Donald Trump arrived in Beijing this week with an unusual delegation. Alongside the traditional cabinet officials sat Apple's Tim Cook, Tesla's Elon Musk, and Nvidia's Jensen Huang. The guest list signalled what the summit's published agenda did not: technology, and AI specifically, has become as central to US-China relations as trade balances or Taiwan. What Trump and Xi Jinping discuss when the doors close is partly about chips, partly about the shape of warfare, and partly about whose model of the future wins.
The conventional story of the past two years has been that the United States is winning the AI race. American firms produce the leading frontier models. American hyperscalers operate more than 5,500 data centres, roughly ten times any other country. American capital has poured into AI in volumes that no rival can match. By every metric that the US tech industry chose to emphasise, the gap was widening.
Then it wasn't. In January 2025, the Chinese lab DeepSeek released its R1 model, which matched leading American systems on most benchmarks and was reportedly trained for less than $6 million on chips it was not supposed to have. Nvidia and Broadcom lost 17 percent of their market value in a single day. Analysts called it America's Sputnik moment, and the comparison stuck. Stanford's annual AI Index Report concluded last month that the model-performance gap between US and Chinese frontier systems has "effectively closed."
The deeper finding is more interesting than the headline numbers. The two countries are not really racing toward the same finish line. The American AI sector is organised around frontier model capability and the pursuit of artificial general intelligence: a small number of labs spending enormous sums to build the next, bigger thing. China is organised around diffusion, the integration of AI across manufacturing, logistics, healthcare, and government. Beijing has mandated 70 percent AI penetration in key industries by 2027. JD Logistics offers 12-hour delivery in major Chinese cities using AI routing. Manufacturing adoption sits around 67 percent, nearly double the equivalent US figure.
So which side is ahead depends entirely on what you measure. The US still produces more high-impact patents and more top-tier models. China leads in publications, in physical AI and robotics, and in the raw electricity needed to run any of it. In 2025, China added 540 gigawatts of new power capacity. The US added 54. American data centres are projected to consume 9 percent of the country's electricity by 2030, and the bottleneck is no longer chips but generation. China has fewer such constraints.
None of this seems likely to be settled at the summit. A breakthrough on semiconductor export controls is widely seen as off the table. The realistic agenda involves AI-enabled warfare, a category that has moved from theoretical to operational over the past year through deployments in Gaza, Iran, and the recent Venezuelan raid. There are also, paradoxically, the beginnings of bilateral conversation about constraints. Both governments have at least floated agreements around AI in nuclear command and control. The framing is starting to resemble the Cold War arms control playbook, which is either reassuring or alarming depending on your reading of that history.
One subplot worth noting: the same week, Anthropic released a cyber-focused model called Mythos to a handful of businesses and security firms but withheld it from the public, citing "unprecedented cybersecurity risks." The implication is that frontier AI is now considered a potential national security asset in its own right, not just a productivity tool. That changes what export controls would have to police, and it changes what Xi and Trump are actually negotiating about.
The summit will produce a communiqué. It will be heavily managed, carefully translated, and parsed for tone by every analyst in both capitals. Whether the substance moves at all is less clear. What seems already clear is that the framework everyone inherited, of two countries racing for the same prize using comparable tools, has stopped describing the actual competition. Both sides have shifted ground. The summit is the moment that becomes harder to ignore.