The Global Fever

Policy

Trump Xi AI Cooperation at APEC Meets New US Chip Curbs on China

Trump said the US opposes "globalist regulation" of AI while Xi urged human control, days after Washington tightened chip access for Chinese firms. Here is what was said and agreed.

Wasif Mumtaz

Founder and Editor · November 23, 2026 · 7 min read

Waving flag of China with yellow stars on red fabric
Photo: Magnific

Trump Xi AI cooperation was on the agenda at the APEC leaders' meeting in Shenzhen on November 19, but the two presidents described very different ideas of how AI should be governed. Trump said the United States opposes "any form of globalist regulation" of AI. Xi urged both countries to keep AI "under human control." Days earlier, Washington had issued a rule that restricts Chinese companies' access to AI computing power, which is why the summit's friendly language sits awkwardly beside the policy.

We previewed the meeting in our report on APEC trade openness as leaders met in Shenzhen, which was written before the outcome was known. This article covers what the two leaders said and how it fits with the new US controls.

Trump Xi AI cooperation: what each leader said in Shenzhen

Trump spoke at the leaders' meeting on November 19. AFP, as carried by Euronews, quotes him saying the United States opposes "any form of globalist regulation" of what he called "super intelligence," and that it would not let anyone control or stifle it. Euronews reported that he warned such regulation could hold the US back in its race with China. Outlets quote the line with small differences in wording, so the exact phrasing varies by report.

Xi took a different line. Xinhua reported that he called for stronger US-China communication and coordination on AI safety risks, and for AI to stay under human control. Chinese state media accounts, summarized by the Latin Times, also have him describing AI as a driver of innovation-led growth and urging regional cooperation on data governance and secure cross-border data flows.

The bilateral meeting took place the same afternoon. China Briefing reports that Trump proposed more cooperation with China on farm products, energy and "super intelligence," and that Xi asked for a longer list of cooperation items and a shorter list of problems. The two agreed to draft an action plan for "a constructive China-U.S. relationship of strategic stability," but no timeline or content was given.

What the leaders agreed on AI, and what they did not

The formal outcome was modest. According to the Latin Times, all 21 APEC economies adopted the Shenzhen Leaders' Declaration by consensus. It is voluntary and non-binding. China Briefing says its "Shenzhen Outlook" calls for facilitating cross-border data flows while protecting data and privacy, and that Xi's speech set out no new AI-specific measures. The Latin Times could not say whether China's proposed AI and digital economy statement was adopted, and it summed up the summit as leaving the AI divide between Washington and Beijing unresolved.

The clearest picture of what both governments wanted from AI talks comes from before the summit. On November 10, Treasury Secretary Scott Bessent told the National Committee on US-China Relations that the Shenzhen meeting would launch a bilateral dialogue on "superintelligence." The goals he listed were sharing best practices for testing advanced models and keeping "human judgment firmly in command," along with a direct channel for warning each other of serious AI incidents, according to The Standard. That channel was first agreed at a September meeting. Bessent also said the countries should "harbor no illusions" about their differences.

In the sources we reviewed, there is no published readout of the incident-channel details from Shenzhen. Some social media posts claimed the leaders announced joint AI guardrails, but we found no official White House or Chinese government text naming such a document, so we treat those claims as unconfirmed.

The new US measures on China's access to AI chips

The Commerce Department issued an interim final rule on November 16, according to TechCrunch. It requires licenses to export, reexport or transfer AI chips and servers to data centers anywhere in the world if those facilities may provide computing power to companies in China or other restricted countries. It also covers access through intermediaries and applies to data centers or networks above 12 million total processing performance, roughly the power of 750 Nvidia H100 or H200 chips.

Malaysia, Singapore, Indonesia, Thailand and Brazil are newly subject to licensing, with the rule citing a "higher-than-normal risk of diversion." Chips destined for US-based companies in those countries are exempt, and existing operators have 90 days to obtain licenses. The rule also rescinds the Biden administration's January 2025 AI Diffusion Rule. Commerce described the new control as "a logical extension of existing controls."

The rule took effect on November 19, the day of the leaders' meeting, after Federal Register publication. Our report on US export controls on consumer chips covers TSMC's reported pauses on some China-linked orders and Bessent's November 20 comment, reported by the Associated Press, that "The proper chip restrictions are in place." Washington's stated worry is diversion. The AP reported that the US is investigating Megaspeed, a Singapore-based company, over the possible diversion of restricted Nvidia chips to China.

Chip controls were not on the table in Shenzhen. China Briefing reports that US officials said export controls would not be negotiated with China during APEC, and that technology restrictions run on a separate track from tariff talks and could keep tightening.

Why AI diplomacy and chip policy are moving apart

The two tracks cover different subjects. The AI dialogue Bessent described is about safety practices: model testing, incident warnings and human oversight. The chip rule is about who can reach the computing power used to train advanced models, which is why Commerce focused on remote access through overseas data centers.

Trump's remarks fit that split. He opposes international rules on AI while arguing the US leads China and must stay ahead. The Standard reported that in September he was reluctant to add new guardrails because he wanted to preserve US leadership, even as the two sides agreed to set up the incident channel. On November 20 he told reporters that US companies building AI data centers abroad would face "a tremendous tax," which he tied to the race with China, as we covered in our report on the proposed data center tax.

Our reading is that the positions are consistent with one another even though they send mixed signals. Washington is willing to talk with Beijing about shared risks while limiting what Chinese firms can run on advanced chips. Whether talks can survive tighter controls is untested. Digitimes reported that China vowed to retaliate against the expanded controls, though we could not find details of any countermeasures.

What it means for businesses and what to watch next

China Briefing advises companies that rent cloud computing in Southeast Asia to check whether their providers are affected. Alibaba has warned that the rule could disrupt some of its global data center expansion plans. For operators already serving China-headquartered customers, the first compliance deadline falls about 90 days after the rule took effect.

Several things will show whether diplomacy and controls converge. The incident channel details and the action plan on "strategic stability" are still unpublished. Commerce may clarify the rule's scope, since an interim final rule can change after publication. The Standard reports that a fourth Trump-Xi meeting is expected in December at the G20 summit in Miami.

This article reports public statements and press coverage and is not investment or legal advice.

Frequently asked questions

What did Trump say about AI at APEC?

On November 19 in Shenzhen, Trump said the US opposes "any form of globalist regulation" of AI, according to AFP via Euronews. He reportedly warned that regulation could slow the US in its race with China. Wording differs slightly between outlets.

What did Xi say about AI cooperation?

Xinhua reported that Xi called for stronger US-China communication and coordination on AI safety risks and for AI to remain under human control. He also urged regional cooperation on data governance and cross-border data flows.

Did the US and China agree on AI rules at APEC?

We found no published agreement on AI rules. The leaders agreed to draft an action plan on a relationship of strategic stability, with no content or timeline given, and the APEC declaration is voluntary and non-binding.

What are the new US measures on China's AI chip access?

A Commerce rule issued November 16 requires licenses to send AI chips and servers to data centers that may provide computing power to companies in China or other restricted countries. It also adds licensing for data centers in Malaysia, Singapore, Indonesia, Thailand and Brazil.

Were the chip curbs announced during the APEC summit?

The rule was issued on November 16, three days before the leaders' meeting, and took effect on November 19. China Briefing reports that US officials said chip export controls would not be negotiated with China at APEC.

ShareXLinkedInWhatsApp

Sources

Related stories

Policy

Democratic Congress and AI Regulation: What Investors Face as Fed Hikes Loom

Democrats won both chambers in November, and the Fed is signaling more hikes. Investors now weigh limited federal AI action against state data center pauses and pricier financing.

Policy

Trump Threatens Tax on AI Data Centers Abroad as Senate Weighs AI Committee

Trump says US firms that build AI data centers abroad will face "a tremendous tax" but gave no details. Sen. Hickenlooper separately proposed a Senate select committee on AI.

Policy

US Export Controls on Consumer Chips Widen as Bessent Says Nvidia Curbs Are Enough

A Commerce rule in force since November 19 now reaches chips used in consumer devices, while Treasury Secretary Bessent says Nvidia's older chips need no further limits.