Factbox-AI rivalry hangs over Trump-Xi talks

WASHINGTON, Sept 16 (Reuters) – The race for AI supremacy is set to hang over next week’s meeting between U.S. President Donald Trump and Chinese President Xi Jinping in Washington.

Washington and Beijing view advances in AI as critical to their economic and military competitiveness, but they remain at odds over access to advanced chips, allegations of technology copying and regulation of the AI industry.

U.S. Treasury Secretary Scott Bessent said he will meet Chinese Vice Premier He Lifeng this weekend to discuss AI-related issues before the leaders’ summit. Here are some of the key talking points and disputes.

MILITARY ATTACKS

An AI system that interfered with a nuclear command network or launched a military cyber operation could leave either government with minutes to decide whether the other government had attacked.

U.S. and Chinese security experts are urging both governments to prepare for such scenarios. They have proposed red lines around nuclear systems, human control over consequential cyberattacks, and a hotline for incidents involving autonomous AI.

Open questions include how to stop an accident from being mistaken for an attack, which military decisions machines should never make and how nuclear-armed rivals communicate when technology behaves unexpectedly. 

COPYING U.S. TECHNOLOGY

Copying U.S. technology through a technique known as “distillation” has become a flashpoint in the battle over AI model development.

Distillation is the process of training smaller AI models using output from larger, more expensive ones to lower the costs of training a new AI tool.

The U.S. government accused six Chinese companies in September of maliciously copying American AI products using distillation, with law enforcement officials saying in a statement that the firms did so “likely with Chinese government awareness.”

The U.S. ⁠made a similar accusation in April ahead of Trump’s visit to Beijing. The Chinese embassy in Washington called the allegations “baseless.”         CHIP EXPORTS TO CHINA

Chinese access to key U.S. technologies underpinning the artificial intelligence boom is a point of friction between the two countries.

Since 2022, the U.S. has imposed sweeping restrictions on shipments of advanced chips and chipmaking equipment to China.

There is evidence of advanced AI chips being smuggled into the country, though the scale is unknown.

The U.S. has been considering restrictions on the ability of Chinese companies to remotely access data centers in other countries to prevent those firms from benefiting from Nvidia’s advanced chips.

There have been some exceptions to this effort. Late last year, Trump decided to allow sales of Nvidia’s H200 AI chips. H200s are not the most sophisticated chips available, but their sale prompted concerns from China hawks in Washington who argued the chips could bolster China’s AI capabilities.        REGULATION

The two leaders have fundamentally different perspectives on how involved they should be in the AI sector.

Trump has brushed aside calls for new federal AI rules, saying on Monday that the U.S. only needs a strong president, which the country has “in spades.” He said calls from AI executives for regulation of the technology are a “hoax.”

The Chinese government has rules governing AI algorithms and the data used to train them, including requiring content generated by AI to be clearly labeled.

US VS. CHINA AI DEVELOPMENT RACE

The estimates U.S. officials typically give put Chinese AI models just a few months behind their American counterparts, although industry leaders increasingly describe them as neck and neck.

In 2025, then-White House AI adviser David Sacks said China was likely three to six months behind the United States on AI development. Earlier this year, the U.S. Commerce Department put DeepSeek V4 Pro about eight months behind frontier models.

Many in the industry say the gap is shrinking quickly. Last month, Hugging Face Chief Executive Clement Delangue said China was winning the race as far as open-weight models were concerned — and likely was not far off in terms of frontier models either. Frontier models are the most advanced AI tools and open-weight models have publicly accessible core elements that can be downloaded, inspected and modified.

“I wouldn’t be surprised if they start dominating at the frontier either by the end of this year or next year at the rate of progress,” Delangue told broadcaster CNBC on August 3.

(Reporting by Courtney Rozen and Raphael Satter in Washington, Karen Freifeld in New York, Eduardo Baptista in Beijing; Editing by Rod Nickel)