China Semiconductor Bifurcation

The China semiconductor bifurcation is the most consequential structural change in the global semiconductor supply chain since the foundry model emerged in the 1980s. For forty years, semiconductor supply chains were globally integrated by economic logic: design wherever the talent is, manufacture wherever the equipment and expertise are best (Taiwan, South Korea, Japan), package wherever labor is cheapest, and sell everywhere. Export controls, starting with the 2019-2020 Huawei Entity List actions and accelerating through the October 2022 BIS rules and their subsequent tightening, have broken this logic. The bifurcation is not a policy outcome that may or may not materialize — it has already materialized. Two parallel semiconductor supply chains now exist: one centered on TSMC, ASML, Applied Materials, Lam Research, KLA, SK Hynix, Micron, and Samsung serving the US-allied technology ecosystem; and one being built by SMIC, Hua Hong, CXMT, YMTC, NAURA, AMEC, and Huawei/HiSilicon serving the Chinese domestic technology ecosystem. These two supply chains are not equal in capability — the Western-aligned chain leads by approximately one decade in process technology — but they are increasingly separate, and the degree of technical separation grows with each successive round of export control tightening.

The SX framework for the bifurcation tracks it across six layers of the semiconductor supply chain: foundry (logic chip manufacturing), memory (DRAM and NAND), equipment (lithography, etch, deposition, metrology), materials (chemicals, gases, wafers), EDA/IP (design tools and standard cell libraries), and AI compute (accelerators and the software ecosystems that run on them). Bifurcation is most advanced in AI compute (Huawei Ascend ecosystem is functionally separate from NVIDIA/AMD), partially advanced in foundry and memory, and least advanced in equipment (SMIC and CXMT remain dependent on Western tools purchased before controls tightened). The key analytical question for each layer is not "is China capable of self-sufficiency?" but rather "what is the pace of domestic capability development relative to Western technology advancement, and does the gap narrow or widen over time?"

Related Coverage: Huawei / HiSilicon Spotlight | Spotlights Hub | Process Nodes & Lines | Fab Clusters | AI & ML Sector | HBM Memory | Bottleneck Atlas

Bifurcation Status by Supply Chain Layer

The EUV Export Control — The Mechanism Maintaining the Process Gap

The most important single supply chain control in the US-China semiconductor bifurcation is the EUV lithography export restriction. ASML's EUV scanners — the only tools capable of printing sub-7nm features in a single exposure — have been restricted from export to China since 2019 under Dutch export control regulations coordinated with US policy. This restriction is not about a single chip or a single product; it is about the manufacturing tool without which no leading-edge chip can be made at competitive volume and yield. Every TSMC N3/N2 chip, every Samsung SF3/SF2 chip, every Intel 18A chip in the world was made possible by ASML EUV scanners. China does not have a single ASML EUV scanner and, under current policy, will not receive one. The entire performance gap between SMIC N+2 and TSMC N2 — approximately five process generations — is maintained primarily by this one export control.

Understanding the EUV control requires understanding what it does and does not prevent. It does not prevent China from manufacturing chips — SMIC has demonstrated 7nm-class production without EUV using aggressive DUV multipatterning, confirming that the control's purpose is not absolute manufacturing prohibition but pace-of-advancement limitation. What EUV controls prevent is China achieving sub-7nm density at the yield rates that make commercial AI accelerator production economically viable. SMIC's DUV multipatterning approach requires more mask layers per wafer (higher cost, lower throughput), achieves lower final density (fewer transistors per mm²), and produces lower yields due to accumulated overlay errors across multiple exposures. The result is that SMIC's 7nm-class chips cost more per die, perform less per die, and are produced less efficiently than TSMC's EUV 5nm chips. EUV export controls do not eliminate China's semiconductor manufacturing capability; they cap its performance ceiling and inflate its production cost — which is sufficient to maintain a commercially significant performance and efficiency gap for as long as the controls hold.

The domestic Chinese response to EUV restriction has two tracks. The first is perfecting DUV multipatterning at SMIC — engineering excellence at maximizing what is achievable without EUV. This has been more successful than most Western analysts predicted: the Kirin 9000S in 2023 proved 7nm-class density is achievable at commercial volume with DUV-only tools. The second track is developing a domestic EUV alternative. SciCarrier (backed by Huawei) and SMEE (Shanghai Micro Electronics Equipment) are working on domestic lithography. SMEE has demonstrated 28nm capable tools. Domestic EUV is qualitatively different from SMEE's 28nm tool — it requires a plasma extreme ultraviolet light source, tin droplet targeting lasers, multilayer optics, and precision mechanics all developed from first principles, representing arguably the most technically demanding manufacturing engineering project in China's history. Independent assessments place genuine domestic EUV capability — not demonstrated, but commercial-volume capable — at 5-10 years away under sustained investment. China's semiconductor self-sufficiency at leading-edge is primarily a lithography problem.

China's Counter-Leverage — Gallium, Germanium, and Critical Materials

The bifurcation is not a one-sided story of Western export controls constraining Chinese semiconductor capability. China has deployed its own supply chain leverage through export restrictions on materials where it holds dominant global production positions. Gallium and germanium — both critical for compound semiconductor manufacturing — are produced globally in significant volumes only in China. Gallium is a byproduct of aluminum smelting and is used in GaN semiconductors, III-V compound semiconductors for defense RF, solar cells, and LEDs. China produces approximately 80% of the world's gallium. Germanium is used in fiber optic cables, infrared optics, and as a substrate for specialized high-efficiency solar cells and some semiconductor processes. China produces approximately 60% of global germanium.

In July 2023, China imposed export restrictions on gallium and germanium, requiring export licenses for compounds and metals of both elements. In December 2024, China expanded restrictions to include antimony (used in flame retardants and some semiconductor processes) and in 2025 extended to graphite (critical for EV battery anodes and some semiconductor processes). These restrictions are calibrated for signaling as much as for commercial effect — gallium and germanium can be produced outside China but at substantially higher cost and with significant lead time to develop new supply. Defense and aerospace programs that depend on GaN-on-SiC or GaAs semiconductors are most acutely exposed because they cannot easily substitute materials or qualify alternate supply chains quickly. The restrictions establish that the bifurcation imposes costs on both sides and that China has a degree of reciprocal supply chain leverage over the Western ecosystem even as the Western ecosystem maintains a larger technology capability advantage over China's domestic supply chain.

The 28nm Frontier — The Next Policy Battleground

The most actively debated supply chain policy question in the US-China semiconductor context as of 2026 is whether to extend export controls to the 28nm node. The current control architecture focuses on sub-14nm processes and the equipment (primarily EUV) and software (advanced node EDA) necessary to produce them. SMIC is aggressively expanding 28nm capacity at Shanghai, Shenzhen, and Beijing, and is doing so with equipment already purchased from Western suppliers before controls tightened. The concern in Washington is that a large, low-cost Chinese 28nm manufacturing base serves the global market for mature node chips — MCUs, analog ICs, RF front-ends, power management — that are critical to automotive, industrial, and consumer electronics supply chains globally. Chinese 28nm chips at SMIC prices could displace Western mature node producers (GlobalFoundries, TSMC's mature nodes, United Microelectronics) commercially.

The policy argument against extending controls to 28nm is that it would be extremely difficult to enforce (28nm equipment is widely deployed globally and cannot be meaningfully restricted without alienating allies who themselves produce 28nm) and that the commercial disruption of global supply chains for mature node chips — where China is already a major supplier — would impose significant costs on US and allied industries without a proportionate security benefit. The policy argument for extension is that allowing China to dominate the 28nm commodity layer of the semiconductor supply chain creates economic leverage and supply chain dependency that is strategically problematic even if mature node chips do not directly enable military capability. The 28nm frontier debate will shape the next phase of the bifurcation more than any development in advanced node technology — because it determines whether the bifurcation remains a capability-tier phenomenon (advanced nodes separate, mature nodes integrated) or becomes a comprehensive stack-level decoupling.

Bifurcation Scorecard — What China Has Achieved and What Remains Constrained

Key Questions — China Semiconductor Bifurcation

Is the bifurcation reversible? At the AI compute layer, almost certainly not under current political conditions — H20 ban has eliminated all legal NVIDIA products for Chinese buyers, Beijing's GPU directive has mandated Ascend adoption, and two years of infrastructure investment on both sides have created ecosystem commitments that would take years to unwind even if policy reversed completely. At the foundry layer, reversal would require ASML to begin exporting EUV to China — politically inconceivable in the current US-Dutch-Japan technology coalition context. At the mature node and memory layers, the bifurcation is less complete and more reversible in principle, but the commercial investment in domestic alternatives makes reversal economically irrational for Chinese companies even if policy permitted it. The operational conclusion is that the bifurcation should be treated as structurally permanent at the advanced node and AI compute layers for strategic planning purposes.

Does China's algorithmic innovation compensate for hardware capability gaps? Partially and increasingly. DeepSeek R1's demonstration that frontier-class AI model results are achievable with dramatically less compute than US labs assumed was necessary — using techniques like mixture of experts, distillation, and speculative decoding that reduce compute requirements — represents a genuine algorithmic response to hardware constraints. If Chinese AI researchers systematically develop algorithmic approaches that require less compute for equivalent results, the hardware performance gap matters less than raw FLOPS comparisons suggest. However, there are absolute limits: training at the frontier (models larger than current state-of-the-art) requires raw compute scale that algorithmic efficiency can only partially offset. China's ability to maintain frontier AI development is bounded by both HBM supply (which constrains how many Ascend chips can be packaged) and by inference serving at scale (which requires the cluster architecture that Huawei is building but at a significant power efficiency disadvantage).

What is the most underappreciated supply chain risk in the bifurcation? The Western tool and consumable dependency of SMIC and CXMT. Both companies expanded aggressively by acquiring Western equipment before controls tightened, and both are running fabs full of Applied Materials CVD chambers, Lam Research etch tools, KLA metrology systems, and Tokyo Electron coaters that require ongoing supply of Western-sourced consumables (photoresists from JSR, Tokyo Ohka, and Sumitomo; process gases; chemical mechanical planarization slurries). These consumables cannot be substituted with domestic alternatives in the near term without significant yield degradation. China's domestic semiconductor manufacturing capability is not nearly as independent as the existence of SMIC N+2 production suggests — it is dependent on ongoing imports of specialty chemicals and spare parts that, if effectively controlled, would constrain SMIC's ability to maintain yield at current levels. The enforcement of consumables controls is the underreported supply chain risk that could matter more than any new equipment restriction in the near term.

Related Coverage

Huawei / HiSilicon Spotlight | NVIDIA Spotlight | Samsung Semiconductor Spotlight | Intel Foundry Spotlight | Process Nodes & Lines | Fab Clusters | HBM Memory | AI & ML Sector | Bottleneck Atlas | Spotlights Hub