For the first time since ChatGPT catapulted artificial intelligence into the public consciousness, a plurality of Americans now believe China, not the United States, is winning the race to build the most advanced AI. That is the headline finding from a Pew Research Center survey released Thursday, a data point that carries unusual political weight two months before President Donald Trump hosts Xi Jinping in Washington for talks that, by Trump’s own confirmation, will put AI at the center of the agenda.
According to a report by the South China Morning Post summarizing the Pew findings, 36% of Americans surveyed said they believe China is more advanced than the United States in AI. Only 12% said the United States is ahead. Another 18% viewed the two countries as roughly equal, while 33% said they were unsure — a large “don’t know” bloc that, in itself, signals how much the once-clear narrative of American technological supremacy has fractured over the past eighteen months.
That fracture has an obvious catalyst. When ChatGPT launched in late 2022, US labs were widely thought to hold a lead of three years or more over anything coming out of China. DeepSeek’s R1 in early 2025 compressed that gap to months. And this July, the near-simultaneous release of Moonshot’s Kimi K3 and Alibaba’s Qwen 3.8, the latter a 2.4-trillion-parameter model that Alibaba claims trails only OpenAI’s Fable 5, has convinced many analysts, and evidently a growing share of the public, that the frontier gap is now measured in weeks. The public’s shifting mood tracks the technology.
A Partisan Split — With a Twist
The Pew data reveals a partisan divide that runs counter to some standard assumptions about who worries most about China. Democrats were significantly more likely than Republicans to say China has pulled ahead in AI: 43% of Democrats surveyed named China as the leader, compared with 30% of Republicans.
Republicans, meanwhile, were more likely to say the United States is still on top, though even in this camp the numbers are strikingly modest. Just 17% of Republicans named the US as the AI leader, versus only 9% of Democrats. In other words, even among the president’s own political base, fewer than one in five voters is willing to affirm that America is ahead in the technology that his administration has repeatedly described as central to national power.
The gap partly reflects broader partisan patterns in how each side views China’s rise, but it also reflects a communications problem for the White House. Trump officials have spent much of 2026 arguing publicly that Chinese labs are only closing the gap because they are distilling American frontier models, an accusation the administration formally leveled against Moonshot’s Kimi K3 earlier this week, as EastFrontier reported. Yet the Pew numbers suggest that framing has failed to persuade the very voters most inclined to trust the administration’s read on Beijing.
Trump: “Whoever Wins That Race Is Probably Going to Win”
The survey lands as Trump prepares to host Xi at the White House on September 24, a meeting the president has said will include AI as a headline topic. Speaking to reporters this week, Trump doubled down on the framing his administration has used since January.
“We’re leading China. We want to keep it that way,” Trump said, according to the SCMP account. He went further in describing the stakes: “Whoever wins that race is probably going to win. Period.”
The rhetoric is characteristically absolute, but it also underscores why the Pew numbers are politically awkward. If the president’s message is that America is winning and must protect that lead, a survey showing that three times as many Americans believe China is ahead as believe the US is ahead complicates the case for the sweeping domestic mobilization with export controls, chip subsidies, restrictions on Chinese open-weight models that the administration has been assembling.
The September summit itself is expected to be the first substantive US-China dialogue dedicated specifically to AI. As covered in our earlier reporting on the scheduled September talks, the two governments have been quietly building an agenda that touches on model safety, export controls, and, according to some briefings, the possibility of information-sharing on catastrophic-risk incidents. Public perception of who is ahead will shape how much political room Trump has to make concessions or demands.
Why Perception Has Shifted
Several concrete developments over the past six months help explain the Pew results. Chinese open-weight releases have become genuinely competitive: Kimi K3, Qwen 3.8, and the imminent DeepSeek V4 have each traded blows with the top US closed models on public benchmarks. David Sacks, the former White House AI tsar, recently observed that “Kimi K3 just fixed 15 critical security bugs that Codex and Fable refused because of ‘cyber guardrails,'” a comment widely circulated among developers.
The infrastructure story is shifting too. Last week Z.AI, formerly known as Zhipu, powered up a one-gigawatt data center built entirely on Chinese-made chips, a milestone that undercuts the assumption that Beijing’s AI ambitions are perpetually bottlenecked by Nvidia’s export-controlled hardware. And at the World Artificial Intelligence Conference in Shanghai, Xi Jinping delivered a keynote address in which 28 countries, mostly from the Global South, signed on to China’s “AI for all” framework.
For American voters watching the headlines, Chinese open models topping leaderboards, Chinese data centers going online without Nvidia, Chinese diplomacy locking in Global South partners, the impression that the United States is losing ground is not irrational. It is, in significant part, an accurate reading of the surface signals.
What Comes Next
Pew’s numbers will feed directly into the political economy of the September summit. Congressional hawks will use them to argue for tougher restrictions, including the broad ban on Chinese open-weight models that Axios has reported some Trump officials are pushing. Industry voices, including the 200 Silicon Valley founders who signed an open letter this week opposing such a ban, will argue that the survey shows exactly why the US must double down on openness and speed rather than retreating behind walls.
What the Pew survey cannot resolve is the underlying question. Model benchmarks, compute buildout, developer adoption, and diplomatic reach all now tell overlapping but distinct stories about who is “ahead.” What the survey does prove is that, in the court of American public opinion, the presumption of US dominance in AI has ended. Whether Trump can restore it, or whether the September meeting with Xi becomes an acknowledgment of parity, will define the next phase of the technology contest.