Stop Selling the Rope

Executive Summary

The United States and its allies hold decisive advantages in artificial intelligence semiconductor design and manufacturing—two of the few critical technology sectors where American leadership remains intact. This paper presents an unequivocal America First framework for growing and maintaining an American advantage vital to both economic competitiveness and national security: prohibit all sales of advanced AI chips and semiconductor manufacturing equipment to China. 

China’s domestic AI chip production remains severely constrained, with chips lagging U.S. technology by at least one generation. Recent procurement data has identified $16 billion in Chinese orders for American chips, and the People’s Liberation Army has identified AI as central to achieving military modernization and advantage. The dual-use nature of AI technology and China’s Military-Civil Fusion doctrine make civilian/military distinctions effectively meaningless.

Since China’s accession to the WTO, wave upon wave of western corporations have surged into the Chinese market, hoping to gain a foothold that could translate into long-term growth and profits. The result has always been the same: technology transfer, rapid gains by CCP-favored indigenous firms, loss of leadership. The United States cannot allow this to happen in AI.

This paper recommends five measures:

1. Prohibit the sale of advanced AI chips to PRC-based entities
2. Prohibit the sale of advanced semiconductor manufacturing equipment to PRC-based entities 
3. Increase enforcement resources at the Bureau of Industry and Security, whose current budget is dwarfed by documented smuggling profits
4. Mandate location verification for controlled semiconductors and semiconductor manufacturing equipment
5. Establish secure channels for reporting export violations

Introduction

The United States today faces a choice: learn the lessons of the past quarter-century, or trade away American leadership in a vital future technology with the hope that following the same playbook will somehow yield a different outcome.

Over the past 25 years, the U.S. has surrendered sector after sector to China in a misguided pursuit of the promises of globalization. From solar panels and electric vehicles to telecommunications to rare earths, China now has effective dominance, leaving American supply chains vulnerable to severe disruption, American producers with little market share, and American workers with fewer opportunities and lower wages.

The design of advanced semiconductors, especially the GPUs used to train and run AI models, remains one domain where the United States maintains a decisive global advantage. Firms from the U.S. and allied nations, most notably Japan, the Netherlands, and Taiwan, as well as Germany and South Korea, also maintain a decisive advantage in the machinery and facilities needed to produce those chips. As a result, China is currently lagging well behind the U.S. in its ability to produce leading-edge “compute”—the computational power that allows for the training and running of AI models. 

China already has the motive to close that gap. By selling either chips or chipmaking equipment to the CCP, the U.S. is providing the means. If the U.S. chooses an America First chip export strategy, by contrast, our leaders have the opportunity to establish definitive American leadership in advanced AI. Unfortunately, the same arguments made by business leaders decades ago, which led to trading away past technological leadership in critical sectors, are being echoed today in this sector: America needs to “addict” China to American technology. If U.S. firms gain a toehold in China, their technologies will become the Chinese standard. But if we stay out of China, the Chinese will quickly develop alternatives. 

These arguments were bad in 1999. They are unforgivable in 2025, when everyone knows that China grants access to its market only for as long as it takes to steal technology, transfer know-how, and then push foreign firms out. Rather than transfer its advantage and enrich an adversary, the United States should maintain the advantage and transform it into era-defining leadership. 

Failure to act poses a severe risk to U.S. national security and economic competitiveness.

Any export of advanced AI chips or chipmaking equipment to China must be understood as supporting its military, either by empowering its leading AI labs or by directly providing computing power to the People’s Liberation Army (PLA). Ultimately, the dual-use nature of AI makes civilian/military distinctions meaningless—a reality the CCP will exploit through its Military-Civil Fusion approach to critical technology.¹U.S. Department of State, U.S. Technology in the Military-Civil Fusion Strategy Revision, August 27, 2025. The same AI capabilities used in commercial markets to optimize supply chains can be used to coordinate military logistics; the same optical systems used to inspect factories can guide weapons.

With trillions of dollars of potential value on the table, from both sales of AI chips and AI-driven productivity gains across sectors, the economic stakes are also extremely high. At best, selling advanced AI chips to China would mean accelerating China’s economic progress and strengthening its competitiveness relative to America’s. At worst, it would mean selling the Chinese Communist Party the rope with which to hang the United States, its allies, and the democratic world broadly, trading away our hard-earned advantage for short-sighted, short-term gains today.

The United States must act quickly to assert an America First chip export strategy. The guiding principle of a truly America First strategy is simple, sensible, and urgent: prohibit the sale of advanced AI chips and manufacturing equipment to China, working with allies to prioritize American and allied control of advanced AI capacity.

China’s Disadvantage, America’s Window to Act

The United States and its close allies dominate the advanced AI chip supply chain. Given the technological complexity of the design and manufacturing processes, China is and will remain reliant on these suppliers for these chips for years to come. 

The U.S. directly controls the design stage of the AI chip supply chain, with three American companies accounting for over 75% of advanced chip design.²TrendForce, “Global Top 10 IC Design Houses See 49% YoY Growth in 2024, NVIDIA Commands Half the Market,” March 17, 2025. U.S. partners control the manufacturing process, particularly Taiwan: TSMC manufactures 80–90% of sub-7nm chips (mainly in Taiwan, though with increasing output in Arizona).³TrendForce, “TSMC’s Advanced Processes Remain Resilient Amid Challenges,” April 8, 2024. Two of the three companies that produce high-bandwidth memory (HBM) are Korean (Samsung and SK Hynix) and one is American (Micron).⁴Select Committee on the Chinese Communist Party, Selling the Forges of the Future, October 7, 2025; Jeongku Choi and MS Hwang, “Samsung’s Q2 2025 Memory Performance Disappoints but Signals H2 Recovery,” Counterpoint Research, July 31, 2025.

The U.S. and a few key allies together control the production of the semiconductor manufacturing equipment (SME) required to conduct that manufacturing. The Dutch company ASML is the sole producer of the extreme ultraviolet (EUV) lithography machines needed to make the most advanced chips. According to the U.S. government, “the United States, Japan, and the Netherlands account for 90 percent of semiconductor manufacturing equipment market share.”⁵U.S. Department of Commerce, Bureau of Industry and Security. Assessment of the Status of the Microelectronics Industrial Base in the United States. December 21, 2023. “This is largely on the strength of several key companies: U.S.-based Applied Materials, Lam Research, and KLA Corporation, Japan-based Tokyo Electron, and Netherlands-based ASML.”  Private-sector sources concur:

U.S.-headquartered companies lead in design and core IP, and command almost half the global market share in semiconductor manufacturing equipment (SME). Most of the remaining global market share for SME is in allied countries, including the Netherlands and Japan, whose companies conduct significant manufacturing and R&D in the United States.⁶Semiconductor Industry Association, 2025 State of the U.S. Semiconductor Industry, July 10, 2025.

This concentration leaves China extremely dependent on U.S., Japanese, and Dutch SME and makes effective control possible—if the U.S. acts now and denies China’s attempts to use access to U.S. technology to steal the advantage.⁷Select Committee on the Chinese Communist Party, Selling the Forges of the Future.

With respect to SME, the United States already imposes various export restrictions on sales to China, as do key SME-producing allies like Japan and the Netherlands. These restrictions have successfully impeded China’s advanced production capabilities. However, as detailed in a recent investigation by the House Select Committee on the Strategic Competition between the United States and the Chinese Communist Party, this control regime has significant gaps.⁸Ibid. The Select Committee finds that “five restricted (i.e., on the U.S. Government’s Chinese military company list, investment restriction list, or Entity List) semiconductor companies in China were also top customers of each Toolmaker [meaning the key U.S., Japanese, and Dutch SME producers] from 2022 to 2024.” For four of these producers, currently restricted Chinese companies accounted for 45% of combined revenue between 2022 and 2025, demonstrating the appetite of Chinese purchasers—a majority of which are state-owned enterprises—for whatever SME is available.⁹Ibid.

The U.S. and its allies apply some countrywide export controls on SME to China, but only on the most cutting-edge technology. This has succeeded in denying China access to ASML’s EUV lithography (capable of producing chips 5nm and under, though also used for 7nm chips), for example—but not to all of ASML’s less advanced but still highly capable deep ultraviolet (DUV) lithography, which is used to produce chips between 40nm and 7nm. According to the Select Committee report, China is attempting “to use DUV to produce even smaller nodes, including 5nm,” and while such SME produces advanced chips at reduced efficiency and with reduced yield, China is attempting to make up the difference through sheer volume by actively acquiring as many of ASML’s DUV machines as possible.¹⁰Ibid. Seventy percent of ASML’s DUV sales in 2024 were to China.¹¹Ibid. Nor do American countrywide export controls apply to node-agnostic SME, the equipment that is needed to produce chips of any kind, advanced or otherwise.

The U.S. also imposes entity-specific SME export controls, which can reduce acquisition by entities of concern, but ultimately represents merely a hurdle to overcome, not an outright denial. Chinese companies are adept at “entity obfuscation,” and BIS itself has approved a majority of license applications for exports of SME even to listed entities.¹²Ibid. Nor are U.S. and allied export control regimes well-aligned. U.S. SME producers face greater constraints relative to allied producers, which continue to sell to China in large volumes. 

While working cooperatively with allies to impose controls is ideal, these meaningful gaps in the current SME control regime must be addressed—unilaterally if necessary. Because U.S. technology is so deeply integrated throughout the supply chain, the U.S. can and should use the Foreign-Produced Direct Product Rule (FDPR)—which allows the Department of Commerce to restrict foreign-to-foreign transactions of products that incorporate U.S. technology—to exercise unilateral control over export of components that sit outside of direct American control but utilize American technology. In order to minimize conflict with allies over unilateral action, the U.S. should consider establishing an escalation process, whereby allies can meet key thresholds for imposing their own export controls before FDPR kicks in. In addition, the U.S. should begin by taking a surgical approach to FDPR enforcement, focusing on the least intrusive means of accomplishing the goal of limiting SME exports to China. (For example, restricting the sale by Japanese firms of a critical component of DUV lithography machines when the final buyer is in the PRC.)¹³These suggestions are broadly consonant with the Select Committee’s recommendation to “dramatically expand country-wide controls for the PRC, with a licensing policy of presumption of denial, to apply to any SME and related components and consumables, that can be used in an advanced or foundational fab, utilizing FDPR if necessary.” Finally, the controls should come with positive incentives for allies. What incentives are most suitable will depend on the interests of the nation in question, but increased cooperation on critical mineral supply chain resilience would be a good place to start. China has repeatedly used its chokehold on the global critical mineral supply chain as a tool of geopolitical power—including as a means of signaling its own aspiration to control the global semiconductor supply chain and AI stack.¹⁴Keith Bradsher and Meaghan Tobin, “China’s Rare Earth Exports Face New Scrutiny Amid Global Supply Concerns,” The New York Times, October 9, 2025. The U.S. and its allies will require mutual cooperation on this front.¹⁵Zolan Kanno-Youngs, “To Counter China’s Hold on Rare Minerals, Trump Turns to Australia,” The New York Times, October 20, 2025.

When it comes to selling chips themselves, the United States has vacillated from prohibition to permissiveness. In response to restrictions imposed by the Biden administration, Nvidia designed its H20 chip specifically for sale to China. In April 2025, the Trump administration announced it would ban the H20’s sale to China. A few months later, the administration reversed course, announcing it would allow H20 sales.¹⁶NVIDIA Newsroom, “NVIDIA CEO Jensen Huang Promotes AI in Washington, DC and China,” NVIDIA Blog, July 14, 2025; Demetri Sevastopulo and Michael Acton, “Nvidia and AMD to Pay 15% of China Chip Sale Revenues to US Government,” Financial Times, August 11, 2025. As of October 2025, the administration had not yet articulated its preferred long-term approach to the strategic question of advanced AI chip exports to China. 

Despite its military and economic ambitions, China meaningfully trails the United States in AI capability, due to its limited capabilities in the design and manufacture of advanced AI chips.¹⁷By “advanced AI chips,” we mean high-performance semiconductors—such as graphics processing units (GPUs), application-specific integrated circuits (ASICs), or field-programmable gate arrays (FPGAs)—engineered to accelerate neural network training and inference through massive parallel computation and high-bandwidth memory. Producing leading-edge chips requires both sophisticated designs and the ability to produce chips at the 3nm node, both of which China currently lacks. On design, China lags the U.S. by several years, in large part because there is no reason to design chips they cannot produce. On production, however, they are far behind. Despite limitations, current export controls on SME have kept China from producing leading-edge chips, while limits on imported HBM (which is packaged in the final chip) have constrained overall capacity. 

According to SemiAnalysis, China’s production capability is severely limited: about 500,000 Ascend units (mostly 910Bs) in 2024 and about 800,000 (about 650,000 of which were the more advanced 910Cs) in 2025. Citing limitations on HBM, SemiAnalysis predicts production of only 250,000–300,000 Ascend 910Cs in 2026.¹⁸Dylan Patel et al., “Huawei Ascend Production Ramp: Die Banks, TSMC Continued Production, HBM is The Bottleneck,” SemiAnalysis, September 8, 2025. Chinese production is severely hampered by HBM availability and low manufacturing yields (the percentage of functional chips produced from a wafer). Reliable estimates are hard to come by, but industry experts suggest a range of 30–50% yields, compared to more than 90% yields for U.S.-allied manufacturers. This is compared to U.S. production of 3.67 million Nvidia B300-equivalent chips in 2025. (Each B300 is roughly five-times more powerful than a 910C chip, so this is the equivalent of about 18 million 910C-equivalent chips.)¹⁹Institute for Progress, forthcoming. Even as China reportedly works to grow its domestic semiconductor capacity,²⁰Zijing Wu, “China Seeks to Triple Output of AI Chips in Race with the US,” Financial Times, August 27, 2025. it remains hamstrung by controls on semiconductor manufacturing equipment, which cap Chinese firms’ ability to compete at the leading edge.

This hardware deficit is serious—for now. Of 22 notable Chinese AI models, only two were allegedly trained on Chinese chips.²¹Erich Grunewald and Tim Fist, “Countering AI Chip Smuggling Has Become a National Security Priority: An Updated Playbook for Preventing AI Chip Smuggling to the PRC,” Center for a New American Security, June 11 2025; Veronika Blablová and Robi Rahman, “Why China Isn’t About to Leap Ahead of the West on Compute,” Epoch AI: Gradient Updates, July 26, 2025. DeepSeek’s market share collapsed after it confronted a shortage of Nvidia H20 chips, despite its cost advantages.²²Wei Zhou, AJ Kourabi, and Dylan Patel, “DeepSeek Debrief: >128 Days Later,” SemiAnalysis, July 3, 2025; Qianer Liu and Juro Osawa, “DeepSeek’s Progress Stalled by U.S. Export Controls,” The Information, June 26, 2025. Chinese clusters need 300,000 Huawei Ascend 910C chips to match 100,000 Nvidia B200s.²³Lennart Heim, “China’s Models Close the Gap,” ChinaTalk, April 29, 2025. Huawei CEO Ren Zhengfei’s admission to People’s Daily is telling: “The United States has exaggerated Huawei’s achievements. Huawei is not that great. We have to work hard to reach their evaluation. … Our single chip is still behind the U.S. by a generation.”²⁴Brenda Goh, “Huawei Chips Are One Generation Behind U.S. Peers but Firm Finding Workarounds, CEO Says,” Reuters, June 10, 2025.

The path to realizing Xi Jinping’s dream of a fully modernized PLA with world-beating “intelligent warfighting” capability, supported by and protective of a globally dominant economy that leads on AI, depends on access to chips to train and run leading-edge models. Reports that the PLA has submitted multiple requests for Nvidia H20 chips in the past year,²⁵Sobolik, Michael (@michaelsobolik). “The People’s Liberation Army put in multiple requests for H20 chips over the past year. You can call it ‘virtue signaling’ or ‘faux hawkery’ all you want, but the PLA is serious about its chip procurement.” X, October 23, 2025. the $16 billion in orders for H20 chips that Chinese companies placed in Q1 2025,²⁶Tao Burga, Arushi Gupta, and Tim Fist, “The H20 Problem: Inference, Supercomputers, and US Export Control Gaps,” Institute for Progress, April 15, 2025. and the 700,000 orders for H20s after the administration reversed its export ban²⁷Qianer Liu, “China Demands Companies to Halt Nvidia Chip Orders Over Security Concerns,” The Information, August 12, 2025. should be understood in this context.

The National Security Threat

In the Chinese military’s view, mitigating the U.S. advantage in AI and allowing rapid AI development in China would enable it to achieve parity with American military capabilities—or leapfrog them.²⁸Sam Bresnick, China’s Military AI Roadblocks: PRC Perspectives on Technological Challenges to Intelligentized Warfare, Center for Security and Emerging Technologies, June 2024. The Chinese Communist Party (CCP) frames its AI ambitions in these terms, invoking the desire not only for economic strength but military might. AI dominance is integral to the CCP’s military strategy.

President Xi Jinping has expressed his desire for the People’s Liberation Army (PLA) to “basically complete” its modernization by 2035.²⁹U.S. Department of Defense, Military and Security Developments Involving the People’s Republic of China 2024, Annual Report to Congress, December 18, 2024. The PLA conceptualizes this modernization as a process of technological development: from “mechanization” (deploying modern weapons systems and other military platforms) to “informatization” (linking those modern systems to GPS and other networks) and finally to “intelligentization” (the integration of AI and other leading tech for purposes of “intelligent warfare”).³⁰In Chinese: mechanization 机械化, informatization 信息化, and intelligentization (智能化). See Bresnick, China’s Military AI Roadblocks; Jacob Stokes, Military Artificial Intelligence, the People’s Liberation Army, and U.S.-China Strategic Competition, testimony before the U.S.-China Economic and Security Review Commission, February 1, 2024; Kevin Pollpeter and Amanda Kerrigan, The PLA and Intelligent Warfare: A Preliminary Analysis, Center for Naval Analyses, October 2021. As far back as 2020, the PLA understood itself to have “basically achieved mechanization and made significant progress in informationization.”³¹“Regular Press Conference of the Ministry of National Defense on November 26,” Ministry of National Defense of the People’s Republic of China, November 29, 2020. Intelligentization is the domain in which the PLA believes it has a “rare strategic opportunity…to overtake others.”³²Sam Bresnick, China’s Military AI Roadblocks: PRC Perspectives on Technological Challenges to Intelligentized Warfare, Center for Security and Emerging Technologies, June 2024; see also “Lieutenant General Liu Guozhi: The development of military intelligence is a strategic opportunity for our army to overtake others,” [刘国治中将:军事智能化发展是我军弯道超车的战略机遇], CCTV News, October 22, 2017. The PLA is aggressively seeking to demonstrate significant progress on this front by 2027—the PLA’s centenary and the next CCP-set milestone on the path to 2035 and 2049.³³U.S. Department of Defense, Military and Security Developments Involving the People’s Republic of China 2024.

One of the PLA’s “core operational concepts” is that the future of military conflict will require dominance in “Multi-Domain Precision Warfare.”³⁴U.S. Department of Defense, Military and Security Developments Involving the People’s Republic of China 2022, Annual Report to Congress, November 29, 2022. According to Jacob Stokes of the Center for a New American Security, the concept:

… posits that the very networking that gives the U.S. military its power creates interdependencies between its forces, which are also vulnerabilities that can be exploited. Thus, rather than needing to destroy U.S. enemy forces directly—ship-to-ship or tank-to-tank—China can attack the weak points linking U.S. systems and domains together and thereby neutralize or overwhelm U.S. advantages. Those weak points can include internet, satellite, or electromagnetic communications links as well as logistical supply systems. AI is a critical part of this strategy because, in the dynamic environment of an actual conflict, identifying and targeting U.S. vulnerabilities will require sensing, relaying, and processing vast amounts of information at a speed only computers can match.³⁵Stokes, Military Artificial Intelligence.

“PLA strategists,” according to the U.S. Department of Defense, “view warfare as a science,” and believe that achieving fully “intelligentized” warfighting capability requires AI leadership across a range of applications:

PLA strategists have stated new technologies will increase the speed and tempo of future warfare, and that operationalization of AI will be necessary to improve the speed and quality of information processing by reducing battlefield uncertainty and providing decision making advantage over potential adversaries. The PLA is researching various applications for AI including support for missile guidance, target detection and identification, and autonomous systems. The PLA is exploring next-generation operational concepts for intelligentized warfare, such as attrition warfare by intelligent swarms, cross-domain mobile warfare, AI based space confrontation, and cognitive control operations. The PLA also considers unmanned systems to be critical intelligentized technologies, and is pursuing greater autonomy for unmanned aerial, surface, and underwater vehicles to enable manned and unmanned hybrid formations, swarm attacks, optimized logistic support, and disaggregated ISR, among other capabilities.³⁶U.S. Department of Defense, Military and Security Developments Involving the People’s Republic of China 2022.

According to the U.S. Intelligence Community’s 2025 Annual Threat Assessment report, “China almost certainly has a multifaceted, national-level strategy designed to displace the United States as the world’s most influential AI power by 2030.”³⁷Office of the Director of National Intelligence, 2025 Annual Threat Assessment of the U.S. Intelligence Community, March 18, 2025. This includes the CCP’s Military-Civil Fusion strategies, through which “the CCP is acquiring the intellectual property, key research, and technological advancements of the world’s citizens, researchers, scholars, and private industry in order to advance military aims,” and which thus render the distinction between military and civilian technological development meaningless.³⁸U.S. Department of State, What Is MCF One-Pager, May 2020; U.S. Department of State, U.S. Technology in the Military-Civil Fusion Strategy Revision. This strategy is personally supervised by Xi and results in “Joint research institutions, academia, and private firms… all being exploited to build the PLA’s future military systems—often without their knowledge or consent.”³⁹U.S. Department of State, What Is MCF One-Pager.

Senior U.S. State Department officials confirm PLA integration of commercial AI systems, with DeepSeek models directly aiding military applications.⁴⁰Michael Martina and Stephen Nellis, “DeepSeek Aids China’s Military and Evaded Export Controls, U.S. Official Says,” Reuters, June 23, 2025. The PLA also reportedly submitted multiple requests for Nvidia H20 chips in 2025, underscoring the military applications of this technology.⁴¹Michael Sobolik (@michaelsobolik), “The People’s Liberation Army put in multiple requests for H20 chips over the past year. You can call it “virtue signaling” or “faux hawkery” all you want, but it’s reality,” X, August 16, 2025.

Advanced AI capacity will be critical to the future of warfare and military dominance. The CCP understands itself to be in an aggressive arms race with the United States for that capacity. The United States currently has a decisive advantage in that race. The CCP’s military ambitions require closing that gap; its quickest path to that goal is American willingness to freely sell it the chips and chipmaking equipment to do it.

The Economic Urgency

AI is poised to become the next general-purpose technology, with applications ranging from biomedical research to education to fashion design to drone warfare. The United States currently dominates much of the AI “stack,” from leading-edge semiconductor design to world-leading labs that train and deploy AI models. If the U.S. can maintain this lead, the benefits will compound over time, creating lasting advantages. If it chooses to carelessly sell its edge away to a near-peer adversary, the loss will be generational.

As AI diffuses throughout the economy, its benefits will directly scale with compute. With ten times China’s compute capacity, America can deploy proportionally more AI capacity across every economic sector, creating compounding advantages in drug discovery, logistics, research, manufacturing, and innovation itself.⁴²Heim, “China’s Models Close the Gap.” The U.S. currently hosts roughly 75% of global compute, while China lags behind at 14%.⁴³Konstantin F. Pilz et al., “AI Supercomputers: Performance Share by Country,” Epoch AI, June 5, 2025.

For its part, China is desperate for an AI-driven economic boom. In August 2025, the State Council of China released its official view on the matter.⁴⁴State Council of the People’s Republic of China, “Opinions of the State Council on Deepening the Implementation of the ‘Artificial Intelligence Plus’ Initiative,” August 26, 2025. Its Opinions on Deeply Implementing the “Artificial Intelligence Plus” Action calls for embedding AI in every facet of economic life, from industry and consumer applications to science and technology, with the goal of creating an AI-powered “intelligent economy”—and ultimately an “intelligent society”—in the next few years.⁴⁵Peng Zhang, “China Releases ‘AI Plus’ Policy: A Brief Analysis,” Geopolitechs, August 26, 2025. The State Council’s stated goal is to

[P]romote the broad and deep integration of artificial intelligence with all sectors and domains of the economy and society, reshape paradigms of human production and daily life, foster a revolutionary leap in productivity and profound changes in production relations, and accelerate the formation of a new form of intelligent economy and intelligent society…Guided by Xi Jinping Thought on Socialism with Chinese Characteristics.⁴⁶State Council, “Opinions on AI+ Initiative.”

When CCP leaders say they are betting China’s future economic dominance on AI, U.S. policymakers should believe them.

With respect to who can purchase those chips that are produced, economic tradeoffs need to be considered. The AI chip industry faces significant bottlenecks and has repeatedly struggled to meet demand. In early 2025, OpenAI CEO Sam Altman reported that its GPT 4.5 model release was delayed due to a lack of compute capacity.⁴⁷Jowi Morales, “OpenAI has run out of GPUs, says Sam Altman — GPT-4.5 rollout delayed due to lack of processing power,” Tom’s Hardware, February 28, 2025. Upon release, Nvidia’s Blackwell chip was reportedly sold out for the following year.⁴⁸Anton Shilov, “Nvidia’s Blackwell GPUs are sold out for the next 12 months — chipmaker to gain market share in 2025,” Tom’s Hardware, October 11, 2024. High-bandwidth memory (HBM) supplied by SK Hynix is similarly constrained, with reports in May 2025 noting that its 2025 capacity was nearly sold out.⁴⁹Joyce Lee and Heekyong Yang, “Nvidia Supplier SK Hynix Says HBM Chips Almost Sold Out for 2025,” Reuters, May 2, 2024. On Nvidia’s latest earnings call in late August 2025, CEO Jensen Huang boasted that “everything’s sold out,”⁵⁰NVIDIA Corporation, “Q2 FY2026 Earnings Call Transcript,” Yahoo Finance, August 27, 2025. while the Wall Street Journal reports that “supply-chain constraints” are a key limiting factor on the company’s growth.⁵¹Asa Fitch, “Nvidia’s ‘Wow’ Factor Is Fading,” Wall Street Journal, September 9, 2025.

American chip designers may wish to expand their market share in China to further globalize their reach, but in some cases their exports do not simply boost China’s fortunes but actively reallocate capacity away from the United States in the process.

Most analysts estimate the economic value of that computing power in the trillions of dollars,⁵²Jennifer Sor, “Morgan Stanley Sees the AI Productivity Boom Adding $16 Trillion to the Stock Market’s Value,” Business Insider, August 18, 2025. dwarfing any incremental revenue that chipmakers could secure by filling Chinese demand.⁵³Fazian Farooque, “Nvidia’s Big Week: Earnings Loom as China Deal Clears Path,” Yahoo Finance, August 25, 2025. In celebrating the mutual gains from trade, economists assume that the parties to the exchange face the full costs and capture the full benefits of the deals they make. And indeed, selling chips to China may be an entirely rational choice for the profit-maximizing chipmaker. But where the security risk and economic opportunity for the nation dwarf the narrow pecuniary concerns of the deal, free trade can grievously harm the national interest. It is for policymakers, not CEOs, to decide whether the United States benefits more from dominating AI compute or cashing a few extra checks.

Learning from Past Mistakes

On March 1, 2000, the Clinton administration issued a press release celebrating the U.S.-China WTO Accession deal. Calling the deal “A Clear Win for U.S. High Technology,” the White House gloated that China would “open its internet and telecom markets to investment and services, and provide stronger protection of intellectual property. This will allow the United States to participate in building China’s information infrastructure.”⁵⁴The White House, “President Clinton Announces New Initiatives to Improve Education,” March 1, 2000. A group of 200 high-tech CEOs agreed, declaring that not only would permanent normal trade relations (PNTR) with China expose “the Chinese to our American business practices, values and perspectives,” but that it would “also facilitate the development and spread of information technologies, including the Internet, ensuring that American firms will be in a position to help shape the continued evolution of China’s economic and social environment.”⁵⁵The White House, “Fact Sheet on Granting China Permanent Normal Trade Relations,” April 3, 2000.

Of course, the United States did indeed participate in building China’s information infrastructure—by handing over American technology, which China studied, replaced domestically, and then sought to displace globally. The CCP playbook is consistent:

Acquire → Indigenize → Dominate → Weaponize

China obtains foreign technology through joint ventures or forced transfers, develops domestic alternatives, uses predatory subsidies to destroy global competition, then leverages market dominance for geopolitical ends. The “addiction thesis”—the belief that China would remain dependent on U.S. technology, benefiting both nations and creating American leverage—proved decisively and disastrously wrong. In industry after industry, the U.S. traded away technological dominance for short-term financial gain and is now paying the price.

This pattern is demonstrable in many sectors. Huawei is now the leading provider of 5G infrastructure in the world.⁵⁶Ketan Gandhi, “Top 5G Infrastructure Companies in the World | 2025,” Expert Market Research. China’s global market share of solar panels grew to over 80% by 2021, despite the United States pioneering much of the underlying technology.⁵⁷International Energy Agency (IEA), Solar PV Global Supply Chains, IEA, Paris, August 2022. China controls 90% of global rare earth processing⁵⁸Keith Bradsher, “China’s Rare Earth Export Restrictions Could Disrupt Global Supply Chains,” The New York Times, June 3, 2025. and 75% of global lithium-ion battery production, both of which are other sectors once dominated by the United States before offshoring to China.⁵⁹Teo Lombardo, Leonardo Paoli, Araceli Fernandez Pales, and Timur Gül, “The Battery Industry Has Entered a New Phase,” International Energy Agency, March 5, 2025.

Advanced AI chip design and manufacture is one of the few critical sectors in which the U.S. and its allies retain the kind of dominance the U.S. once enjoyed in many others. The Trump administration understands the danger of unfettered trade with China better than any administration in living memory. It is critical that it not repeat America’s failed experiments with attempting to control the CCP through the free trade of American technology into China.

The answer is not to send China the technology the CCP wishes to dominate. It is a policy of full denial of access to the technology that can most effectively accelerate China’s AI progress. Without that access, the CCP will continue to struggle. Even exporting non-leading-edge (by American standards) chips to China, such as the Nvidia H20, extends the country’s AI capabilities by months, if not years.⁶⁰Burga, Gupta, and Fist, “The H20 Problem.” Conversely, constraining China’s compute access has compounding benefits over time: as each generation of AI model demands more compute, it will fall further and further behind.

Some argue that gaining a foothold in China is critical to maintaining the U.S. lead because it will disincentivize Chinese industry from building its own indigenous capacity. Yet this is precisely the playbook China has utilized repeatedly to great success. Selling SME to China will only facilitate development of its own leading-edge chips and semiconductor ecosystem. Access to this technology, and the expertise required to deploy it, will only accelerate development of indigenous alternatives that can and will be substituted when they are ready. The CCP will not relent in its push for indigenization, and the American technology will simply function as a bridge until Chinese firms can catch up. 

Direct chip sales are no safer. Access to more compute now will not slow China’s push for indigenous alternatives, which is driven not by firm-level profit motives but by the CCP’s commitment to its geopolitical and economic goals. While it is true that advanced AI chips require proprietary software ecosystems that benefit from scale, making models built for those ecosystems difficult to port elsewhere, there is no reason to believe that China’s short-term operation within those ecosystems would slow the development of Chinese alternatives. Faith in the power of an American ecosystem fails twice over:

1. It assumes that expanding China’s AI compute capacity in the short-term in return for a slightly slower path to indigenous alternatives will leave China in a weaker position. It will not. The capacity provided by American chips today will simply serve as a bridge until China is able to fully indigenize.
2. It is already evident in the U.S. market that firms can and will resist lock-in to one platform. Hyperscalers are developing their own chips; leading AI labs are shifting to be chip-agnostic. If American companies can see through the problem of dependence on a single chip provider and invest to avoid it, so can the Chinese Communist Party.

From a corporate perspective, the massive Chinese market is tempting. But short-term corporate profits for a few American companies and minimal revenue to the U.S. Treasury are not worth trading away the most valuable economic advantage in modern history or helping America’s main geopolitical adversary achieve military parity.

Policy Framework

The goal of an America First chip export policy should be to assert American and allied control throughout the entire AI supply chain and the full advanced AI stack, from chipmaking tools to model training, using that control to maximize the U.S. advantage over China in AI compute and capability. 

As the President’s AI Action Plan states, “Advanced AI compute is essential to the AI era, enabling both economic dynamism and novel military capabilities. Denying our foreign adversaries access to this resource, then, is a matter of both geostrategic competition and national security.”⁶¹The White House, America’s AI Action Plan, July 2025.

1. Restrict Chip Sales to China
   • Ban physical and remote access to controlled chips for all PRC-based entities, including shell companies whose beneficial owner is based in China. Controlled chips should include:
     • All data-center chips; andAll chips meeting the following criteria:
       • Total processing performance (TPP) > 4800; or
       • Total processing performance (TPP) > 2400 and performance density > 1.6; or
       • Memory bandwidth > 1.4 TB/s; or Interconnect bandwidth > 1.1 TB/s; or
       • Combined memory + interconnect bandwidth > 1.7 TB/s.
     • These criteria align closely with current U.S. export controls to restrict advanced AI chips capable of training cutting-edge AI systems, capturing chips like Nvidia’s H100/H200. The thresholds target chips with the processing power, performance density, and memory bandwidth necessary for frontier AI development, consistent with existing Department of Commerce Bureau of Industry and Security (BIS) regulations.
2. Restrict SME Sales to China
   • Prevent chip fabrication and high-bandwidth memory companies from providing advanced chip production to Chinese entities and crack down on lax due diligence efforts that have allowed shell companies to place orders for restricted chips and memory for Chinese buyers.
   • Ban the export of all U.S.-made advanced semiconductor manufacturing equipment and subsystems to China countrywide, including all Deep Ultraviolet (DUV) immersion systems and components.⁶²The Trump administration’s AI Action plan calls for the Department of Commerce to develop new export controls on semiconductor manufacturing subsystems, many of which currently lack controls. This should be a top priority of the administration. The White House, America’s AI Action Plan, p. 21.
   • While pursuing agreements with allies to further restrict export and servicing of advanced semiconductor manufacturing equipment and subsystems to China would be ideal, the stakes are too high to wait. The administration can unilaterally ban these exports by using the Foreign Direct Product Rule, which was used to strong effect by the first Trump administration. The administration should use this authority and follow through with meaningful fines and litigation against foreign companies that violate it, and with meaningful trade consequences for countries that permit the violation.
3. Increase Funding for Enforcement
   • Meaningfully increase BIS resources to crack down on smuggling, in collaboration with the U.S. intelligence community. The profits from just three reported smuggling cases are more than double the BIS’s annual budget for export control enforcement.⁶³Grunewald and Fist, “Countering AI Chip Smuggling.” As the President’s AI Action Plan notes, greater capacity for “complete coverage” and “enhanced monitoring” is essential.⁶⁴The White House, America’s AI Action Plan.
4. Mandate Location Verification
   • Implement a location verification requirement on all controlled chips, in line with the President’s AI Action Plan recommendation to leverage “new and existing location verification features on advanced AI compute to ensure that the chips are not in countries of concern.”⁶⁵The White House, America’s AI Action Plan. While this would not itself solve the problem of chip smuggling, it would help provide BIS with information needed to target its enforcement efforts.⁶⁶Chip Security Act, H.R. 3447, 119th Cong., 1st sess., 2025.
   • Implement a location verification requirement on all controlled SME, in line with the Select Committee’s recommendation that BIS “mandate the use of location-tracking technology in appropriate SME so that SME companies can continually verify locations of SME and report those locations to BIS in instances where suspected violations have occurred.”⁶⁷Select Committee on the Chinese Communist Party, Selling the Forges of the Future.
5. Establish Whistleblower Platform
   • Establish a public whistleblower platform that can securely and confidentially receive reports on illegal smuggling of advanced AI chips. Chip smuggling to China is pervasive and lucrative. Fighting it will require incentivizing those with direct knowledge to speak up and protecting them when they do.⁶⁸See, for example, The Stop Stealing our Chips Act of 2025, U.S. Senator Mike Rounds, “Rounds Introduces Legislation to Prevent Smuggling of American AI Chips into China,” April 10, 2025. See also the recommendation in Select Committee on the Chinese Communist Party, Selling the Forges of the Future.