Exploiting the DUVi Loophole for Advanced Manufacturing
The architecture of the United States export controls, designed to stifle China’s advancement in artificial intelligence and semiconductor manufacturing, is facing a critical vulnerability. A comprehensive report by the American Enterprise Institute (AEI) details how Chinese chipmakers are successfully exploiting a significant loophole in the regulations enacted by the Biden administration in 2022. While the US and its allies have effectively denied China access to the extreme ultraviolet (EUV) lithography machines necessary for the most advanced chip production, they left a critical gap regarding older-generation equipment.
Chinese entities, spearheaded by Semiconductor Manufacturing International Corporation (SMIC) and its primary customer, Huawei, are aggressively acquiring deep ultraviolet immersion (DUVi) lithography tools. Rather than using these machines for their intended legacy applications, Chinese engineers are repurposing and upgrading them to fabricate near-frontier chips. This process, while highly inefficient and punishingly expensive compared to EUV manufacturing, is allowing China to scale its domestic production of advanced semiconductors far faster than US policymakers had anticipated, effectively circumventing the intent of the export controls.
SMIC’s Progress and the Threat to the US Compute Monopoly
The scale of this workaround is massive. According to the AEI report, Chinese entities accounted for a staggering 70 percent of ASML’s DUVi tool sales in 2024. This influx of equipment has enabled SMIC to produce and sell chips that are increasingly competitive with the US-led technological frontier. The report notes that SMIC has already achieved significant milestones, including the successful production of 7-nanometer-class processors that power Huawei’s latest smartphones, demonstrating a capability that rivals chips released by TSMC in 2018.
More alarmingly for US strategic interests, SMIC is actively developing a 5-nanometer process using these repurposed DUVi machines. While this process is undoubtedly much more expensive and yields fewer usable chips than the TSMC equivalent, it provides Huawei with a sanction-proof pathway to manufacturing near-frontier AI accelerators at scale. The AEI report warns that China’s fleet of several hundred DUV lithography machines will almost certainly be capable of printing millions of high-end logic dies by 2026, which Huawei will then package into advanced AI accelerators, directly challenging the United States’ near monopoly on the compute needed to win the global AI race.
The Inefficiency of the Process vs. Strategic Imperative
The economic realities of using DUVi for advanced nodes are stark. The process requires multiple patterning steps, significantly increasing the time, complexity, and cost of manufacturing each wafer, while simultaneously driving down yield rates. In a purely commercial, globally integrated market, this approach would be financially ruinous and uncompetitive. However, the AEI report emphasizes that Beijing is not operating under standard market constraints. The strategic imperative of achieving semiconductor self-sufficiency far outweighs the economic inefficiencies of the DUVi workaround.
For the Chinese government, the cost curve is a secondary concern compared to the national security and economic risks of relying on foreign-controlled technology. The massive state subsidies and domestic demand from companies like Huawei provide SMIC with the financial runway to absorb the high costs of this inefficient manufacturing process. This dynamic highlights a fundamental miscalculation in the initial design of the US export controls, which assumed that the economic penalties of using older technology would naturally deter China from pursuing advanced chip production.
Recommendations for Closing the Lithography Gap
To address this critical vulnerability, the AEI report outlines three complementary policy recommendations aimed at closing the DUVi loophole. First, it advocates for a shift toward capability-based export controls. Instead of regulating lithography tools based on specific performance thresholds or what buyers declare they will be used for, the controls should focus on what the machines are physically capable of producing when pushed to their limits or upgraded.
Second, the report recommends implementing a countrywide presumption of denial for all lithography exports to China. This would effectively halt the flow of even older-generation machines, recognizing that any addition to China’s manufacturing capacity can be leveraged for advanced production. Finally, and perhaps most contentiously, the report calls for restrictions on ASML’s ability to service and maintain the DUVi machines already deployed in Chinese fabrication plants. This measure aims to degrade the operational efficiency of China’s existing fleet, further increasing the friction and cost of their workaround strategies.
The Future of the US-China Semiconductor Rivalry
The exploitation of the DUVi loophole underscores the immense difficulty of containing technological diffusion in a highly interconnected global industry. The US strategy of “chokepoint” controls is being actively subverted by Chinese ingenuity and massive state investment. If left unaddressed, the AEI report warns, this loophole risks undermining the entire architecture of US export controls on AI chips.
The situation presents a complex challenge for US policymakers. Tightening controls further, particularly by restricting the servicing of existing machines, risks escalating trade tensions and potentially alienating key allies like the Netherlands, home to ASML. However, failing to act allows China to continue building a robust, albeit inefficient, domestic semiconductor ecosystem capable of fueling its AI ambitions. EastFrontier has previously reported how China’s AI chip self-sufficiency is already at 41%. The resolution of this issue will be a defining factor in the ongoing technological rivalry between the two superpowers, determining who controls the foundational hardware of the artificial intelligence era.