U.S. China AI debate moves toward Trump-Xi talks in September
Trump plans to discuss AI with Xi as Chinese open-source models gain users, raising policy questions beyond tariffs and chips.
By Maya Okafor · Markets Writer
· 4 min read
The U.S. China AI debate is moving from chip controls and tariffs into diplomacy, as President Donald Trump says he plans to raise artificial intelligence with Chinese President Xi Jinping in September. For investors, the fight matters because AI spending, semiconductor restrictions and model access are now tied to both tech-company growth and U.S.-China policy risk.
CNBC reported that Beijing is increasingly using its AI capabilities as a tool of foreign influence while Chinese startups release new models that more American companies are using. Xi gave a speech on AI on July 17, the same day Moonshot released Kimi K3, a model CNBC described as a rival to Anthropic’s Claude Fable 5.
The release intensified discussion in Washington over Chinese open-weight AI models, according to Ryan Fedasiuk, a fellow at the American Enterprise Institute. Open-weight models make key model components available for others to use or adapt, which can speed adoption but also raises security and policy concerns.
Fedasiuk told CNBC that Kimi K3 had set off a sharp debate in Washington over how to respond to Chinese open-weight models. That debate is no longer limited to regulators and engineers.
What will Trump and Xi discuss about AI?
Trump said late last week, while speaking about U.S. data center policy, that he plans to discuss AI with Xi during a U.S. visit in late September, CNBC reported. China’s foreign ministry said Beijing was ready to work with the United States on AI.
Lex Zhao, a Chinese-born venture capitalist and Boston-based managing partner at One Way Ventures, told CNBC that the direction of U.S.-China cooperation or rivalry on AI will be highly consequential. Zhao, whose firm oversees $150 million in assets, said many American startups use Chinese AI models and that many companies in his firm’s portfolio build on top of them.
Zhao said the two countries each have leverage over the other, which could limit how far either side goes in restricting AI links. That commercial overlap turns a tech-policy fight into a diplomatic issue: cutting off access could affect companies already using Chinese models, while leaving access open could sharpen U.S. security concerns.
George Chen, partner and chair of digital practice at The Asia Group, told CNBC after attending China’s World AI Conference in Shanghai and APEC Digital Weeks in Chengdu that Beijing now sees AI as a diplomatic tool for attracting allies and partners. Nearly two dozen senior tech officials attended the Chengdu events, including officials from the U.S., according to CNBC.
Chinese models are improving, but compute is still a bottleneck
China’s AI sector has advanced despite four years of U.S. limits on access to advanced chips used to train AI systems, CNBC reported. Those restrictions still show up in capacity constraints.
AI consultant Shu Weibing told CNBC that during a visit to Z.ai’s offices this month, he found a three-month waitlist for companies seeking access to the startup’s most advanced GLM 5.2 coding tools, which were released in mid-June. CNBC said Z.ai did not respond to a request for comment.
Moonshot also said a few days after launching Kimi K3 that it was pausing new subscriptions because of limits on computing power. Separately, CNBC reported that a person familiar with the matter confirmed Z.ai is expanding a 1-gigawatt data center.
Demand is not only domestic. CNBC reported that U.S.-based Hugging Face said it used Z.ai’s GLM 5.2 to address a cyberattack involving a rogue OpenAI model. Fedasiuk told CNBC the incident showed why open source is necessary for American enterprise and said a middle ground may be possible.
Other countries are watching the policy choices. APEC representatives in Chengdu issued a statement supporting open-source AI models that use strong security assurance, according to CNBC.
Zhao said companies are best served by systems that let them switch among different AI models instead of depending on one. As Chinese models improve and spread, the policy question for Washington is getting harder: how to protect national interests without cutting U.S. companies off from tools they are already using.
This story draws on original reporting from CNBC.
