China's Homegrown EUV Gamble: What a Domestic Lithography Breakthrough Could Mean for the Global Chip Race

ASML's extreme ultraviolet (EUV) lithography machines are the most complex manufacturing equipment ever built. Each machine contains 100,000 parts, takes 40 Boeing 747s to ship, and costs approximately $380 million. Without EUV, you cannot manufacture chips at 7nm and below — which means you cannot make the most advanced AI processors, smartphones, or high-performance computing chips.

ASML is Dutch. The Netherlands has banned EUV exports to China since 2019 under US pressure. China has not received a single EUV machine.

This is the central bottleneck in the US strategy to contain China's semiconductor advancement. And China is spending whatever it takes to build a domestic alternative.

Where China Is Today: The SMEE Baseline

China's primary lithography equipment maker is SMEE (Shanghai Micro Electronics Equipment Group). SMEE's most advanced publicly disclosed machine is a 28nm DUV (deep ultraviolet) immersion lithography system — comparable to equipment ASML and Nikon were selling in the mid-2010s.

SMEE delivered its first 28nm production-capable tools to Chinese fabs in late 2023. This is a significant achievement — domestic 28nm production reduces China's dependence on legacy ASML DUV equipment for mature-node chips — but it is a long way from the EUV needed for 7nm and below.

The Huawei 7nm Question: Huawei's Kirin 9000s chip, manufactured by SMIC, was found to use a 7nm process in teardowns of the Mate 60 Pro in August 2023. This created significant confusion in Western intelligence and policy communities — how was SMIC manufacturing 7nm without EUV?

The answer: SMIC is using multi-patterning techniques with existing DUV equipment to approximate 7nm geometry. This is extremely expensive (each wafer requires multiple lithography passes), yields are reportedly low (30-40% vs TSMC's 80%+ at equivalent nodes), and throughput is a fraction of EUV-based production. It is technically a 7nm process but at a cost and volume that makes it uncompetitive for mass market products.

The EUV Development Programme

China's domestic EUV programme is classified. What is publicly known comes from academic papers, patent filings, equipment procurement records, and intelligence assessments:

The physics challenge: EUV lithography uses 13.5nm wavelength light generated by directing high-powered laser pulses at tin droplets, producing plasma that emits EUV radiation. The light source, optical system (using Zeiss precision mirrors — now also export-controlled), and stage system all require components at the absolute frontier of precision manufacturing. ASML took 30 years and approximately $8 billion in R&D to reach current EUV capability.

China's reported progress: Multiple intelligence assessments (US, Dutch, and South Korean) reported in 2024-2025 that China had achieved EUV-range wavelength generation in laboratory conditions but had not demonstrated an integrated, production-capable EUV tool. The gap between laboratory demonstration and production-capable tool is measured in years, not months.

The mirror problem: EUV optical systems require precision multilayer mirrors with surface roughness measured in picometres, manufactured by Zeiss (Germany). Germany has restricted Zeiss exports to China for EUV applications. China is developing domestic mirror capability, but precision optics at this level is not a near-term gap to close.

Timeline estimates: The most optimistic assessments — from Chinese government-linked research publications — suggest a production-capable EUV tool by 2030. Western intelligence assessments typically give a 2032-2035 range for a production tool capable of volume manufacturing. Both timelines acknowledge significant uncertainty.

What China Has Done Without EUV: The SMIC 7nm Path

The more immediately significant story is not the EUV programme — it is what China has accomplished without EUV.

SMIC's N+2 process (marketed internally as 7nm equivalent) uses a technique called self-aligned quadruple patterning (SAQP) with DUV immersion tools. Each layer requires four separate lithography passes instead of one EUV pass. This multiplies equipment cost, time-per-wafer, and defect probability.

Production volumes reported in 2024-2025: SMIC manufactures approximately 3,000-4,000 Kirin 9000s wafers per month at N+2. TSMC manufactures approximately 100,000-120,000 A17 Pro wafers per month at N3 for Apple. The volume gap is approximately 30x.

For Huawei's purposes — making enough chips for its premium smartphone and AI accelerator products — SMIC's 7nm capacity is sufficient. For competing with TSMC in high-volume consumer electronics or AI hardware at scale, the cost and yield disadvantage makes SMIC N+2 noncompetitive on the open market.

GLM-5 and the Ascend Proof of Concept

The most concrete demonstration that China's domestic chip ecosystem can train frontier AI models is the GLM-5 achievement: Zhipu AI trained a 744 billion parameter model entirely on Huawei Ascend 910B clusters in 2025, achieving near-parity with Western frontier models on standard benchmarks.

Ascend 910B is manufactured at SMIC using 7nm-equivalent DUV multi-patterning. The fact that it can support frontier model training — while less energy-efficient than TSMC-fabbed Nvidia H100 — demonstrates that China's domestic semiconductor path can support leading-edge AI development.

DeepSeek's decision to optimise V4 for Huawei Ascend hardware (rather than Nvidia) is the strategic statement: China's top AI labs are committing to the domestic hardware stack, validating Huawei's chip programme with real workloads.

Export Control Escalation: The US Response

The US has progressively tightened semiconductor export controls on China:

• October 2022: A100, H100 restricted
• October 2023: H800, A800 restricted (the workaround chips Nvidia designed for China)
• April 2024: H20 restricted (the next-generation workaround chip)
• 2025-2026: Restrictions expanded to cover EDA (electronic design automation) software, advanced packaging equipment, and components used in advanced chip testing

Each escalation has pushed Chinese developers and fabs deeper into the domestic ecosystem. The irony: aggressive export controls are accelerating China's domestic semiconductor investment rather than preventing it.

The Dutch government, under US pressure, extended export control licensing requirements to ASML's DUV immersion tools (the ones SMIC uses for 7nm multi-patterning) in January 2023. This does not immediately stop SMIC — they already have the tools they need — but it prevents SMIC from purchasing additional DUV capacity to scale volume production.

What This Means for Developers and Procurement Teams

Two AI hardware supply chains are emerging. The US-centric chain (TSMC-fabbed Nvidia/AMD/Google/Apple chips) and the China-centric chain (SMIC-fabbed Huawei Ascend, with lower but improving capability). Any developer or enterprise planning AI infrastructure over a 5-year horizon should model scenarios where they need to operate in one or both ecosystems.

China will reach volume 5nm-equivalent production by 2028-2030 without EUV, using progressively refined multi-patterning techniques. This is sufficient for many AI workload requirements. The gap with TSMC N3/N2 will persist but narrow.

EUV remains the critical bottleneck. Even the optimistic Chinese government timeline puts EUV production capability at 2030. Until then, China cannot manufacture at TSMC N3 equivalent node costs and yields. This is the strategic window the US export control strategy is buying — and the question is what happens inside that window.