Trump's AI-China strategy ahead of Xi summit

Asked point-blank whether he worries AI could wipe out humanity, Trump answered flatly, 'No, I don't have any,' and recast the real danger as competitive rather than existential: 'I have concerns that if we don't win AI, we're going to be put in a very bad position.' ^[1]That dismissal puts him at odds with researchers inside the very labs racing to build the technology he wants America to dominate. Anthropic's Jacob Coxon warned that developers 'are racing straight to self-improving superintelligence and gambling with our lives,' and colleague Evan Hubinger, the company's alignment science lead, put the probability of human extinction from AI at more than 10 percent within the next decade. ^[1]The dissent has migrated to Capitol Hill: Sen. Bernie Sanders and Rep. Greg Casar introduced the Ban Artificial Superintelligence Act, with Sanders arguing 'the future of humanity cannot be left in the hands of a handful of Big Tech oligarchs.' ^[2]On X, the highest-engagement posts of the news cycle echoed the White House's triumphalist framing that whoever wins the AI race wins outright, a sharp contrast with the alarm coming from Anthropic's own safety researchers. The rift is not activists versus industry - it is the administration versus the scientists building the technology it is racing to win.

Trump has put a number on America's AI advantage: 'We are leading China right now by a pretty good period, I would say a year, which is, you know, considered a lot.' ^[3]The Center for Strategic and International Studies takes a starkly different view, concluding that 'the U.S. technical lead in frontier model development can be measured in mere months, as opposed to years.' ^[4]The gap is closing because Chinese developers, including DeepSeek, Zhipu AI and Moonshot AI, have shown they can match Western model quality at a fraction of the cost using distillation and efficiency techniques such as mixture-of-experts and sparse attention, and because China locked in AI partnerships with 28 Global South nations at Shanghai's World Artificial Intelligence Conference in July 2026. ^[5]A widely viewed CNN segment carried a similar warning from a China technology analyst, who argued that the more meaningful measure of Chinese AI strength is not benchmark scores but how deeply the technology has already embedded itself into ordinary citizens' daily lives, a shift that accelerated after DeepSeek's early-2025 model release. Trump's public confidence and the more sober assessments from research institutions and journalists covering China describe two different races, and only one of them assumes the United States is comfortably ahead.

For all the talk of an AI arms race, the Trump-Xi summit agenda has produced remarkably little on the issue that actually constrains China's frontier compute: export controls. Despite Nvidia CEO Jensen Huang's presence on the US delegation, 'there were no discussions of chip export controls' at the summit, and while an export license for Nvidia's H200 chip technically exists, no shipments have gone through because of Chinese purchase restrictions. ^[4]Those restrictions are not incidental. Even after Trump's Commerce Department cleared the H200 sale, with the US Treasury taking a 25 percent cut of every transaction, China directed its buyers to Huawei instead, a snub that Reddit's technology communities read as evidence Beijing no longer trusts an on-again, off-again US supply chain enough to stay dependent on it. The diplomatic track is similarly stuck: Beijing and Washington are discussing folding a long-planned intergovernmental AI dialogue into the broader economic meeting ahead of the summit, but the two sides are struggling to make progress because 'the Chinese side is said to view US requests as limiting both harmful outcomes from AI and Beijing's ability to develop its own capabilities.' ^[6]China also pushed back publicly, just two weeks before the summit, on US claims that Chinese labs are engaged in 'malicious' AI model distillation. ^[7]Commentators covering the summit's preview described an agenda of trade, rare earths, semiconductors and agriculture on paper, but predicted Trump would likely settle for symbolic wins rather than a substantive AI or tech accord, with Xi reportedly seeking softer US language on Taiwan in exchange for economic concessions.

While the administration frames AI purely as a contest to win, the CEO of one of its most important labs is arguing for something closer to a negotiated ceasefire on safety pacing. Dario Amodei has called for AI labs to give safety measures time to catch up, proposed a narrow antitrust waiver so competitors could jointly set safety standards, and floated coordinating with other governments, including authoritarian ones like China's, so that no single actor accelerates unchecked while others hold back. ^[8][9]That is a markedly different posture than either Trump's hands-off competitiveness or the more hawkish currents circulating elsewhere in Washington. Reddit's policy-focused communities have been reacting with alarm to reports that factions inside the administration discussed restricting or banning foreign open-weight Chinese models such as Kimi K3, a move critics see as protecting incumbent US labs' market position rather than genuinely countering China, and to a think-tank report that raised, as a hypothetical, whether the US should consider physical strikes on Chinese AI data centers to prevent a rival from reaching artificial general intelligence, an idea most commenters dismissed once they read past the headline. Sen. Bernie Sanders and Rep. Greg Casar's push for a legislative ban on superintelligence adds yet another lane: Casar has said 'Congress should immediately ban AI systems too powerful to control.' ^[2]Between Amodei's cooperative slowdown, Congress's prohibition push, and the more hawkish ideas percolating in policy circles, the administration's simple 'race to win' framing is competing with several very different visions of how the US should actually handle Chinese AI competition.