictions as early as last year. The recent rise of China's Kimi K3 model has reignited the debate.

On July 30, Senator Tom Cotton wrote to Lutnick, demanding any company doing business with the U.S. government should be banned from using Chinese open-source models. Around the same time, DoorDash, the country's largest food delivery platform, faced questions from lawmakers for using a Chinese open-source model.

Sohail Doshi, founder of AI startup Particle, told international media that restricting access to Chinese open-source models would cause "hundreds of U.S. companies to go bankrupt instantly."

That is not an overstatement. Chinese open-source models are gaining ground quickly among U.S. businesses. According to OpenRouter, an AI model aggregation platform, U.S. companies using the platform spent more than 30 percent of their weekly usage on Chinese open-source models from February to July this year, peaking at 46 percent in a single week. OpenRouter data analyst Justin Summerville said Chinese models are 60 percent to 80 percent cheaper than leading U.S. models — a figure that may be a conservative estimate.

On July 31, DeepSeek-V4-Flash launched its public beta. According to Artificial Analysis, a U.S. testing platform, the cost per task for this model is less than one percent of Anthropic's flagship model, Fable 5.

A security tool, not just a cheap option

Chinese open-source models have also proven useful in cyber defense. Recently, OpenAI's closed-source model had a loss-of-control incident that breached Hugging Face, a major open-source AI community. The community eventually fought off the attack with help from the Chinese open-weight model GLM-5.2.

In sharp contrast, U.S. AI companies like OpenAI and Anthropic have restricted users from using their closed models for cyber defense. As a programmer who goes by the handle "Benniji" on social media put it: If the U.S. blocks Chinese open-source models, it removes one of the most effective tools currently available to defend against AI-driven cyberattacks.

Protectionism disguised as security

The push to restrict Chinese open-source models appears to have two goals: slowing down China's AI development and shielding U.S. AI firms from price competition. With expensive models from OpenAI and Anthropic struggling to compete on cost, some U.S. policymakers are turning to non-market measures — a pattern previously seen in 5G, drones and robotics.

But protectionism often backfires. Competition from Chinese open-source models has already pushed U.S. closed-source providers to cut prices, making AI more affordable across the U.S. economy. On July 30, OpenAI announced price cuts on two models, with one reduced by as much as 80 percent — a move widely seen as a response to competition from Chinese open-source alternatives.

If access to Chinese models is blocked, U.S. AI giants would face less competition, and their incentive to keep improving cost-performance could weaken. That would ultimately hurt U.S. businesses and consumers. As Nvidia and its co-signers wrote in their letter: "Open weights also strengthen competition and competition is what keeps the gains of AI broadly shared rather than concentrated in a few hands."

The U.S. government does need to regulate large AI models, but the focus should be on its own companies' closed-source models to prevent further loss-of-control incidents and cyber risks. As for restricting Chinese open-source models, Washington should listen to industry experts and avoid another self-defeating move driven by an overstretched idea of national security.