How the US Uses Semiconductor Restrictions to Constrain China’s AI and Military Capabilities
Export controls on advanced chips have become Washington's primary tool for slowing Beijing's technological and military advance, but enforcement depends on Taiwan's manufacturing monopoly and fragile allied coordination.
The United States has turned semiconductor export controls into its most powerful non-kinetic weapon against China, restricting access to advanced chips essential for artificial intelligence development and military modernisation. These controls target specific chip architectures—particularly high-performance GPUs like Nvidia’s H200 and A100 series—that enable the training of large language models and the operation of autonomous weapons systems. The policy rests on a structural advantage: Taiwan Semiconductor Manufacturing Company produces over 90% of the world’s most advanced chips, giving Washington leverage over a supply chain Beijing cannot yet replicate domestically.
Why Chips Matter for Strategic Competition
Advanced Semiconductors sit at the centre of both AI development and modern military capability. Training frontier AI models requires massive parallel computing power delivered by specialised GPUs, while precision-guided munitions, hypersonic missiles, and next-generation radar systems depend on cutting-edge processors. China’s People’s Liberation Army has publicly identified AI as a core component of its military modernisation strategy, seeking to achieve ‘intelligentised warfare’ by 2035. Denying access to the chips that enable this transformation has become Washington’s preferred method of maintaining technological superiority without direct confrontation.
The restrictions specifically target chips with performance above defined thresholds. The October 2022 rules from the Bureau of Industry and Security prohibited exports of chips exceeding certain computational density and interconnect bandwidth levels—metrics designed to capture AI training accelerators while allowing less capable processors for consumer electronics. Nvidia’s H200, with 141 gigabytes of high-bandwidth memory and performance optimised for transformer model training, falls squarely within the prohibited category.
16%
~5%
600+
The Mechanisms of Control
The US enforces chip restrictions through three overlapping mechanisms. The Entity List maintained by the Commerce Department designates over 600 Chinese companies—including leading AI firms like SenseTime and Megvii, plus the entire Chinese military-industrial complex—as prohibited recipients. Any US company selling controlled technology to these entities faces criminal liability. The Foreign Direct Product Rule extends this control to foreign-made products that incorporate US technology, equipment, or software. Since nearly all advanced chipmaking relies on American lithography software, inspection tools, or design platforms, this provision gives Washington extraterritorial reach over the global semiconductor supply chain.
The third mechanism targets manufacturing equipment itself. ASML, the Dutch company that makes extreme ultraviolet lithography machines essential for producing chips below 7 nanometres, has been barred from selling to China since 2019 under coordinated US-Netherlands Export Controls. Without access to these machines, Chinese foundries cannot produce the transistor densities required for competitive AI accelerators. China’s leading chipmaker SMIC has achieved 7nm production using older deep ultraviolet tools, but the process is inefficient and cannot match TSMC’s 3nm technology used in the latest Nvidia chips.
Enforcement relies on end-use verification and supply chain monitoring. US officials conduct site visits at foreign semiconductor facilities, review sales records, and track chip shipments through customs data. The Bureau of Industry and Security has increased staffing for these audits by 40% since 2022, according to agency budget documents. Violations carry severe penalties: in 2023, the Commerce Department fined a Singaporean distributor $300,000 for diverting restricted chips to China through Hong Kong shell companies.
Allied Coordination and Its Limits
The effectiveness of US restrictions depends entirely on allied cooperation, particularly from the Netherlands, Japan, and South Korea—the only countries with the equipment and expertise to manufacture advanced chips. The 2023 Hiroshima G7 statement committed members to ‘de-risking’ technology supply chains, but national economic interests create friction. South Korean chipmakers Samsung and SK Hynix derive roughly 40% of revenue from Chinese customers; strict enforcement would devastate their business models.
Japan and the Netherlands have implemented parallel export controls on lithography and deposition equipment, but with narrower scope than US rules. Tokyo’s restrictions exempt equipment for chips above 14nm—a threshold that allows continued sales to Chinese factories producing older-generation semiconductors. This creates a policy gap: China can still access tools for 28nm production, sufficient for many military applications including missile guidance systems, even as cutting-edge AI chips remain blocked.
The more fundamental challenge is enforcement beyond allied borders. Third countries—particularly in Southeast Asia and the Middle East—have become transhipment hubs for restricted chips. Customs data analysed by Center for Strategic and International Studies researchers shows unexplained spikes in chip imports to Malaysia and Vietnam that correspond to similar increases in re-exports to China. Shell companies purchase chips through legitimate distributors, relabel them, and ship via circuitous routes. The Commerce Department has no jurisdiction to inspect facilities in non-allied countries, relying instead on intelligence gathering and voluntary cooperation that varies widely.
The semiconductor restrictions represent a reversal of decades of US trade policy that prioritised market access and integration. From the 1990s through the mid-2010s, Washington encouraged technology transfer to China as a path toward economic liberalisation and eventual political reform. The shift to explicit technological containment began during the Trump administration’s first term and accelerated under Biden, reflecting bipartisan consensus that China’s authoritarian system will not converge with Western norms and that maintaining technological superiority requires active denial of critical capabilities.
Taiwan’s Structural Leverage
The entire US strategy rests on Taiwan’s manufacturing monopoly. TSMC produces 92% of the world’s most advanced logic chips and 100% of those below 5nm, according to TrendForce industry data. No other foundry has mastered the complex multi-patterning techniques required for extreme-scale integration at 3nm geometries. This gives the island extraordinary geopolitical importance: if China seized control of TSMC’s fabs, it would eliminate the primary constraint on its AI and military development while simultaneously crippling US technology companies that depend on those same facilities.
Washington has responded with a two-track approach. Diplomatically, State Department officials have repeatedly stated that any attempt to take Taiwan by force would trigger comprehensive sanctions, though they carefully avoid specifying whether this includes military defence. Industrially, the CHIPS and Science Act allocated $52 billion to rebuild domestic semiconductor manufacturing, with TSMC receiving $6.6 billion in grants to construct two fabs in Arizona capable of producing 4nm chips by 2025.
But the domestic buildout faces significant obstacles. TSMC’s Arizona project has encountered delays due to workforce shortages—advanced chipmaking requires specialised technicians that take years to train. The company publicly complained about higher US construction costs and regulatory complexity. Even if the Arizona fabs reach full capacity, they will produce only a fraction of TSMC’s Taiwan output, leaving the US dependent on cross-strait stability for the foreseeable future. This creates a paradox: the more successfully Washington uses chip restrictions to pressure China, the greater Beijing’s incentive to secure the Taiwan fabs that enable those restrictions.
“Whoever controls semiconductor manufacturing controls the future of AI, and therefore military power in the 21st century. Taiwan is not just a political issue—it’s the strategic high ground of the entire technology competition.”
— Former US National Security Advisor, Congressional testimony, 2024
China’s Response and the Limits of Containment
Beijing has pursued three parallel strategies to counter the restrictions. The first is massive state investment in domestic semiconductor capabilities—the National Integrated Circuit Industry Investment Fund has committed over $150 billion since 2014. Progress has been uneven: China now produces competitive chips at 14nm and has achieved limited 7nm production, but remains years behind in sub-5nm processes and entirely dependent on imported lithography equipment.
The second approach targets alternative architectures. Chinese AI labs have optimised algorithms to run on less powerful hardware, achieving competitive performance with older chips through software efficiency. Huawei’s Ascend 910B processor, manufactured domestically at 7nm, reportedly delivers 70% of Nvidia A100 performance at AI inference tasks—sufficient for deployment of trained models, though not for training new frontier systems. Open-source model weights from Western labs, freely downloadable despite export controls, allow Chinese researchers to fine-tune capable AI systems without needing to train from scratch.
The third strategy is aggressive acquisition through grey-market channels. Chinese procurement networks have proven adept at obtaining restricted chips despite controls. A Reuters investigation found restricted Nvidia chips openly advertised on Chinese e-commerce platforms, routed through intermediaries in Singapore and Taiwan. The scale is difficult to quantify, but US officials estimate thousands of high-end GPUs reach China monthly through these channels—enough to support significant AI research, though insufficient for the compute clusters needed to train the largest models.
These countermeasures mean the restrictions slow rather than stop China’s AI development. The policy buys time for the US to maintain a technological lead, but the advantage is narrowing. Chinese labs published over 40% of top-tier AI research papers in 2025, up from 25% in 2020, despite restricted access to cutting-edge hardware. The question is whether the time purchased through export controls is sufficient for the US to establish an insurmountable lead—or whether China’s scale advantages in data, engineering talent, and state resources will eventually overcome the hardware gap.
| Capability | United States | China |
|---|---|---|
| Advanced logic production (sub-7nm) | Via TSMC partnership, Intel 4nm in development | Limited 7nm via SMIC, no sub-5nm |
| Lithography equipment | Full access via ASML partnership | Blocked from EUV, DUV available |
| GPU design expertise | Nvidia, AMD dominance | Huawei Ascend, limited ecosystem |
| AI chip access | Unrestricted domestic use | Restricted imports, grey-market leakage |
| Fab construction timeline | TSMC Arizona: 2025-2028 | SMIC expansion: ongoing, 2-3 generations behind |
The Diplomatic Backlash
The chip restrictions have become a focal point of US-China tensions, with Beijing framing them as economic coercion designed to perpetuate American hegemony. Chinese officials argue the controls violate World Trade Organization principles of non-discrimination and free trade. While the WTO allows export restrictions for national security, China contends the broad scope of US controls—covering commercial AI applications with no direct military use—exceeds that justification. The dispute has not been formally litigated at the WTO, partly because both sides recognise the organisation lacks enforcement mechanisms for such politically charged issues.
Recent diplomatic attempts to use chip access as a bargaining chip have exposed the limits of the strategy. When the Trump administration authorised limited H200 exports to China in May 2026 as part of broader trade negotiations, Beijing responded by blocking the imports domestically—a move that humiliated Washington and demonstrated China’s willingness to sacrifice technological access rather than appear to submit to US pressure. The episode suggested that semiconductor restrictions, while effective at slowing China’s AI development, cannot easily be converted into diplomatic leverage without appearing to reward exactly the behaviour the policy aims to prevent.
Long-Term Sustainability Questions
The semiconductor export control regime faces structural challenges that may limit its effectiveness over time. Technology evolves: today’s cutting-edge 3nm chips will be commoditised within a decade, and China is investing heavily to ensure it controls next-generation technologies like photonic computing and quantum processors that may bypass silicon limitations entirely. The restriction strategy assumes continued US dominance in chip design and allied control of manufacturing—assumptions that become more precarious as China’s domestic capabilities improve and as manufacturing diversifies beyond Taiwan.
The policy also carries economic costs. Nvidia’s data centre revenue from China fell 80% following the October 2022 controls, representing billions in lost sales. Prolonged exclusion from the Chinese market may motivate Beijing to succeed at building indigenous alternatives, permanently reducing demand for US chip technology. Some analysts argue the restrictions accelerate the very decoupling they aim to manage, creating parallel technology ecosystems with reduced interoperability and increased geopolitical instability.
Ultimately, semiconductor restrictions represent a wager that the US can maintain technological superiority through denial rather than competition alone. The strategy’s success depends on variables largely outside Washington’s control: Taiwan’s continued autonomy, allied cohesion despite divergent economic interests, and China’s inability to overcome a multi-decade manufacturing gap through state-directed investment. Whether those conditions hold for another decade remains the central uncertainty in technology-driven great power competition.
Related Coverage
For analysis of the current diplomatic tensions around chip restrictions, see our reporting on the Trump administration’s failed H200 export authorisation and Beijing’s decision to block imports despite US approval. Our coverage of the semiconductor market volatility surrounding recent summits examines how investors price policy uncertainty, while our analysis of TSMC’s Arizona expansion details the domestic manufacturing buildout. For broader geopolitical context, see our pieces on China’s leverage in the current negotiating environment and the competition over AI model architectures.