Chapter 1
Since October 2022 the United States has tried to hold a lead in…
Since October 2022 the United States has tried to hold a lead in artificial intelligence by controlling the machines that make it. The rule the Bureau of Industry and Security issued that month created chip classifications aimed at China, set fabrication thresholds carrying a presumption of denial, and for the first time barred American citizens from supporting advanced chip production at Chinese plants. Everything since has been a variation on it: a 2023 rewrite that redefined the controlled chips after Nvidia engineered around the first set, a January 2025 framework that tiered the entire world, and that framework's rescission four months later in favour of case-by-case discretion.
Chapter 2
The theory was denial
The theory was denial, and it has not held cleanly. In September 2023 a teardown of a Huawei phone found a 7-nanometre chip made without the lithography the controls were meant to withhold. On inauguration day in 2025 the Chinese lab DeepSeek released R1, a reasoning model built despite the controls, which took more than $600 billion off Nvidia's market value in a single session and became the central exhibit for the argument that restrictions were spurring China rather than slowing it. By August 2025 the policy had inverted far enough that Nvidia and AMD agreed to hand the US government 15% of their China AI-chip revenue in exchange for export licences — a national-security control administered as a revenue share, without clear precedent. Stanford's 2026 index put the measured performance gap between the best American and Chinese models at 2.7%.
Chapter 3
In 2026 the instruments multiplied and the target moved from chips to…
In 2026 the instruments multiplied and the target moved from chips to models, and then to the machines they run in. In June the government used export-control authority against a named frontier model for the first time, cutting off two Anthropic models for every foreign national including those inside the United States, then withdrew the order eighteen days later without publishing either decision. The Pentagon's annual list of Chinese military companies added Alibaba and Baidu, on indirect state and ministry affiliation and alleging nothing about their AI. In July the Federal Communications Commission barred new foreign-made humanoid robots from the American market, a category in which two Chinese firms each shipped more than 5,000 of the roughly 15,000 units sold worldwide in 2025. The argument turned personal in the same weeks: the White House science adviser accused a Chinese lab of distilling an American model to build Kimi K3 and published no evidence, the Treasury Secretary put sanctions on the table, and the two governments' own evaluators then measured that model and found it well below the frontier on cyber capability. In September the allegation became a document: three US agencies named six Chinese AI companies in a joint advisory accusing them of distilling American models at industrial scale, likely with Beijing’s awareness — an advisory that cites no statute, publishes no evidence and carries no penalty.
Chapter 4
The competition now has two blocs
The competition now has two blocs, and they are drawn on opposite principles of disclosure. The United States launched Pax Silica in December 2025 and had 24 signatories to its declaration by the second summit in June 2026, alongside a separate statement on AI opportunity signed by 35 economies — text and roster published in full, commitments non-binding and unfunded until a $50 million customs-credentialing notice in August. Three weeks after that summit, states meeting in Shanghai founded a World Artificial Intelligence Cooperation Organization: the count given at the signing was 29, Beijing now says 38, no member list has been published at either figure, and the founding agreement has never been produced. Kazakhstan is so far the only country reported in both. Then in September 2026 the United States hosted a G20 innovation ministerial in North Carolina at which both China and Russia were listed among those present for a consensus statement on emerging technology — so the same year that produced two rival rosters also produced a table with everyone at it, and a text agreed there that binds none of them. What the American record does not have either is an instrument for its own model-level actions: the chip rules were published in full and can be read line by line, while the orders that reached a frontier model in 2026 arrived with no classification, no regulation cited and no notice — and the agency that administers all of it now reports licence processing running at 62 days against the 38 in its last published report. Three weeks after the North Carolina meeting, Xi Jinping’s state visit to Washington produced something the two governments built together: a standing dialogue on AI’s risks and benefits, a channel for reporting incidents, and a next meeting due in November — announced in two separate texts that do not agree on its name, and neither of which mentions chips. Whether the observable record supports the arms-race framing, or something more tangled, still turns on documents that have not been produced.
And now
Both governments are tightening what crosses the border, and Beijing has now put a date on part of it. On 19 October China issued export controls on long-range drones and their engines, autopilots and autonomous-navigation technology, effective 18 November. Its proposed controls on 23 AI technologies, from large models to computing clusters, remain open for comment until 15 November.
In Washington, a Homeland Security order published on 20 October bars DeepSeek’s products from federal civilian agencies, under the supply-chain law also used against Anthropic. DHS says data held in China could give Beijing intelligence on Americans; that is its own assessment. On 19 October the State Department sanctioned Shanghai Zhisuan, a start-up founded by a former Google engineer, and the fund that backed it, for trade-secret theft — the government’s findings, not a court’s. The Treasury has issued its first penalty for an unreported American stake in a Chinese AI company, $200,000, and on 26 October proposed banning such investments outright; that rule is not yet adopted.
Washington and Beijing agreed in September to a standing AI dialogue, with a first exchange due in November.
Where it stands today →