The Enforcement Record That Indicts the Policy
Federal prosecutions do not typically mark the success of an export control regime — they mark how thoroughly a regime has been routed. The March 2026 indictments charging three individuals connected to Super Micro Computer with diverting $2.5 billion in Nvidia chips to China described encrypted communications and layered intermediaries that had been operating long enough to produce a paper trail measured in billions of dollars. The ABC News framing — a 'tangled web of lies' involving American-made AI servers secretly shipped to China — captured the operational scope. What it did not capture is the structural implication: a smuggling network that large is not a failure of a few bad actors, it is evidence that the legal supply was insufficient to meet demand, and that the gap was priced and routed around at scale. Prosecution closes individual channels. It does not reduce the demand that created them.
Jensen Huang's Zero Percent Is a Policy Verdict
When the CEO of the world's dominant AI chip company describes his firm's market share in the world's second-largest AI market as 'zero percent' and characterizes the policy that produced that outcome as having 'largely backfired' , it is not a complaint about quarterly earnings — it is an assessment of whether the policy achieved its stated objective. Nvidia had halted production of its China-intended chips by February 2026 , and the U.S.-approved versions designed to comply with export rules remained unsold months later, with Nvidia openly worried that Huawei and other domestic Chinese firms were absorbing the enterprise AI demand that Nvidia had been frozen out of . The licensed path that emerged in May 2026 offers a theoretical route back, but the customer relationships, the deployment momentum, and the trust that Chinese enterprises extend to a supplier they can rely on — those do not return on the same timeline as a regulatory filing. Nvidia's China position has not been paused. It has been redistributed.
China's Domestic Silicon Bet Is Not a Contingency Plan
The distinction between a contingency response and a long-term industrial strategy matters here. China's push toward 70% homegrown silicon wafer use, with domestic manufacturers scaling 12-inch production , did not begin as a reaction to the latest export control round — it reflects a state-directed investment thesis that predates the current restrictions by years. The American Enterprise Institute's analysis of what it called the 'lithography loophole' characterized China's manufacturing approach as a path to chip self-sufficiency that works around equipment access constraints rather than depending on them being lifted . The U.S. Commission warning about China's use of open-source AI models adds a parallel vector: hardware denial was never the only axis of competition, and on the software and methodology axis, the access controls simply do not exist. China is not trying to rebuild what it cannot import. It is building what it never had — and the process is already past the point where export control relaxation would slow it down.
Washington's Incoherent Signal Has Its Own Costs
The Trump administration's chip policy posture has been inconsistent enough to impose costs independent of whatever the actual rules are. A House chair identified AI chip sale limits to China as a legislative priority in February 2026 , while by May the U.S. trade chief was characterizing export controls as not a major topic in bilateral talks . Reuters documented a dynamic in which Washington was pulling back from aggressive restriction at the same moment Beijing's own export controls were maturing into a meaningful sovereign lever . The Fulcrum's analysis framed Trump's approach as a 'nuanced' chip export policy designed to preserve U.S. dominance — but what nuance looks like from inside a supply chain is unpredictability, and unpredictability freezes procurement decisions. Nvidia's approved China chips sitting unsold is not just a Nvidia problem; it is evidence that the buyers who would have purchased them could not assess whether buying was safe. Policy incoherence does not split the difference between restriction and access — it produces the worst outcomes of both.
The Race Has Two Clocks Running at Different Speeds
The debate over who is winning the U.S.-China AI hardware competition has been conducted almost entirely on the wrong time horizon. Near-term, U.S. restrictions have cost Nvidia measurable revenue and market share , and China's frontier model development has faced genuine compute constraints. On a longer arc — the one that determines whether China achieves semiconductor self-sufficiency — the restrictions have accelerated exactly the industrial investment they were designed to prevent. The BBC's assessment that China is winning one AI race while the U.S. wins another is accurate but incomplete: the race China is winning is the one that makes the other race less consequential. A domestic Chinese silicon supply chain capable of meeting 70% of its own wafer needs is not catching up to TSMC — it is building the foundation for a compute infrastructure that does not depend on the geopolitical stability of the Taiwan Strait. The enforcement actions and licensing exceptions are quarterly adjustments. China's domestic chip build-out is a decade-long structural bet already far enough along that the United States cannot spend its way out of the outcome.