Reshma Kapadia

President Donald Trump and Chinese leader Xi Jinping will likely have little choice but to discuss artificial intelligence and its dizzying advancement when they meet this week. Their countries, after all, are engaged in a race to be the world's pre-eminent AI power.

Seemingly everyone is talking about artificial intelligence: Industry leaders are calling for slowing the pace of development; researchers are warning about catastrophic risks; and concerns about AI going rogue are mounting. But few expect Trump and Xi to get very far on the contentious topic when they meet Thursday in Washington, D.C.

In fact, the bar for their second summit is relatively low. Analysts expect they will focus most on reinforcing the fragile stability between the two countries. That's no easy task.

The world's two largest economies share a growing list of frictions, including U.S. restrictions on China's access to advanced chips and China's grip on rare-earth minerals. Earlier this month, U.S. intelligence agencies said Chinese AI start-ups are inappropriately tapping U.S. frontier models to get ahead, though Beijing has repeatedly rejected that characterization.

Strategists say both countries want the current detente to last long enough for each economy to become more self-reliant in key sectors, such as critical minerals for the U.S. and advanced semiconductor chips for China. Both are essential in their AI race.

Trump initiated the idea of a U.S.-China AI dialogue over the summer. According to geopolitical consultants, the two sides had not been able to agree to an agenda for AI even at a working level, nor who should be in the room, just days ahead of the summit. On Sunday, though, Treasury Secretary Scott Bessent, after meeting with China's vice premier He Lifeng, said the two had agreed to an AI dialogue and the U.S. had proposed a "notification mechanism" for national security-related incidents

Still, the recent spotlight on AI's threats has created a further opening for a conversation between Trump and Xi. Last week, OpenAI reported that swarms of AI agents escaped testing and launched a cyberattack. Earlier this month, top researchers resigned from Anthropic and Google, sounding an alarm about safety concerns and the risk that advanced AI could even wipe out humans.

Some notable AI leaders at the center of the debates on AI safety-including OpenAI Chief Executive Sam Altman and Nvidia CEO Jensen Huang-are expected to attend the state dinner Trump is hosting for Xi on Thursday.

While China has flagged concerns about AI safety in recent years, analysts expect any discussion Thursday to be narrow and more surface-level than initially expected. One reason is U.S. lawmakers can't agree on whether or how to create safeguards for the technology. Despite growing pressure from Congress, the national security community, and the industry itself to develop guardrails, Trump has called AI safety fears a hoax. Others in the administration warn any slowing would compromise America's position in the AI race with China.

As the debate intensifies, Trump on Saturday said he would create an AI force and appoint an "AI czar." The economic stakes when it comes to AI remain especially high, with Morgan Stanley estimating an eventual addressable market of $40 trillion. It is also critical for the global economy, markets and geopolitics.

"Whoever can disseminate and lock in their technology around the world is going to reap extremely high economic returns and put the other country in much worse position," says Aalok Mehta, director of the Wadhwani AI Center at the Center for Strategic and International Studies.

U.S. AI labs like Anthropic and OpenAI are trying to build the most powerful and cutting-edge technology in the world, helped by the U.S. advantage of controlling about 75% of global compute. China is taking a more pragmatic approach, in part out of necessity because of less access to compute. It is focusing on cheaper open-source models and wide-scale adoption around the world, so countries in Africa and Asia are using China's tech stack.

Still, Chinese AI start-ups' open-source models have closed the gap with U.S. frontier models-even amid U.S. export restrictions limiting their access to advanced technology. That has contributed to Beijing's increased confidence.

"Xi sees China as the new internet once in a lifetime opportunity for Xi. If he wants to build a legacy, how great if China can beat America or even be in No. two in a new technology? China has never been in such a stage where it's so close to America," said George Chen, chair of digital policy at geopolitical consultancy the Asia Group, who formerly worked on policy at Meta Platforms, in a news briefing.

Many liken the AI battle between the two countries to a nuclear arms race, but Alvin Wang Graylin, a senior fellow on technology at the Asia Society Policy Institute's Center for China Analysis, sees it as more of a space race. He notes that Bessent and Trump's recent comments that winning AI is everything has added to Beijing's wariness.

"We are looking at something that should be public good and treating it like a nuclear weapon and the Manhattan project," Graylin says. "We need to separate safety from national security from economic competition. You can't race your way to win against bad actors or runaway AI."

Even if the focus of any dialogue is squarely on shared safety risks, analysts aren't hoping for much headway. Gregory Allen, founder of Decision Tree Research, which focuses on geopolitics, policy and technology, doesn't see anything analogous to arms inspections as "even remotely possible" for AI at this stage.

The lowest of the low-hanging fruit for Trump and Xi would be to reaffirm the agreement the two countries reached during the Biden administration: to prohibit AI from being involved in the chain of command for nuclear weapons. Allen also says the two could agree to take steps to reduce the risk of unintentional cyberattacks, AI escaping human control in any way, or AI providing capabilities to those looking to develop bioweapons or pathogens.

Asia Society's Graylin has an even lower bar: Just an agreement on how to define AI safety and to continue negotiations to create an AI framework would be a win.