Spreading Cheap Chinese AI: Are US Containment Strategies Reaching Their Limits?

As China's open-weight artificial intelligence (AI) models increase their competitiveness in the global market, assessments have emerged that traditional protectionist strategies aimed at blocking China's AI technological advancement are failing to work.

The diagnosis is that because Chinese models are gaining popularity within the US private sector, it has become even more difficult to apply regulations without harming American users and businesses.

Bloomberg reported on the 21st local time that China's AI models have become an unavoidable choice for resource-constrained countries due to their low cost and ease of use, and as a result, Chinese AI models are bound to spread anyway, causing the US containment strategy to hit fundamental limits.

The US government has previously blocked their spread within the United States through measures such as imposing high tariffs on Chinese electric vehicles and smartphones or banning the use of cutting-edge components.

However, in the case of open-weight AI models, once they are released, other countries can download the weights, modify them, and use them, making them difficult to block.

According to the AI model marketplace OpenRouter, as of the first week of July, Chinese models accounted for 63% of US companies' token usage on OpenRouter.

This is a sharp increase from less than 10% a year ago.

Among them, Chinese models such as DeepSeek, Qwen, Kimi, and GLM accounted for 46% of the total.

DeepSeek alone accounted for 16.3%, outperforming Google, Anthropic, and OpenAI combined.

Delivery service DoorDash and hospitality company Airbnb are also utilizing Chinese AI to cut costs.

Remarks hinting at sanctions and regulations are emerging from the US administration.

Jamieson Greer, the United States Trade Representative (USTR), appeared on CNBC and said, "We are looking very closely at how China is proliferating its AI technology."

Treasury Secretary Scott Bessent said in an interview with Fox Business that if Chinese models "are found to be misappropriating the fruits of our companies for foreign models, we have the authority to sanction them."

He added that Chinese models have sent "hundreds of millions of calls" to "US models," and that "this is unacceptable, and we will thoroughly investigate this issue within days or weeks."

He also noted, "Another issue could be whether (US) companies using Chinese models should be required to disclose that fact to their customers," adding that "(customers) should not be using counterfeit goods."

This is interpreted as hinting at regulations against US companies using Chinese AI.

However, despite the government's stance, it remains questionable whether a ban on use can be applied down to private enterprises.

Saif Khan, a researcher at the Institute for Progress, said, "If the government tries to ban even the private use of Chinese open-weight models, it will face opposition from the open source community, parts of the AI research community, and Silicon Valley."

Strong backlash is also expected from AI infrastructure companies such as Baseten and Fireworks.

Christie Loh, a researcher at Mats Research, said, "This is very different from when the use of 5G was banned for China's Huawei in 2018. (Banning the use of AI models) is on a completely different level."

Regulatory strategies are also failing to converge into a single approach.

Dean Ball, head of strategy and future at OpenAI and a former AI advisor to President Donald Trump, said, "The government's best strategy is to create tremendous regulatory risk around the use of Chinese open-weight models."

He meant that by instilling fear, uncertainty, and doubt, they can discourage people from using them.

On the other hand, David Sacks, co-chair of the President's Council of Advisors on Science and Technology, countered, "Regulatory decisions must always be justified and based on facts, logic, and evidence, and should not intentionally exploit fear and uncertainty."

The US originally pursued a strategy of supplying lowered-performance Nvidia chips to China so that China would become "addicted" to US advanced technology, but now the situation has flipped where US companies depend on cheap Chinese models, making it so that if the government regulates this, US companies will face hardships.

Researcher Loh predicted, "Containing Chinese AI models will be a very difficult task," adding, "Chinese models will spread anyway, and various countries around the world will continue to use them."
※ Please note: This article was translated by AI and may contain errors.