US Export Controls on AI and Semiconductors: Two Divergent Visions
Introduction
The United States currently stands at a critical strategic crossroads regarding its policy on advanced semiconductor exports to China, particularly those used for artificial-intelligence (AI) systems. The Biden administration implemented unprecedented restrictions on chip exports and semiconductor manufacturing equipment in October 2022,[1] which were subsequently expanded and refined in October 2023[2] and December 2024.[3] These decisions have significantly shaped the technological landscape and set the stage for what may become a defining issue for the new Trump administration.
At the heart of this policy debate is a fundamental question about timelines and predictions: how quickly will transformative AI capabilities develop, and what will China’s indigenous semiconductor capacity look like when they do? The answer to these questions profoundly affect whether current export-control policies will achieve their strategic objectives or potentially backfire.
The case for maintaining or strengthening chip export controls, championed by figures like Anthropic CEO Dario Amodei,[4] rests on the prediction that transformative AI capabilities will emerge relatively soon (within two-to-three years). If this “short-term advancement” scenario holds true, then denying China access to cutting-edge chips could meaningfully diminish their ability to deploy advanced AI systems at-scale, even if they were to be successful in either developing or copying the models. As Nathan Lambert notes, “a large part of export controls, if they work, is just that the amount of AI that can be run in China is going to be much lower.”[5]
Conversely, skeptics of chip-export controls, like Ben Thompson, question the long-term efficacy of such restrictions.[6] If truly significant AI advancements take longer (perhaps 10 or more years), China will likely develop its own chip-manufacturing capabilities during that timeframe. Under this “long-term independence” scenario, the primary effect of existing controls may be to deny revenue to U.S. firms like Nvidia Corp., while accelerating China’s push for technological self-sufficiency. As Dylan Patel observes: “China will win because of these restrictions long-term, unless AI does something in the short-term.”[7] Notably, some observers combine skepticism of export controls on chips with support for strengthening export controls on the equipment needed to manufacture chips.[8]
From a law & economics perspective, the question turns both on what the United States aims to accomplish by restricting chip exports and AI tools, and which future scenario is most likely.
This issue brief does not advocate for a specific approach to export controls. Instead, it aims to illuminate the assumptions, tradeoffs, and considerations that should inform this critical policy choice. By clarifying how differing forecasts about AI development and China’s technological trajectory recommend different optimal policies, we hope to provide the new administration with a framework to align its export-control strategy with its broader technological and geopolitical objectives.
I. Background: The Current Landscape
In recent, competition between the United States and China to develop their respective capacities for AI deployment and semiconductor manufacturing has intensified dramatically. This has been marked by several major developments that illustrated both the potential value of current U.S. export controls, and the limits of that strategy.
In January 2025, the Chinese AI firm Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co. (DeepSeek) released its R1 reasoning model, which appeared to demonstrate capabilities that were competitive with the U.S.-based OpenAI’s o1.[9] What made this announcement particularly significant was that DeepSeek claimed to have accomplished this with substantially less computational resources than American companies typically employ—reportedly training the model for approximately $5-6 million, rather than the billions that some had assumed necessary. While some questioned these figures—noting they excluded research costs, distillation from other models, and human labor—the development nonetheless demonstrated China’s ability to innovate within existing constraints.[10]
This followed the 2023 revelation that Huawei, despite being subject to some of the strictest U.S. controls, had produced its Mate 60 Pro smartphone featuring a domestically manufactured 7nm processor made by SMIC.[11] While this doesn’t match Taiwan’s most advanced nodes (currently at 3nm), it represented a significant achievement for a company operating under significant restrictions. As Ben Thompson observed: “The existence of this chip wasn’t a surprise for those paying close attention: SMIC had made a 7nm chip a year earlier.” Nonetheless, the announcement triggered “overwrought reaction in Washington D.C.,” which led to further restrictions.[12]
The current U.S. export-control regime, established during the Biden administration, operates through several mechanisms. Fundamentally, it restricts the export to China of cutting-edge graphics processing units (GPUs) used for AI applications. The initial controls focused on controlling both interconnection bandwidth and computing performance, as measured by floating-point operations per-second (FLOPS). Later iterations primarily targeted computing performance. Alternate versions of the same chips (such as Nvidia’s H800 and H200) were created specifically for the Chinese market with reduced capabilities (so-called “nerfed” chips) in order to comply with these restrictions.
Beyond chip exports, the controls also restrict semiconductor-manufacturing equipment (SME) needed to produce advanced nodes, with particularly tight controls on extreme-ultraviolet (EUV) lithography machines produced solely by Dutch firm ASML Holding. The restrictions also limit the ability of “U.S. persons” to support or service advanced semiconductor-manufacturing facilities in China, with rules that target both hardware and knowledge transfer.
The export controls invoke the Foreign Direct Product Rule (FDPR) to extend U.S. jurisdiction over foreign-made items that incorporate U.S. technology, or that were manufactured using U.S. equipment. As Gregory Allen notes, the December 2024 update to the U.S. export controls even created “new FDPRs and updated de minimis provisions” that expanded unilateral U.S. authority, potentially capturing “all of the SME made by any company on Earth.”[13]
These restrictions have created a complex set of responses and adaptations. While they have slowed China’s advancement at the cutting edge, they haven’t stopped it entirely. As Allen noted, “SMIC was already producing and selling 7nm chips no later than July 2022 and potentially as early as July 2021, despite having no EUV machines.”[14] But the controls have “dramatically constrained SMIC’s ability to scale up 7nm production,” limiting output to the “low tens of thousands” of wafers monthly, instead of the “hundreds of thousands” originally planned.[15]
The restrictions have also spurred efficiency innovations. DeepSeek, for example, developed advanced techniques to overcome bandwidth limitations in the H800 chips they could legally access, programming “20 of the 132 processing units on each H800 specifically to manage cross-chip communications” by working at a lower programming level than Nvidia’s CUDA (for “Compute Unified Device Architecture”) parallel-computing platform.[16] Such adaptations reflect China’s determination to progress in AI development, despite the constraints imposed by U.S. export controls.
Meanwhile, the global semiconductor landscape remains dominated by Taiwan Semiconductor Manufacturing Co. (TSMC), which produces the most advanced chips worldwide. As Ben Thompson notes, Taiwan’s proximity to mainland China, and China’s longstanding claims on the territory, contribute to a precarious geopolitical situation in which “TSMC’s foundries — and Samsung’s — are within easy reach of Chinese missiles,” which in turn presents “a major issue if you are a U.S. military planner.”[17] This dependence on Taiwan has motivated both U.S. political efforts to onshore chip manufacturing and Chinese ambitions for AI self-sufficiency.
It is against this complex backdrop that it now falls to the Trump administration to reassess U.S. export-control policy, weighing both the existing controls’ demonstrated effects and the potential trajectories for both AI development and Chinese semiconductor capacity in the coming years.
II. Key Decision Factors
A. Timeline Considerations
The most crucial factor in determining optimal export-control policy is predicting the timeline for transformative AI development. These predictions shape whether export controls will secure a meaningful advantage, or simply accelerate China’s push for independence.
1. The short-term AI-advancement scenario
Proponents of the short-term AI-advancement scenario argue that truly transformative AI capabilities are imminent. Anthropic CEO Dario Amodei has suggested that “super powerful AI” could emerge by 2026-27, providing whichever nation possesses it with significant military advantages.[18] This prediction is used to justify maintaining strong export controls to ensure the United States and its democratic allies maintain their lead during this critical window.
Under this scenario, denying China access to cutting-edge chips would effectively limit that nation’s ability to deploy advanced AI models at-scale, even if the underlying technology could be developed or copied. As Dylan Patel notes:
To some extent, training a model does effectively nothing… The thing that Dario [Amodei is] (…) speaking to is the implementation of that model, once trained to then create huge economic growth, huge increases in military capabilities… But that requires a significant amount of compute.[19]
This argument is bolstered by accounts from Chinese AI firms themselves. DeepSeek CEO Liang Wenfeng has admitted that “money has never been the problem for us; bans on shipments of advanced chips are the problem.”[20]
2. Long-term AI-advancement scenario
Those who foresee a longer timeline for transformative AI argue that export controls could be counterproductive. Ben Thompson contends that denying China access to advanced chips primarily serves to sew “the seeds for competition in an industry — chips and semiconductor equipment — over which the U.S. has a dominant position.”[21] In other words, if significant AI breakthroughs take a decade or more, the current restrictions may simply motivate and accelerate China’s development of indigenous chipmaking capabilities.
The timeline debate reflects fundamental uncertainty about the pace and trajectory of AI progress. As Nathan Lambert observes:
if you’re making me give a year, I’m going to be like, “Okay, I have AI CEOs saying this. They’ve been saying two years for a while. (…) I need to take their word seriously, but also understand that they have different incentives.” So I would (…) add a few years to that. Which is how you get something similar to 2030 or a little after 2030..[22]
B. Technical Realities
In addition to the importance of predictions about AI timelines, U.S. export-control policy must also take account of practical technical considerations.
1. Training versus inference compute
Export controls must distinguish between computational power (commonly referred to as “compute”) used for training new AI models and for inference (deploying existing models). While training frontier models requires enormous compute resources, inference (running those models) also demands significant hardware, especially for advanced reasoning capabilities. Nathan Lambert points out that reasoning models like OpenAI’s o1 require especially significant amounts of computational power.[23]
2. China’s optimization innovations
One counterargument to export controls comes from China’s demonstrated ability to optimize AI systems under hardware constraints. DeepSeek’s R1 model achieved capabilities competitive with OpenAI’s o1 despite being forced by U.S. sanctions to use less-powerful H800 GPUs with constrained memory bandwidth.[24] AI investor Nat Friedman poses the question:
is it in fact the case that, if you impose sanctions on China so that they can’t get as much compute, then all you do is give them this constraint to optimize against, which says, ‘How can we squeeze every little bit of IQ out of every flop that we’ve got?’, and they just find clever ways of doing a lot more with a lot less.[25]
But optimization of this kind also produces tradeoffs; a company that focuses on optimization does so at the expense of putting their scarce top engineers on other tasks. It is at least possible that U.S. researchers who do not face such constraints will use their time to develop even better AI applications.
3. Manufacturing challenges beyond EUV
While export controls on EUV-lithography equipment have successfully prevented China from producing the most advanced logic chips, Chinese manufacturers have demonstrated the ability to produce 7nm chips using older deep ultraviolet (DUV) lithography through techniques like multi-patterning. As Gregory Allen notes: “SMIC was already producing and selling 7 nm chips no later than July 2022 and potentially as early as July 2021, despite having no EUV machines.”14 This raises questions about the long-term effectiveness of equipment-focused export controls.
In response, Ben Thompson has argued that “it’s reasonable to assume that [SMIC’s] fab won’t progress further without a Chinese supplier developing” EUV.[26] Thompson advocates strengthening export controls on equipment that would allow China to make state-of-the-art chips, while abandoning export controls on finished chips.[27]
C. Economic Impacts
Export-control decisions carry significant economic consequences that must be weighed against potential security benefits.
1. Effects on US semiconductor companies
The impact of export controls on U.S. semiconductor companies has been debated extensively. Wafer-fabrication-equipment (WFE) suppliers have argued that export controls threaten their business model, with some politicians claiming that the companies face a “death spiral.”[28] The industry-research firm SemiAnalysis counters, however, that “[d]espite a short-term shock or loss of business, the slack is taken up by customers ex-China within a few quarters” and that “the 24 months under export controls have been among the best in history for American WFE suppliers.”[29]
For AI-chip manufacturers like Nvidia, the picture is more complex. Export controls harm Nvidia directly by reducing demand for their chips, but potentially help Nvidia indirectly by making it more difficult for Chinese competitors to develop products that perform as well. As Gregory Allen notes, Nvidia would likely prefer not to be bound by export controls.[30] On the other hand, however, “there are some elements of the new export control package that actually help Nvidia by hurting its Chinese competitors.”[31]
2. Taiwan’s economic security
The semiconductor industry, particularly TSMC, is critically important for Taiwan. As home to the world’s leading advanced-chip manufacturer, Taiwan has historically occupied a unique strategic niche, with its economic value serving as a potential deterrent against Chinese military aggression. U.S. export controls that look to hinder China’s technological advancement thus may paradoxically erode this carefully balanced deterrence mechanism. This shift could fundamentally alter the calculus of potential conflict by making Taiwan less economically indispensable to China, and thereby increasing the risk of military action.
As Ben Thompson notes:
… both China and the U.S. need access to the best chip maker in the world, along with a host of other high-precision pieces of the global electronics supply chain. That means that a hot war, which would almost certainly result in some amount of destruction to these capabilities, would be devastating…one of the risks of cutting China off from TSMC is that the deterrent value of TSMC’s operations is diminished.[32]
There are two mechanisms at play. First, to the extent that Chinese businesses cannot buy TSMC’s products, then a disruption due to war would not worsen their situation. Second, export controls serve to encourage the development of domestic manufacturing capacity in China for state-of-the-art chips. When Chinese manufacturers achieve this goal, then it may no longer be possible to reinstate Chinese economic dependence on Taiwan.
3. Effects on US economy beyond semiconductor manufacturing
The short-term effects of chip export controls on the U.S. economy may be small beyond companies like WFE suppliers and Nvidia. But to the extent that they serve to reduce China’s dependence on Taiwanese manufacturing, the controls may increase the likelihood of a war over Taiwan. A hot war, disrupting all semiconductor manufacturing in Taiwan, would likely have very significant consequences for the U.S. economy. Taiwan manufactures not only the state-of-the-art (leading-edge) semiconductors used in iPhones and Nvidia’s top AI-focused products, but also other commodity (trailing-edge) chips that are indispensable for vast swaths of the modern economy, used in “everything from cars to stereos to refrigerators.”[33]
D. Geopolitical Dimensions
Export controls function within a broader geopolitical context that shapes their implementation and consequences.
1. Allied cooperation challenges
Effective export controls require cooperation from key allies, particularly Japan and the Netherlands, which host critical semiconductor-equipment manufacturers. Gregory Allen notes that: “White House officials have been discussing restrictions on capital equipment with counterparts in the Hague and Tokyo since Biden’s inauguration,” but that “the Netherlands apparently does not yet see eye-to-eye with U.S. assessments on the need to set the threshold at 16/14 nm or smaller for logic chips.”[34]
This cooperation challenge reflects differing economic interests and threat perceptions. In 2023, 29% of Dutch firm ASML’s sales were to customers in China, creating a strong disincentive to further restrict exports.[35]
2. Taiwan’s vulnerability
As noted in the previous section on economic impacts, Taiwan’s geopolitical vulnerability represents perhaps the most critical consideration in export-control policy. If export controls reduce China’s dependence on Taiwan, while maintaining Taiwan’s importance to the United States, this could create dangerous incentives for Chinese action. With broad and effective U.S.-imposed export controls on chips, China doesn’t risk disrupting their semiconductor supply chain by attacking Taiwan. On the other hand, a war in Taiwan would likely create massive disruptions for the U.S. economy due to the level of U.S. dependence on Taiwan-made chips.
III. Conclusion
The debate surrounding AI-related U.S. export controls reflects a fundamental uncertainty about technological trajectories. While the actual pace of AI advancement will ultimately determine policy efficacy, decisionmakers must act based on incomplete information. This brief has outlined two plausible scenarios—one in which transformative AI capabilities emerge rapidly and another in which they develop more gradually—and described how each suggests different optimal approaches to export controls.
For policymakers navigating this uncertainty, several principles warrant consideration, regardless which scenario materializes. First, export controls should be designed with sufficient flexibility to adapt as the technological landscape evolves. Static policies risk becoming either irrelevant or counterproductive as conditions change. Second, policy effectiveness depends heavily on multilateral cooperation. Unilateral actions by the United States face significant limitations, as demonstrated by China’s ability to procure restricted technologies through third-party countries. Third, the economic impacts of export controls extend beyond immediate revenue considerations to long-term market positioning and technological leadership. This includes the possibility that they will make it likelier that China will attack Taiwan, and the disruption to the U.S. economy that would entail.
Ultimately, the export-control decisions made in 2025 will reflect implicit forecasts about AI development trajectories. By acknowledging these forecasts explicitly and establishing clear metrics to evaluate them, policymakers can create more resilient policies capable of adaptation as technological realities unfold. Whichever approach the administration pursues, it should be implemented with clear objectives, regular reassessment mechanisms, and recognition of the inherent uncertainties in predicting technological futures.
[1] Martijn Rasser & Kevin Wolf, The Right Time for Chip Export Controls, Lawfare (Dec. 13, 2022), https://www.lawfaremedia.org/article/right-time-chip-export-controls.
[2] Gregory Allen, Understanding the Biden Administration’s Updated Export Controls, Cent. Strateg. Int. Stud. (Dec. 11, 2024), https://www.csis.org/analysis/understanding-biden-administrations-updated-export-controls.
[3] Id.
[4] Dario Amodei, On DeepSeek and Export Controls, Dario Amodei (Jan. 2025), https://darioamodei.com/on-deepseek-and-export-controls.
[5] Lex Fridman, Dylan Patel, & Nathan Lambert, DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters, Lex Fridman Podcast (Feb. 3, 2025), https://lexfridman.com/deepseek-dylan-patel-nathan-lambert-transcript.
[6] See, e.g., Ben Thompson, DeepSeek FAQ, Stratechery (Jan. 27, 2025), https://stratechery.com/2025/deepseek-faq.
[7] Fridman, Patel, & Lambert, supra note 5.
[8] Ben Thompson, AI Promise and Chip Precariousness, Stratechery (Feb. 25, 2025), https://stratechery.com/2025/ai-promise-and-chip-precariousness.
[9] Thompson, supra note 6.
[10] Id.
[11] Id.
[12] Id.
[13] Allen, supra note 2.
[14] Id.
[15] Id.
[16] Thompson, supra note 6.
[17] Ben Thompson, Chips and Geopolitics, Stratechery (May 19, 2020), https://stratechery.com/2020/chips-and-geopolitics.
[18] Amodei, supra note 4.
[19] Fridman, Patel, & Lambert, supra note 5.
[20] Jordan Schneider et al., Deepseek: The Quiet Giant Leading China’s AI Race, ChinaTalk (Nov. 27, 2024), https://www.chinatalk.media/p/deepseek-ceo-interview-with-chinas; see also Allen, supra note 2.
[21] Thompson, supra note 6.
[22] Fridman, Patel, & Lambert, supra note 5.
[23] Id.
[24] Thompson, supra note 6.
[25] Ben Thompson, An Interview with Daniel Gross and Nat Friedman About Models, Margins, and Moats, Stratechery (Jan. 23, 2025), https://stratechery.com/2025/an-interview-with-daniel-gross-and-nat-friedman-about-models-margins-and-moats.
[26] Thompson, supra note 8.
[27] Id.
[28] Dylan Patel, Jeff Koch, & Sravan Kundojjala, Fab Whack-A-Mole: Chinese Companies Are Evading U.S. Sanctions, SemiAnalysis (Oct. 28, 2024), https://www.semianalysis.com/p/fab-whack-a-mole-chinese-companies.
[29] Id.
[30] Allen, supra note 2.
[31] Id.
[32] Ben Thompson, Taiwan and Tech’s Geopolitical Realities, ARM on Mac?, TSMC’s Choice, Stratechery (May 11, 2020), https://stratechery.com/2020/taiwan-and-techs-geopolitical-realities-arm-on-mac-tsmcs-choice.
[33] See Thompson, supra note 8.
[34] Allen, supra note 2.
[35] Toby Sterling, ASML Expects US, Dutch Export Rules to Hit China Sales by 10-15%, Reuters (Jan. 24, 2024), https://www.reuters.com/technology/asml-expects-us-dutch-export-rules-hit-china-sales-by-10-15-2024-01-24.