US semiconductor export controls targeting advanced computing chips and chipmaking equipment have undergone successive tightening rounds since October 2022, with each revision closing loopholes that Chinese buyers and their suppliers had exploited in prior iterations of the control regime. The October 2023 and subsequent 2024 rule updates have expanded the technical parameter coverage of controlled chips, extended export licence requirements to additional country destinations beyond China, and subjected semiconductor manufacturing equipment to more comprehensive licence requirements that are constraining ASML's ability to serve existing customers and win new tool orders in China. The cumulative effect of these controls is market fragmentation: a compliant global ecosystem of AI chip supply and deployment centred on US-allied geographies and an increasingly separate Chinese ecosystem built on domestically produced alternatives that are technically inferior but commercially available without licence constraint.
The Control Architecture and Its Commercial Impact
The technical parameter thresholds that define US export control applicability to advanced chips have been designed to target AI training hardware specifically, with performance density limits on interconnect bandwidth and total compute throughput that capture data centre GPU products while exempting consumer-grade chips. NVIDIA's A100 and H100 GPU products, which dominate AI training infrastructure globally, were the primary targets of the October 2022 controls, prompting NVIDIA to develop H800 and A800 variants with reduced interconnect specifications designed to fall below the control thresholds. The October 2023 rule tightening closed this workaround by lowering the threshold parameters, capturing the modified variants as well as the original controlled products. NVIDIA's subsequent development of an H20 variant designed to comply with revised controls has been the company's commercial response to maintaining access to Chinese data centre customers while operating within the control regime's revised parameters.
AMD has faced parallel constraints on its Instinct MI300 series AI accelerator products, which achieve performance levels that fall within the control parameter thresholds when configured for high-bandwidth multi-chip deployments. Intel's Gaudi AI accelerator products have navigated the control regime with variants designed to remain below threshold performance specifications, though the competitive positioning of Gaudi in the AI training market against NVIDIA and AMD is complicated by both technical performance gaps and the market disruption of sequential control regime changes. The commercial cost to US chip companies from Chinese market revenue forgone is substantial: NVIDIA alone identified billions in quarterly revenue from China data centre sales that the controls impacted, representing a meaningful portion of its data centre segment revenue at the time of the initial control imposition.
Chinese Domestic Alternatives and the Huawei Ascend Programme
Chinese domestic AI chip development, led by Huawei's HiSilicon subsidiary through the Ascend 910B and Ascend 910C processor products, has achieved performance levels that are commercially viable for AI inference and increasingly competitive for training workloads, albeit at a performance deficit relative to NVIDIA H100 and H800 equivalents in absolute benchmark terms. The domestic Chinese AI chip ecosystem is completing the validation cycle with major Chinese technology companies including Baidu, ByteDance, Alibaba, and Tencent, who are under regulatory and strategic pressure to diversify away from US chip dependency and have operational incentive to qualify domestic alternatives regardless of performance-per-dollar comparisons with controlled foreign alternatives.
The constraint on Huawei Ascend scaling is semiconductor manufacturing capability rather than chip design maturity. The October 2022 controls and their extensions have comprehensively restricted TSMC and other leading-edge contract foundries from manufacturing chips for Huawei above 10nm process node sophistication, forcing Huawei to rely on SMIC, the Chinese domestic foundry, for Ascend processor production. SMIC's demonstrated capability at 7nm process nodes, revealed through the Kirin 9000S chip found in the Mate 60 Pro smartphone in 2023, was a significant revelation, but yields, wafer starts per month, and advanced packaging integration capability at SMIC remain substantially behind TSMC and Samsung at comparable process nodes, constraining Ascend production volumes and cost competitiveness.
ASML and the EUV Equipment Control Dynamic
ASML's Extreme Ultraviolet lithography systems, which are essential for manufacturing chips at 7nm and below, have been subject to export licence denial to China at Dutch government direction since 2019 under US diplomatic pressure. The control has expanded progressively to cover not only EUV systems but also advanced deep ultraviolet immersion lithography systems that are relevant for 14nm and 7nm manufacturing capability, the technology range at which SMIC is operating. ASML's China revenue exposure remains significant despite the EUV restriction because the Chinese semiconductor industry continues to invest heavily in expansion of trailing-edge capacity using older lithography equipment that remains freely exportable. The progressive tightening of DUV export restrictions represents the primary near-term risk to ASML's China revenue that the company has acknowledged in forward guidance.
The geopolitical complexity of ASML's position is unusually acute: the company's product is genuinely singular in its technological capability, the Netherlands government's export licensing decisions are made under pressure from Washington that creates bilateral diplomatic tension, and ASML's long-term revenue planning must account for a China customer base that represents 29 percent of its total 2023 revenue but whose future equipment purchasing capacity is progressively constrained by the control regime. The company's investment in next-generation High NA EUV systems represents a capability that China will be unable to access domestically for a decade or more, which provides some revenue substitution optionality in non-Chinese markets even as China tool sales contract.
Market Bifurcation and Long-Term Ecosystem Implications
The bifurcation of the global AI chip market into compliant and domestic-Chinese ecosystems has implications that extend beyond the immediate commercial revenue impact on NVIDIA, AMD, and ASML. The Chinese AI computing infrastructure that is being built on domestically produced chips and software stacks is developing on frameworks and tools that are increasingly divergent from the CUDA-centric software ecosystem that dominates Western AI development. This software ecosystem divergence may eventually prove more durable than the hardware performance gap, since software libraries, model training frameworks, and developer workflow tools create switching costs and ecosystem lock-in that persist beyond hardware technology catch-up cycles.
The control regime is also creating commercial opportunity for non-US semiconductor companies and software providers who are not subject to the same export control constraints and can serve the Chinese market freely. South Korean memory producers, Taiwanese packaging and test services companies operating China-compliant product lines, and European fabless semiconductor companies whose products do not cross the advanced AI chip threshold parameters are positioned to serve components of the Chinese semiconductor supply chain without the licence requirement friction that burdens US-controlled products.
Top 10 Companies in Semiconductor Export Controls and AI Chip Geopolitics
- NVIDIA -- Dominant AI training chip supplier navigating US export controls through compliance-designed product variants including H20 GPU for China market while maintaining H100 and H200 positioning in Western data centre markets.
- ASML -- Dutch semiconductor lithography equipment monopolist subject to progressive EUV and advanced DUV export control restrictions to China, with High NA EUV next generation system development representing Western exclusive capability.
- Advanced Micro Devices -- AI accelerator competitor to NVIDIA with Instinct MI300 series subject to export controls, developing compliant product variants while expanding AI chip revenue in Western data centre markets.
- Huawei HiSilicon -- Chinese chip design subsidiary developing Ascend 910 series AI processors as the primary domestic alternative to controlled US GPU products for Chinese data centre AI training applications.
- SMIC -- Chinese semiconductor foundry manufacturing Huawei Ascend and other domestic chip designs at the leading edge of its demonstrated process node capability under comprehensive US equipment export restrictions.
- Intel -- US chip company with Gaudi AI accelerator product line and foundry services business navigating export control compliance while competing in AI chip market against dominant NVIDIA positioning.
- Applied Materials -- Leading semiconductor process equipment company with significant China revenue exposure subject to export licence requirements for advanced deposition, etch, and inspection equipment sales.
- Lam Research -- Semiconductor etch and deposition equipment major with China customer base subject to US export control licensing, with China representing a historically significant revenue proportion.
- KLA Corporation -- Semiconductor process control and inspection equipment company with China revenue exposure under export control review, providing yield management tools critical to semiconductor manufacturing quality.
- Cambricon Technologies -- Chinese AI chip startup producing neural processing unit products for edge and cloud inference applications, targeting domestic Chinese AI deployment markets without export control constraints.
Our Take
US semiconductor export controls are achieving their stated near-term objective of constraining Chinese access to the most advanced AI training hardware, but the medium-term strategic analysis is more complicated. The controls are accelerating Chinese domestic semiconductor investment and self-sufficiency programmes at a pace and resource commitment that the absence of controls would not have produced, potentially shortening the technology gap timeline in a way that creates the competitive threat the controls were designed to prevent. For investors, the most important dynamic to track is not the quarterly revenue impact on NVIDIA or the quarterly restriction announcement from BIS, but the speed at which the Chinese domestic AI chip software ecosystem matures to a point where domestic hardware, even at a performance disadvantage, becomes the default platform for Chinese AI development. That ecosystem maturation, more than hardware performance catch-up, determines the terminal market share outcome. -- Elena Vasquez, MarketsNXT