The Stakes of the AI Race & Export Controls
As AI becomes increasingly central to 21st-century power, it is important for the United States to sustain its leadership in frontier AI development in order to support national interests. This requires maintaining a strong position in the capabilities and infrastructure needed to develop advanced AI systems, collectively referred to as the “AI tech stack.” These elements include technical talent, training data, algorithmic innovation, and key hardware components such as energy infrastructure and rare earth materials. Arguably the most critical component in the AI tech stack, however, is the advanced semiconductor chips and specialized tools required to design and manufacture them, which form the computing foundation for frontier AI development and deployment.
Maintaining a US lead in AI relative to competitors like China, therefore, depends on controlling access to these chips and the associated supply chain components. Both the Biden and Trump administrations have recognized this and sought to leverage this comparative advantage through increasingly sophisticated export control regimes.
Maintaining a Comparative Advantage
Export controls serve as the primary mechanism for managing semiconductor access. Unlike Cold War-era controls that focused on denying adversaries specific weapons systems or technologies, modern semiconductor controls aim to slow an entire technological ecosystem—restricting not just AI chips themselves, but also high-bandwidth memory (HBM), semiconductor manufacturing equipment (SME), and the technical know-how required to produce them. This approach principally aims to maintain current US advantages in frontier AI and make it prohibitively expensive for China to replicate US capabilities.
Fortunately, the United States has the upper hand in many dimensions of the AI dominance race. To start, the United States maintains technological leadership in AI in large part because China cannot yet produce the advanced chips required for frontier AI systems at high quality and scale. Preserving this production gap, then, should be the linchpin of American technology strategy. Most critically, the United States maintains a lead in total compute capacity worldwide, owning roughly ten times more advanced AI chips than China for R&D, training, and sustaining the deployment of AI throughout society, as well as dominating the design of these top-end GPUs. When including allied countries like the Netherlands, Taiwan, Japan, and South Korea, the US-led coalition controls virtually every critical node in the advanced semiconductor supply chain, a structural advantage that China cannot easily replicate despite heavy investment in indigenous capabilities.
In Search of a Strategy
However, the US lead in AI is neither self-sustaining nor guaranteed, particularly given the fragmented nature of the current policy approaches. The Department of Commerce’s Bureau of Industry and Security (BIS) has diligently employed its core export control tools to address the challenge, primarily by adding Chinese end-users to the “Entity List” (which restricts their access to controlled items) and denying licenses for those items accordingly. Yet these controls have largely been implemented in isolation, with unclear objectives and haphazard enforcement that allows significant leakage through direct sales, third-party transfers, and inadequate verification. While BIS has refined these tools—such as by implementing performance-based controls that restrict chips above certain computational thresholds and asserting extraterritorial jurisdiction through the Foreign Direct Product Rule to limit foreign-made products incorporating US technology—these measures remain largely reactive rather than proactive, failing to anticipate circumvention tactics.
The challenge facing policymakers is not simply whether to impose controls, but how to wield them as part of an integrated approach to technology competition. Without coordination across government agencies, allied nations, and industry, controls become a game of regulatory whack-a-mole that China has proven adept at circumventing. Intelligence agencies identify new circumvention networks, BIS adds entities to restriction lists, but new front companies and routing mechanisms emerge faster than regulators can respond. This asymmetry favors the evader over the enforcer.
Don’t Trade Away the Lead
There are multiple ways in which the effectiveness of the US export control regime is being undermined. First, direct leakage of controlled chips and SME is helping China narrow its compute shortfall. Export control failures—from an active smuggling black market for banned GPUs, to TSMC’s basic compliance breach that provided Huawei with millions of advanced chips, to active efforts to divert and circumvent controls by commercial actors—have repeatedly undermined US strategy. Second, indirect diffusion occurs when “acceptable” actors acquire materials that subsequently diffuse to the Chinese military. Third, domestic market sales from companies like Nvidia directly into China enable both pathways and effectively support PRC scaling. Finally, policymakers could decide to trade away the US advantage in AI computing directly by using export controls as simple bargaining chips in search of a broader trade agreement, relegating the AI competition from being a central national security issue to one of solely market competition.
Recent US-China negotiations, in which the Trump Administration agreed to suspend the Entity List affiliates program for one year in exchange for China’s suspension of rare earth controls, demonstrate that the Administration views these controls as tradeable. The asymmetry is revealing: China can quietly build out its rare earth export control mechanisms during the suspension, while the US pause in Entity List affiliations enables front companies to proliferate and chips to leak into restricted end-uses. This precedent risks subordinating long-term technological competition to near-term trade agreements.
Each of these pathways reinforces the others. Direct leakage is facilitated by the legitimacy that domestic market sales provide. Indirect diffusion becomes more difficult to control when US companies actively promote their products in the Chinese market. Meanwhile, the possibility of trading away controls creates uncertainty that undermines the core national security concerns at hand and allies’ incentives to cooperate with the United States, while also encouraging companies to maintain Chinese market relationships in anticipation of future easing. Maintaining the advantages the United States has built over the past five years in frontier AI will depend on whether the Trump administration reinforces these restrictions or compromises them in pursuit of a broader trade deal with Beijing.
What’s needed is a fundamental reframe that treats export controls not as standalone regulatory tools but as one element of a comprehensive technology competition strategy. This strategy must integrate denial measures with positive inducements, multilateral coordination with domestic capacity building, and enforcement mechanisms with industry partnership.
Compete, Counter, Cooperate
Addressing these realities requires a comprehensive strategy built on three primary elements. The first pillar is to compete. The Trump administration’s focus on a “promote” pillar that seeks to sell the US AI tech stack abroad aims to expand an affirmative vision for US AI around the world while limiting China’s influence in this space. Continued investment in domestic industrial capacity is also a critical element of this approach. Complementary domestic policy initiatives, including efforts to establish federal preemption of AI governance and standardization through the Commerce Department’s Center for AI Standards and Innovation, create the regulatory coherence necessary for this investment to translate into strategic advantage.
Crucially, the compete pillar is fully compatible with—and indeed depends on—the strategic use of export controls to limit China’s access to advanced chips, SME, and other hardware it still struggles to produce domestically at sufficient quality and scale. This dual approach can work through a deliberate prioritization strategy. US industry should prioritize sales to domestic, allied, and third-country markets rather than treating the Chinese market as equivalent. Companies could also shift from selling chips outright to China to leasing compute through cloud service providers, which would allow the United States to retain ownership and control while still serving legitimate commercial demand. This model already exists: US cloud providers routinely lease compute to foreign actors as a core business practice. The proposal here is to formalize this existing practice as the primary mechanism for providing compute to China, rather than selling physical chips.
The Remote Access Security Act, which passed the House Foreign Affairs Committee unanimously in April 2025, would authorize BIS to issue licenses and impose penalties on exporters of technology accessed over network connections such as cloud computing services. Chinese entities are already exploiting this very model in reverse—companies banned from purchasing Nvidia chips directly, like iFlytek and parts of Shenzhen University, have reportedly accessed restricted H100-equivalent chips through AWS cloud services via intermediaries. This underscores the viability and the urgency of formalizing US control over this key loophole.
Paired with stronger export control monitoring, this approach would allow US industry to compete in the Chinese market while maintaining the ability to cut off access in contingencies. The strategic calculus is straightforward: weigh the marginal revenue from a one-time chip sale to China against the sustained risk of Chinese data centers using American chips for military or surveillance applications beyond US control.
The second pillar is to more aggressively counter China’s efforts to surpass the United States in AI capabilities through the strategic use of export controls. To be effective, export controls must move beyond reactive policy development and violation response toward a proactive enforcement posture. Effective export control implementation requires four elements: (1) maintaining current controls on critical chokepoints, (2) expanding selectively to cover loopholes in chip production capacity, (3) deepening allied coordination, and (4) enabling continuous monitoring of diversions and capabilities.
Maintaining current controls on critical chokepoints is the foundation. These chokepoints represent the narrow set of capabilities where China remains structurally dependent on foreign suppliers, such as producing quality AI chips and the associated software, making them the critical terrain for maintaining strategic advantage. Some argue that selling chips to China creates beneficial dependencies and generates revenue for US firms. However, this misunderstands the nature of the competition. One-time chip sales transfer permanent capabilities to Chinese data centers beyond US control, while simultaneously funding China’s efforts to eliminate the dependency altogether by accelerating indigenous production programs that Beijing will almost certainly pursue in any event. The goal is not to simply create manageable dependencies, but to preserve the structural chip production gap itself.
Existing restrictions, such as controls on extreme ultraviolet (EUV) lithography equipment and key subcomponents like HBM, should be enhanced with greater ex ante flexibility for BIS to act on intelligence proactively rather than waiting for violations to surface. Building on this foundation to close exploitable gaps is urgent, particularly the deep ultraviolet immersion (DUVi) loophole that China is exploiting at scale. Current controls restrict the most advanced DUVi systems but allow older systems into China, provided they are not used for “advanced production” at 14nm and below. Some industry experts claim these older models can be used to produce Huawei’s Ascend AI chips at 7nm, rendering the current restriction ineffective. Therefore, the United States must instead impose country-wide restrictions on all DUVi systems capable of sub-20nm production, as well as precursors that enable HBM manufacturing, like the materials and chemicals used to stack memory layers.
Strengthening BIS’s integration with the US intelligence community is essential for creating a truly proactive enforcement posture. This requires embedding intelligence collectors and analysts at BIS to run targeting cells on military-civil fusion end-users, front companies, and logistics hubs. This transforms end-user validation from a periodic regulatory exercise into an ongoing intelligence-targeting mission that continuously updates enforcement priorities. For example, the US intelligence community could embed targeting officers and analysts at BIS to generate ex ante insight into military-civil fusion end-users, front companies, and hubs for circumvention of export controls, all of which is essential for countering China’s systematic use of civilian companies for military purposes.
The third pillar is to cooperate with allies, third countries, and industry to build a resilient US-led technology order that creates a coalition of advantage. Export controls cannot succeed as a unilateral tool—they require multilateral coordination, allied trust, and industry partnership to prevent China from simply routing around restrictions through alternative suppliers or uncoordinated jurisdictions.
Industry partnership is essential because companies possess the technical knowledge and market intelligence that governments lack. At the same time, lightweight technical standards, like chip location verification technology, can reduce compliance costs while improving assurance, increasing confidence that chips remain where licensed without creating excessive compliance burdens. The key is treating industry as a partner in enforcement rather than simply a regulated entity.
History also offers a valuable template: the Coordinating Committee for Multilateral Export Controls (CoCOM) during the Cold War successfully slowed Soviet technological access through a quiet, intelligence-backed, multilateral process that harmonized denial lists and coordinated enforcement across jurisdictions. While China presents fundamentally different challenges from the Soviet bloc, CoCOM still presents an instructive model. The coalition’s approach follows what Rush Doshi and Kurt Campbell term “allied scaling“: building a robust coalition that pools comparative advantages in the full AI tech stack to out-produce China.
Building a “CoCOM 2.0” for AI and semiconductors requires a true strategic alignment through harmonized export control packages and enforcement that eliminate the arbitrage opportunities China has exploited. The United States should combine the existing “trilateral” (US-Netherlands-Japan) with the “Fab 4” semiconductor alliance (US-Japan-South Korea-Taiwan) and expand to include the UK, the EU, Canada, Australia, Singapore, and Malaysia into a quiet, intelligence-fed coalition. More critically, this requires building the intelligence-sharing architecture that enables coordinated enforcement. To build this coalition, the Trump administration must stay focused on the shared threat: preventing China from acquiring the resources needed to build AI capabilities that pose common national security risks. This would allay concerns among allies about sharing sensitive information that might be leveraged for other political objectives, like a trade deal between only the United States and China.
Conclusion
Export controls must transform from reactive compliance mechanisms into proactive strategic instruments. The Compete-Counter-Cooperate framework demonstrates that controls need not conflict with promoting US AI abroad; rather, it enables this by denying China competitive space while allowing allies to scale, with each pillar reinforcing the others. For example, the GAIN AI Act links domestic capacity building with export control strategy by mandating chipmakers to meet US compute demand first. This approach protects the strategic advantage vis-á-vis China without letting export controls become a reactive system that China can circumvent faster than the United States can respond.
Without “Compete,” the United States cannot maintain advantages worth protecting. Without “Counter,” those advantages diffuse to adversaries. Without “Cooperate,” the United States faces a larger competitor alone. Coordinated implementation of these three pillars creates the foundation for sustained US technological leadership. The question is not whether to use export controls, but whether to wield them strategically.
…
Christian Chung is a former US intelligence officer with over a decade of experience leading analytic and operational teams on Middle East and emerging technology issues. He is currently a Ph.D. candidate at Princeton University focused on technology competition and military innovation, a US Navy Reserve Intelligence Officer, and a Senior Advisor at the Horizon Institute for Public Service. Christian is a Council on Foreign Relations Term Member and a 2025 NextGen National Security Leaders Fellow at the Center for a New American Security.
Image Credit: Markus Spiske, CC0 1.0, via European Alternatives
