Tensions between the United States and China have increasingly centered on high-end artificial intelligence chips. Washington has imposed export controls designed to block China’s access to advanced GPUs and accelerators. These measures were meant to slow progress in military AI while protecting U.S. technological advantages.
In 2025, the U.S. government altered course. Instead of blanket prohibitions, it introduced a revenue-sharing system that allows chipmakers to sell certain processors into China if they surrender a portion of their sales to the U.S. Treasury. This approach mixes trade policy with national security oversight in a way that has not been tried before.
The change has broad consequences. It is beginning to reshape supply chains, shift market expectations, and test the cohesion of allied responses. The effects are being felt in China, the European Union, South Korea, and Japan, each of which is recalibrating its own semiconductor strategy in response.
U.S. Export Controls on AI Chips: Bans, Thresholds, Licensing
The Bureau of Industry and Security issued successive restrictions on advanced computing chips bound for China in 2022 and 2023. Those updates barred sales of flagship GPUs such as Nvidia H100 and H200 to most Chinese end users.
In January 2025, Washington set global performance thresholds that created a permitted band for lower-powered models and blocked the newest, most capable accelerators. The stated objective was to protect U.S. national security and to slow military AI programs in China.
By March 2025, additional Chinese firms were added to the Entity List for activities tied to advanced AI chips and supercomputing. In April 2025, the U.S. banned even the previously permitted band of “green zone” models as rhetoric and scrutiny intensified. The result was tighter licensing, broader screening, and reduced flexibility for Chinese buyers and U.S. suppliers.
These steps defined the control perimeter. They also exposed the commercial and geopolitical costs of a near-total cutoff, which set the stage for the subsequent shift toward a revenue-sharing license framework.
Shift to Revenue Sharing: Terms of the New Deal
In mid-2025, the U.S. government replaced blanket export bans with a revenue-sharing framework. Select AI chipmakers could resume exports to China if they allocated a portion of their sales revenue to the U.S. Treasury. The final rate was set at 15 percent, down from an initial proposal of 20 percent for midrange chips.
Under this framework, Nvidia resumed shipments of its H20 inference GPU, and AMD resumed sales of its MI308 model. The policy imposed limits on performance and model eligibility but removed the total prohibition on these midrange chips.
This arrangement functions as a hybrid of trade policy and export control. By collecting revenue on each sale, the U.S. Treasury gains a return comparable to a tariff while maintaining strategic oversight. The shift marks a clear departure from conventional export bans, establishing a mechanism for regulating Chinese access to AI technology through financial and licensing conditions.
The policy also signaled the commercial stakes for U.S. firms. Nvidia and AMD, which had projected billions in lost sales under full bans, regained access to the Chinese market while contributing to federal revenue. The revenue-sharing model set the stage for China’s response and for broader adjustments in allied semiconductor strategies.
China’s Response and Strategic Adjustments
China responded cautiously to the resumption of AI chip imports. Regulators questioned major domestic firms, including Tencent, ByteDance, and Baidu, on their interest in Nvidia’s H20 chips. Authorities emphasized that these purchases could create exposure to U.S. scrutiny of data and intellectual property. Guidance from China’s Cyberspace Administration advised firms to limit use of imported chips in government or sensitive projects.
No new export bans were imposed. State media and regulators focused on discouraging over-reliance on foreign hardware. At the same time, domestic chipmakers accelerated development of indigenous AI accelerators. Huawei’s HiSilicon and other local firms prioritized newer GPUs to replace capabilities provided by the H20.
China had already begun optimizing mid-range GPUs and advancing domestic alternatives. The strategic approach aims to reduce dependence on foreign chips while mitigating potential security risks associated with imported technology. These measures highlight China’s intent to maintain access to advanced computing capabilities while preserving long-term technological autonomy. The policy shift also signals broader implications for global semiconductor dynamics, setting the stage for coordinated responses by allies and competitors.
Allied Responses: EU, South Korea, and Japan
European authorities have intensified scrutiny of semiconductor exports to China. The European Commission’s 2023 economic security strategy placed AI and related technologies on a risk watchlist and required an assessment of potential security threats. While the EU has not imposed specific export bans on AI chips, officials indicated that curbs on exports or foreign investment could follow depending on the assessment. Coordination with the United States and Japan is under discussion to manage high-end chip access to China.
South Korea, a major memory chip supplier to China, has focused on protecting its domestic industry. The Ministry of Trade emphasized that most Korean exports are outside the scope of U.S. restrictions. South Korean companies obtained Validated End User status for U.S.-origin equipment, which reduced licensing requirements for exports to China. By mid-2025, firms including Samsung and SK Hynix continued sales of memory chips and certain GPUs to Chinese customers, reflecting a balance between trade compliance and market access.
Japan has aligned closely with U.S. policy on semiconductor manufacturing equipment. In 2023, it restricted 23 categories of advanced chip making tools and later extended controls to six additional categories. The measures are intended to prevent the transfer of technology that could enhance China’s advanced chip production. Japanese controls focus on equipment rather than chips themselves, but they contribute to a coordinated strategy to limit sensitive technology access.
Allied approaches highlight a spectrum of responses. Europe emphasizes regulatory oversight and potential curbs, South Korea prioritizes industrial protection, and Japan enforces strict equipment controls. Together, these measures influence the competitive environment for semiconductor supply chains and set the context for how revenue-sharing arrangements will interact with global policy coordination.
Implications for the Global Semiconductor Race
The revenue-sharing model alters the dynamics of U.S.–China technology competition. By allowing AI chips into China for a fee, U.S. firms regained access to a critical market while the Treasury collected a portion of sales comparable to a tariff. Nvidia had projected $5 to $8 billion in lost revenue under a full ban but mitigated part of that loss under the new framework.
For China, the 15 percent levy provides access to midrange GPUs at a lower cost than accelerating domestic development. Authorities emphasize that the H20 is not intended for military or government use and highlight domestic alternatives. The policy demonstrates Beijing’s strategy of balancing immediate technological needs with long-term efforts to reduce reliance on foreign hardware.
The model may set a precedent for other sectors. U.S. officials have suggested export taxes could apply in additional industries. Within the semiconductor sector, uncertainty remains about whether the approach will extend to advanced chips, including Nvidia’s forthcoming Blackwell series.
The episode highlights the strategic tension between short-term market access and long-term autonomy. It also shows how revenue-sharing arrangements can recalibrate global supply chains, influence allied coordination, and serve as a tool to shape competitive behavior in critical technology sectors.
