Photo: liangpv via Getty Images
Since October 2022, the United States has devoted significant resources to restricting China’s access to artificial intelligence (AI) and advanced semiconductor technologies. In the final months of the Biden administration, the Department of Commerce issued four additional far-reaching export control updates. On December 2, 2024, it released two rules that added 140 companies to the Entity List, expanded the scope of the Foreign Direct Product Rule (FDPR), and restricted new technology areas such as high-bandwidth memory, among other measures. In the second week of January 2025, the Department of Commerce issued the AI Diffusion Framework and the Foundry Due Diligence Rule, further shaping the spread of AI and semiconductor technologies throughout the world. Export controls remain front and center for the second Trump administration, which directed an effort to “identify and eliminate loopholes in existing export controls—especially those that enable the transfer of strategic goods, software, services, and technology . . . to strategic rivals and their proxies” on its first day in office.
However, countries like the Netherlands, Germany, South Korea, Japan, and Taiwan continue to control key chokepoints in the AI and semiconductor value chain, making unilateral action only so effective. Furthermore, the existing multilateral export control architecture is neither sufficiently flexible nor fast to allow for the kind of sophisticated, targeted controls that the United States has levied on China. The success or failure of the U.S. export control strategy is thus dependent on its allies’ ability to implement controls outside of this traditional architecture or U.S. extraterritorial regulations covering allies.
This paper provides an in-depth analysis of U.S. allies’ export control authorities related to AI and semiconductor technologies and does the same analysis for China. It demonstrates that U.S. allies often do not have equivalents to U.S. export control authorities and tools like the FDPR and Entity List, but that they generally do have the capability to introduce some controls on advanced semiconductor chips and related equipment not covered by multilateral export control regimes. As a result, lack of alignment with the U.S. export control regime cannot necessarily be attributed to a lack of authorities alone. Allies’ enforcement capacity and willingness to act are also key ingredients in the implementation of effective export controls and are crucial to the success of U.S. and allied technology competition with China. Accordingly, the recommendations section of this paper addresses each of these three elements.
The paper proceeds as follows. First, it identifies key export control authorities used by the United States to slow the progress of China’s AI and semiconductor industries and analyzes which other countries possess these authorities and are thus capable of implementing similar controls. It then surveys the export control policies of key actors, including the European Union, the Netherlands, Germany, Japan, South Korea, Taiwan, and China. It concludes by offering recommendations to U.S. and allied policymakers to make AI and semiconductor export controls more effective.
In December 2023, then–Secretary of Commerce Gina Raimondo clearly outlined the U.S. strategy for China and its AI ecosystem when she said, “America leads the world in artificial intelligence. America leads the world in advanced semiconductor design, period. . . . We’re a couple years ahead of China. No way are we going to let them catch up.” To achieve this goal, the U.S. national security enterprise has undertaken a massive, multifaceted effort to choke off China’s access to cutting-edge AI and related technologies.
Export controls on the high-performance semiconductors used to train and inference state-of-the-art AI models have been at the forefront of this effort so far. On October 7, 2022, the United States enacted a comprehensive set of export controls on advanced semiconductor technologies. Although these controls marked the reversal of nearly 30 years of trade policy, they did not achieve all of their intended goals. U.S. chip designers like Nvidia continued to provide Chinese customers with slightly lower-performance chips, which according to the U.S. Department of Commerce’s Bureau of Industry and Security (BIS), provided “nearly comparable AI model training capability.” In October 2023, the United States updated these controls to cover a much larger set of chips and semiconductor manufacturing equipment (SME). As noted in a previous CSIS report, “The United States is firmly focused on retaining control over so-called ‘chokepoint’ . . . technologies in the global semiconductor technology supply chain.” The United States issued major export control updates in December 2024 and January 2025 that continued in this vein.
Despite the significant resources successive administrations have devoted to restricting China’s access to AI and semiconductor technologies, China’s AI ecosystem remains competitive with the United States’. In May 2024, Chinese AI company DeepSeek released a best-in-class open-weight model reportedly trained on Nvidia A100 chips stockpiled before the October 2022 controls went into effect. The even more impressive Deepseek-R1 reasoning model was trained on Nvidia H800 chips, which were not restricted before October 2023. The performance of DeepSeek-R1 was so far beyond what the financial analyst community had expected a Chinese firm to be capable of that the tech-heavy Nasdaq dropped 3.1 percent in one day as a result of its release. This previous CSIS paper provides an in-depth analysis of DeepSeek and its implications for U.S. export controls.
At the core of the U.S. chip war strategy is a group of iteratively developed and highly sophisticated export control tools. Authorities granted in the Export Control Reform Act of 2018 (ECRA), Export Administration Regulations (EAR), and executive orders offer the U.S national security community a range of options to control the export, reexport, transit, or transfer of select military, dual-use, and purely commercial items and technologies. In the House Committee on Foreign Affairs markup of ECRA, then–committee chairman Congressman Edward Royce stated that ECRA “closes gaps in our export controls that could permit transfers of cutting-edge technology like artificial intelligence and advanced semiconductors to potential adversaries such as Beijing.” More accurately, ECRA represented the codification of U.S. export control powers that had previously relied on authorities in the International Emergency Economic Powers Act (IEEPA) and had a novel focus on emerging technologies.
The U.S. export control regime divides the responsibilities for administering and enforcing export controls on dual-use and military goods and technologies. The Department of Commerce is responsible for regulating the flow of dual-use technologies and less sensitive military items, which covers advanced semiconductors and related technologies, through the BIS. Meanwhile, the Directorate of Defense Trade Controls (DDTC) within the U.S. Department of State oversees and licenses weapons and munitions that are covered by the International Traffic in Arms Regulations (ITAR). The Office of Foreign Assets Control (OFAC) at the Treasury Department oversees restrictions on exports based on financial sanctions.
The four kinds of U.S. export controls most relevant to the Trump and Biden administrations’ efforts to choke off China’s access to AI are described below.
In addition to these four export control types, there are three further export control features that require explanation to understand U.S. AI and semiconductor export controls.
The United States has also employed a range of other economic security tools to restrict China’s access to advanced semiconductor technologies. For example, in August 2023 the Biden administration used IEEPA to issue an executive order directing the U.S. Treasury Department to establish a program to review outbound investments in national critical sectors such as AI and semiconductors. This measure followed reporting that U.S. firms were investing in Chinese AI companies that were on the Entity List. On October 28, 2024, the U.S. Treasury Department issued its final rule on the implementation of the executive order, which went into effect on January 2, 2025, but is now subject to review as a result of the Trump administration’s America First Trade Policy memorandum.
However vigorously the United States pursues its own export control strategy, it will be unable to achieve success without changing the behavior of other countries. While the United States holds a strong position in certain parts of the semiconductor value chain like chip design, other countries such as the Netherlands and Japan also play crucial roles in areas like SME. For example, Dutch company ASML is the sole provider of the latest generation of extreme ultraviolet (EUV) lithography machines needed to make cutting-edge AI chips. The consulting firm Accenture estimated that the inputs to a typical semiconductor chip “could cross international borders approximately 70 or more times before finally making it to the end customer.” In other words, the United States does not control all the relevant chokepoints required to execute its export control strategy. In comments after the October 2022 controls went into effect, one U.S. official told Reuters, “We recognize that the unilateral controls that we are putting in place will lose effectiveness over time if other countries don’t join us.” At the December 2024 Reagan National Defense Forum, Gina Raimondo reiterated this sentiment, albeit more forcefully, saying, “When I set the rules, I have to make damn sure China can’t just buy this stuff from Japan or Korea or the Europeans, so that’s why we have to work with them.”
Fully aware of this reality, Chinese officials have repeatedly threatened countries considering going along with the U.S. controls. For example, in a March 2022 interview, Tan Jian, China’s ambassador to the Netherlands, said, “This will not be without consequences. I’m not going to speculate on countermeasures, but China won’t just swallow this,” in response to the Netherlands’s increasing alignment with the United States on export controls.1
Unfortunately, U.S. allies have not kept pace with the United States’ increasingly restrictive export controls vis-à-vis China. While the U.S. export control system is designed to be able to keep up with technological developments and to be responsive to changing geopolitical circumstances, the regimes of its allies are generally not. As noted in the previous section, other countries with significant semiconductor industries have export control systems grounded in multilateral agreements primarily focused on WMD nonproliferation. These arrangements are significantly less agile than the U.S. system. For example, to add items to a Wassenaar Arrangement control list requires consensus from all participating countries, and opportunities for additions typically occur only once per year. As Wassenaar also includes Russia, which has obstructionist geopolitical goals, finding consensus on issues with geopolitical implications is unfeasible. In total, it can take up to three years to add a new item to Wassenaar control lists, even under favorable circumstances. Furthermore, Wassenaar explicitly prohibits controls targeting an individual country, noting in its founding document: “This arrangement will not be directed against any state or group of states.” Limited improvements have been made to export control coordination between U.S. allies, such as the informal “Wassenaar minus one” approach, an effort by like-minded countries to harmonize controls on items and technologies not listed in traditional multilateral agreements.
But the existing multilateral export control architecture remains inadequate for U.S. national security requirements. As a result, understanding allied countries’ unilateral export control authorities is crucial to the success or failure of the U.S. and allied technology competition with China.
Allied countries possess significantly less comprehensive export control authorities than the United States. For the most part, they do not have an equivalent to key parts of the U.S. AI and semiconductor export control tool kit discussed in the previous sections like the FDPR, U.S. Persons Rule, Entity List, and restrictions that apply on a country-wide basis. This stands in radical contrast to China, which has passed no fewer than five laws since 2020 devoted to building a more sophisticated arsenal of economic security tools. In fact, China’s most recent update to its Export Control Law replicated several key U.S. export control authorities. The following sections compare U.S., allied, and Chinese export control authorities across list-based, end-user, end-use, and services controls.
List-Based Controls
All U.S. allies implement the list-based controls found in the traditional four multilateral export control regimes. However, as shown in Table 1, they generally do have the ability to implement list-based controls outside of those regimes as well. For example, while the European Union does not have the power to require the imposition of export controls on semiconductor technologies, under Article 9 of 2021/821, EU member states are empowered to implement unilateral controls on otherwise uncontrolled items for “reasons of public security, including the prevention of acts of terrorism,” or for “human rights considerations.” The Netherlands specifically referenced Article 9 authorities when it issued its Regulation on Advanced Production Equipment for Semiconductors in June 2023, which introduced a license requirement on the export of previously uncontrolled SME from the country. Similarly, when the Japanese government revealed its plans to place license requirements on 23 types of SME in 2023, it did so with authorities granted under its Foreign Exchange and Foreign Trade Act that was originally passed in 1949.
Although allied countries do have the demonstrated ability to implement list-based controls unilaterally, it is misleading to think of their controls as equivalent to those issued by the United States. For example, no allied country has a meaningful equivalent to the FDPR, nor do they have controls on advanced node semiconductors that apply on a China-wide basis, with the partial exception of Taiwan. On the other hand, recent changes to China’s export control system have given Beijing the authority to implement list-based controls that apply on both a country-specific and extraterritorial basis.
End-User Controls
Multilateral export control regimes do not have an equivalent to the U.S. Entity List. As a result, allies have far less robust end-user restrictions than the United States. For example, the European Union and its member states simply do not have end-user restrictions and are forced to use sanctions authorities to impose controls on specific entities. Table 2 summarizes the end-user control landscape.
While Japan, South Korea, and Taiwan all have the authority to impose end-user restrictions, their end-user authorities are more limited than the U.S. Entity List. For example, Japan’s end-user list only contains entities involved in the development, production, manufacturing, or storage of WMDs, or certain military end-uses. No allied country has put end-user controls on Chinese AI and semiconductor firms like the United States has, at least not publicly. As with list-based controls, allied countries lag far behind China when it comes to end-user controls. For example, in October 2024, China released its new dual-use export control regulation, which outlined a U.S. Entity List equivalent. Other regulations such as the Unreliable Entities List also give China options to target specific actors.
End-Use Controls
Multilateral export control regimes contain end-use (often referred to as “catch-all”) controls related to WMDs and certain military applications. For example, the Missile Technology Control Regime requires members to have end-use restrictions “controlling the export of items not included on a control list when they may be intended for use in connection with delivery systems for WMD[s].” Meanwhile, the Wassenaar Arrangement contains a “catch-all” provision designed to control unlisted dual-use items destined for a country subject to an arms embargo and intended for “military end-use.” As shown in Table 3, allied countries’ end-use controls are thus confined to WMD and military applications, with the potential exception of the European Union.2
In contrast, in October 2022, BIS introduced an end-use restriction related to the production of advanced node semiconductors in China. This move marked a major policy change, as U.S. end-use restrictions had previously been limited to situations involving WMDs. China also has the authority to implement end-use controls for reasons unrelated to WMDs. China’s 2020 Export Control Law provides a catch-all for any end-use that likely would endanger “[China’s] national security or national interests.” These terms are left largely undefined, providing Beijing with the ability to require licenses for a very broad set of unlisted items on an end-use basis if it desires.
Services Controls
Allied countries surveyed in this paper only apply services controls for WMD end-uses. However, there is a key difference in how the United States interprets what is covered under WMDs compared to its allies. In October 2022, BIS informed all U.S. persons that the act of supporting the development or production of advanced node semiconductors and related equipment in China “could involve ‘support’ for . . . weapons of mass destruction-related end uses,” and thus required a license. On the other hand, allied countries do not consider activities supporting the development or production of advanced node semiconductors to be related to WMDs and, as a result, do not imposes services controls on such activities. Table 4 summarizes the services controls landscape.
Despite holding a more limited view of WMD-based export control restrictions than BIS, certain allies do restrict their citizens’ activities related to semiconductor manufacturing through other economic security tools. For example, according to a report by the Semiconductor Industry Association, South Korea has used its Act on Prevention of Divulgence and Protection of Industrial Technology “on numerous occasions to prevent firms transferring to mainland China technology deemed detrimental to the industrial competitive capabilities of Korean companies even in the absence of direct links to national security concerns associated with WMDs or conventional weapons.” Taiwan also restricts its citizens and firms from supporting the Chinese semiconductor industry through a variety of regulations unrelated to its export control regime.
Key Gaps Between U.S. and Allied Export Control Regimes
As demonstrated in the previous sections, allied countries possess far less expansive export control authorities than the United States. Although they are all capable of implementing list-based controls on advanced AI chips and SME, their export control tool kit is far narrower than what is available to BIS. For example, while the list-based controls announced by the Netherlands and Japan as part of a reported trilateral deal with the United States were a legitimate breakthrough in allied export control coordination, their controls were weaker than the U.S. regime in at least three important ways. Former Assistant Secretary of Commerce for Export Administration Kevin Wolf summed up these defects in his February 2023 testimony before the Senate Committee on Banking, Housing, and Urban Affairs, which is quoted below.
[The trilateral deal does not impose] controls on activities of Dutch or Japanese citizens in support of advanced node manufacturing in China. (The new BIS rules prohibit U.S. persons from providing support, even involving uncontrolled foreign-made items, to the development or production of advanced node semiconductors in China.) It will not involve any ally imposing end-user controls such as those related to the EAR’s Entity List. It will also not have controls specific to the . . . production of semiconductor production equipment in China. Thus, Japanese and Dutch companies will still be able to export to China items and services that U.S. competitor companies cannot.
It is worth adding to Wolf’s point that license applications for sales of U.S.-controlled chips to China generally face “a presumption of denial.” Thus, the U.S. policy imposes what is de jure a license requirement but de facto a ban. This is not always the case for U.S. allies. For example, in July 2023 Reuters reported that Japan’s controls would not be implemented on a presumption of denial basis.
Only imposing simple list-based restrictions leaves allied countries vulnerable to the same failure modes that plagued earlier versions of the U.S. controls. For example, in July 2019, Japan issued an update to its export licensing policies, which removed South Korea from its “white countries” list of trusted trade partners and required individual export licenses for fluorinated polyimide, hydrogen fluoride, and related chemicals. A February 2023 World Bank analysis showed that in the absence of an FDPR equivalent, Japanese firms simply relocated production of the controlled chemicals from Japan to subsidiaries located in South Korea. While technically legal, these activities were in stark opposition to the intent of Japan’s controls and greatly limited their strategic impact. Similarly, for years the United States blocked direct sales of cutting-edge chips to customers explicitly affiliated with the Chinese military, while allowing those same chips to flow freely to commercial entities in China. While this policy ended direct sales from U.S. companies to the Chinese military, it was mostly ineffective at stopping indirect sales to the shell companies that helped the Chinese military evade export controls. The October 2022 controls addressed this by implementing restrictions that applied on a China-wide basis. Without equivalent measures, there are good reasons to think that allied countries’ controls will be susceptible to the same diversion risk.
The fact that the Netherlands and Japan used existing authorities to implement unilateral controls on advanced semiconductor technologies also illustrates that analysis of authorities alone yields an incomplete understanding of international semiconductor export control dynamics. Whether Japan and the Netherlands’s reticence can be attributed to poor diplomatic coordination on behalf of the United States or a lack of political will on behalf of its allies is up for debate, but it cannot be entirely attributed to insufficient legal authorities. What is clear is that this delay harmed the U.S. strategic goals vis-à-vis China and its AI and semiconductor ecosystem.
The Dutch and Japanese governments announced their plans to impose controls in line with the October 2022 update in March 2023, but enforcement did not take place until July 2023 in the case of Japan and September 2023 in the case of the Netherlands. This delay allowed Chinese firms to engage in a massive stockpiling effort. According to a Financial Times analysis of Chinese customs data, the total value of Chinese imports of SME increased from $2.9 billion in June and July 2022 to $5 billion over the same two months in 2023. The analysis further found that “most of the imports came from the Netherlands and Japan.”
The December 2024 export control update showed that while U.S. policymakers are providing incentives to allies to align themselves with the U.S. export control regime, they are willing to unilaterally restrict the activity of firms headquartered in unaligned countries. For example, South Korean SME and high-bandwidth memory will be restricted by the expanded FDPR even for sales from South Korea, with a possible future exemption if the country implements controls equivalent to those in the United States.
The following sections survey export control policies for the European Union, Netherlands, Germany, Japan, South Korea, Taiwan, and China.
Gregory C. Allen is the director of the Wadhwani AI Center at the Center for Strategic and International Studies (CSIS) in Washington, D.C. Isaac Goldston is a research associate with the Wadhwani AI Center.
The authors would like to thank former interns Irena Petryk, Samantha Gonzalez, and Teddy Foley for their research assistance. The authors would also like to thank Sadie McCullough, William Reinsch, Jack Chang, Jack Hung, Wonho Yeon, Jeremy Chih-Cheng Chang, Sebastian Bennick, and Kevin Wolf, Chae Soohong, as well as a variety of government officials who wish to remain anonymous for their helpful feedback on earlier drafts of this paper.
This report is made possible by generous support from the RAND Corporation.
The views expressed in this document are those of the authors and do not necessarily reflect RAND opinion.
Please consult PDF for full report and references.
