The Chip Export Controls Saga
How are US chip export controls — from Biden's 2022 regime through Trump's reversals, license deals, and tariffs — reshaping the global AI compute supply chain and US-China technological decoupling?
Canonical Synthesis
Author: terry-tang | Last updated: 2026-04-19
The chip export controls saga is the story of how the United States attempted to weaponize the semiconductor supply chain against China's AI ambitions — and how that attempt has evolved, repeatedly, through policy reversals, informal deals, diplomatic pressures, and the stubborn reality of industrial substitution.
The regime began on October 7, 2022, when the Biden administration's Bureau of Industry and Security published the most expansive expansion of semiconductor export controls in decades. The October 7 rules targeted advanced logic chips, DRAM, and the manufacturing equipment used to make them, effectively cutting China off from the most capable AI accelerators and the means to produce them domestically. The controls were built on a threshold model: chips above specified performance parameters — defined by interconnect bandwidth and aggregate compute — required an export license presumed denied for Chinese end-users.
The October 2023 expansion tightened those thresholds, closing loopholes that Nvidia had exploited with the A800 and H800 chips designed to comply with the original rules. NVIDIA responded with the H20 — a chip purpose-built to fall below the new thresholds — in what became a cat-and-mouse dynamic between American chipmakers seeking Chinese market access and American regulators seeking to deny it.
Biden's final major initiative was the AI Diffusion Rule, finalized in January 2025 and scheduled for May 15 implementation. The rule would have extended export control logic beyond chips to model weights and created a three-tier global licensing system sorting nearly every country into preferred, restricted, or prohibited access categories. It represented the most architecturally ambitious attempt to govern global AI compute distribution through export control authority.
The Trump administration rescinded the Diffusion Rule two days before its effective date, citing complexity and diplomatic damage. What followed was not a coherent replacement policy but a series of ad hoc executive actions: the April 2025 H20 license requirement that cost NVIDIA $5.5 billion in write-downs; the July 2025 informal deal reversing the ban in exchange for a 15% China-sales revenue remittance; and the January 2026 simultaneous BIS H200 case-by-case review policy and Section 232 chip tariff proclamation. Each action reflected different pressures — denial objectives, revenue extraction, diplomatic bargaining — without a unifying strategic framework.
Throughout this period, China's domestic chip industry did not stand still. Huawei's Ascend 910C entered mass shipment in May 2025, offering Chinese AI developers a domestically sourced alternative at roughly 60% of H100 inference performance. The Ascend 920 roadmap pointed toward continued improvement. SMIC's role as Huawei's foundry partner demonstrated that China could produce commercially viable AI chips without TSMC, though at process node disadvantages that remained meaningful.
The Arc
2022: Foundation. The October 7 rules established the principle that advanced semiconductor technology was a national security asset subject to export denial — a departure from the prior presumption of open trade in commercial electronics.
2023: Tightening and Evasion. The October 2023 expansion closed the A800/H800 loophole; NVIDIA responded with the H20. The dynamic illustrated a structural problem: any performance-threshold approach invited threshold engineering.
2024: Preparation. Both the Biden Diffusion Rule drafting process and Huawei's chip development accelerated through 2024. The global response — from the EU, Gulf states, UK, and India — began to materialize in the form of sovereign compute investment.
2025: Disruption. The Trump rescission of the Diffusion Rule, the H20 ban and reversal, and the HUMAIN and Stargate UAE announcements all occurred within six months, producing a policy environment of maximum uncertainty. Huawei's mass shipments demonstrated that denial was incomplete.
2026: Consolidation as Uncertainty. The dual January 2026 actions — Section 232 tariffs and H200 case-by-case review — extended discretionary executive control over the entire Nvidia GPU lineup while raising novel legal questions about the authority's boundaries.
Interpretations
Decoupling-as-strategy
US chip controls represent a coherent strategy of technological decoupling that will ultimately constrain China's AI trajectory, even if Chinese domestic alternatives emerge. The multi-year head start, the process node gap, the EDA software dependencies, and the HBM supply constraints all compound over time. Denial does not need to be complete to be effective.
Proponents: Export control advocates, national security hawks, some academic researchers on technology competition.
Bargaining-chip reading
The revenue-share deal and the pattern of ban-and-reversal suggest that chip controls have become a bargaining instrument in broader US-China economic negotiations rather than a sustained denial strategy. The 15% revenue-share arrangement in particular reflects a view that extraction is preferable to denial when denial is incomplete.
Proponents: Trade economists, some industry analysts, critics of export control coherence.
Industrial-policy reading
The real significance of chip export controls is not denial of China's AI capabilities but acceleration of US domestic semiconductor manufacturing. TSMC's $165 billion Arizona commitment, the CHIPS Act investments, and the TSMC process node advantage are the durable outcomes. The controls create the political justification for industrial policy that market mechanisms alone would not have produced.
Proponents: Industrial policy advocates, semiconductor industry analysts, some CHIPS Act architects.
Open Questions
- Can performance-threshold-based export controls survive indefinitely as a denial tool when Chinese industry has demonstrated the ability to engineer around them?
- Does the Section 232 chip tariff survive legal challenge under WTO commitments and domestic trade law?
- How does the 15% revenue-share deal interact with formal export control regulations, and what is its legal status?
- At what point does Huawei's Ascend roadmap close the performance gap with Nvidia sufficiently to make denial strategy moot?
- Does the TSMC Arizona investment deliver the strategic decoupling benefit its advocates claim, or does it create new interdependencies with Taiwan's political fate?