Border, Toll, Leak — The Week US Chip Controls Split Into Three Regimes at Once

Three days, three contradictory export-control signals — and all of them came from the same Washington.

On Saturday, May 31, the Trump administration cleared Nvidia to ship H200 AI accelerators to China in exchange for a 25% export fee, escalating the "toll booth" framework beyond the H20-only regime that had defined the prior weeks. Forty-eight hours later, the same Commerce Department was clarifying that AI chip export controls extend to overseas subsidiaries and affiliates of Chinese companies — closing the loophole that had let Chinese entities procure restricted Nvidia and AMD silicon through their global arms. And folded in between: the New York Times published an investigation documenting that Chinese military-linked entities had been procuring Nvidia chips through intermediaries for years despite the existing regime.

Read those three together and US export controls are no longer one policy. They are three.

Tier one — the extraterritorial ban. The big regulatory move of the week was the geographic expansion of US controls beyond US borders. At least thirteen separate articles in three days covered Commerce's clarification that the rules now follow Chinese ownership, not Chinese geography. A Chinese firm's Singaporean subsidiary, its Dubai trading desk, its Malaysian back-end partner — all now within scope. This is a categorical shift. Export controls used to police what crossed the US border; they now police what reaches a Chinese-owned buyer anywhere on Earth. Senators have already flagged that loopholes persist, suggesting more perimeter expansion to come.

Tier two — the transactional toll. The H200 / 25% fee is structurally different from a ban. It is a price. The US is now monetizing the regulatory chokepoint rather than slamming it shut. For Nvidia, the math is simple: H200 silicon billed at a 25% premium into China still beats no shipment at all, especially if the alternative is a Huawei Ascend cluster filling that socket. For the Treasury, the toll generates revenue from the very arbitrage the tier-one rules were designed to eliminate. The two tiers are in active tension.

Tier three — the leak. The NYT exposé and the SCMP coverage of how Chinese AI chip designers are re-architecting around domestic process nodes tell the same story from opposite ends. The regime has been porous for years. And the leakage has not just continued — it has shaped the response. Huawei chairman Eric Xu spent the weekend publicly arguing that US export controls "supercharged" China's domestic chip industry. AMD disclosed that its ROCm software stack has captured roughly 4% of China's AI accelerator market — a small share, but a foothold built precisely because Nvidia's CUDA-locked socket cannot legally fill all of China's demand.

For investors, the three-tier framework changes the question. The old question was "will export controls tighten or loosen?" That question now has no single answer. Controls are tightening extraterritorially, loosening transactionally, and leaking structurally — all at once.