China weighs pulling AI models as US reliance comes into focus
Chinese government officials and the country’s top artificial intelligence companies are considering following America’s lead and pulling their top AI models from foreign nations. Closing that spigot would be a major disruption for some U.S.-based companies that have turned to them in the face of sky-high American AI costs.
The talks, first reported by Reuters on Tuesday, have taken place over the past month. The outlet spoke to people familiar with the discussion, who said the restriction could impact China’s most advanced AI models and those the country has yet to release.
This step follows previous U.S. actions to restrict foreign nationals from accessing its leading AI models. Much like the U.S., China is now treating its bleeding-edge AI technology as a national security risk requiring government control.
But as U.S.-made AI costs continue to increase, many U.S. businesses have increasingly relied on Chinese AI models. Restricting access could put those companies in a tough spot.
How would China pull AI models?
Reuters reported that the companies at the talks included major Chinese tech firms, like Alibaba, ByteDance and AI startup Z.ai.
During the meeting, China’s Ministry of Commerce and officials with the companies discussed putting limits on both closed-source and open-source AI models.
Officials discussed making any leak or theft of Chinese AI technology an offense under the country’s national security law, according to Reuters. Additionally, they discussed the possibility of introducing new regulations to limit who can provide funding for domestic AI startups.
The group is still negotiating how wide-ranging the restrictions could be, the sources told Reuters. The sources didn’t say when or if the restrictions would come into effect.
The publication was also unable to learn how overseas restrictions might work, but said previous meetings among Chinese officials have hinted at how the country may proceed.
According to a roundtable discussion about Chinese open-source AI regulations in May, the group discussed a tiered system. The system would require basic, open-source models to submit a simple filing, while more advanced AI systems would face security reviews. But the most sensitive, cutting-edge AI would be barred from public release or restricted to domestic use.
While the U.S. remains the top player in the AI industry, China has quickly caught up with many major U.S. AI companies, as Straight Arrow previously reported. But one area where China beats even the best U.S. AI models is cost.
How dependent are US companies on Chinese AI?
As companies try to capitalize on AI, rising costs have been a major concern. The Wall Street Journal reported in late May that companies have seen costs for AI token usage rise. Companies are now looking at ways to bring down those costs.
One of those solutions is switching to Chinese models, which are much less expensive and comparable to American alternatives. Chinese models charge 18 cents per million tokens, while the average for comparable American models is $4 per million tokens, according to a Citi note.
U.S. companies’ usage of Chinese AI models is increasing, according to data from Openrouter, a platform that allows developers to use a wide range of AI programs. According to CNBC, the data indicates that since Feb. 8, weekly usage of Chinese AI models has consistently remained above 30%, peaking at 46%. This marks a large rise from the previous year’s 11% average, which had plunged to only 4.5% throughout the first half of 2025.
One company that has gone all-in on Chinese tech is AI startup Lindy. In June, the company announced it moved all of its traffic from Anthropic’s Claude to DeepSeek. Company CEO Flo Crivello told CNBC the switch would save Lindy millions of dollars within a few months.
What happens next?
China hasn’t finalized a plan for restricting its cutting-edge AI models. Even if it did restrict its models, the impact may not be immediate.
Chinese AI companies often use open-source platforms for much of their AI software. These open-source models make different parts of the software available for developers to inspect and use. Many models created by American companies, by contrast, are closed and remain proprietary.
Once a model’s weights — the underlying files that make it run — are published online, developers can’t recall or add new safeguards. That means any restrictions China imposes would likely only apply to models the country hasn’t released yet.
That may be part of why some analysts don’t expect a major shift in how U.S. companies use AI, even if China moves forward.
“If an enterprise can deploy a capable open-weight Chinese model on its own infrastructure at low cost, the business case for paying premium prices to U.S. providers weakens considerably,” Nick Patience, AI lead at research firm The Futurum Group, told CNBC in February.
Round out your reading
- Scientists are now eyeing a possible ‘Mega El Niño’.
- Critics mocked Mamdani’s AC request. Republicans made the same ask.
- Social Security was supposed to be a safety net. To young Americans, it’s a broken promise.
- What happens when the water dries up? Much of the American West is close to finding out.
- Political insiders are targeting the two-party system’s grip on America.
Ella Rae Greene, Editor In Chief
