The Toll Booth Pivot — The Week US Chip Controls Became a Two-Tier Tariff

Two Decisions That Looked Like Contradictions

The most decisive signal in US semiconductor news over the past 72 hours isn't the volume of headlines. It's that two seemingly contradictory decisions landed on the same weekend.

On May 31, the Trump administration cleared Nvidia to ship H200 AI accelerators to China — conditional on a 25% export fee — bumping the export-permitted lineup one tier above the H20-only regime. Yet the same weekend, the US Commerce Department began closing the workaround that let Chinese firms procure restricted Nvidia and AMD AI chips through overseas subsidiaries and affiliates. Reported jointly by Bloomberg, Reuters, and SCMP, the new measures extend BIS rules to Chinese-owned entities operating outside mainland China.

Read together, the two moves say one thing: what gets blocked, gets blocked harder; what gets through, pays a toll.

From Sanction to Toll

This is the inflection point where US chip export policy crossed from binary control into monetized gatekeeping. For three years, BIS controls operated as a yes/no switch. The moment a 25% fee attaches to H200, Washington has created its first revenue-bearing export-control instrument. Given that Nvidia's quarterly H200 China revenue could plausibly reach billions of dollars, the 25% fee isn't a one-shot negotiating chip — it's a recurring fiscal lever.

Brussels moved in parallel. The European Commission's draft 'Chips Act 2.0' would — beyond the subsidy-heavy 1.0 — give Brussels authority to intervene in semiconductor supply contracts and levy fines. Where 1.0 said "we will help you build a fab," 2.0 says "we may cut your line or insert ourselves at will." Same logic as Washington's toll booth: convert control into money and authority.

China's Two-Track Response

China is responding in two directions at once.

First, domestic acceleration. BYD on May 29 unveiled what it calls China's first 4nm intelligent-driving chip. The same day, reports surfaced that ByteDance is designing in-house AI inference chips modeled on Groq's LPU architecture, joining Alibaba and Baidu in the captive-silicon push. Huawei chairman Eric Xu went further on May 30, publicly crediting US export controls with "supercharging" Chinese semiconductor development.

Second, rewriting the scaling law. Huawei has pitched a new chip 'scaling law' that bypasses transistor shrinks via chiplet, packaging, and cluster-level architecture — a way around EUV inaccessibility. SCMP and Asia Times framed it as "China rewriting the rules of scaling."