- What happened: The Center for Technology & Statecraft (CTS) estimates that cumulative imports of immersion DUV lithography tools for Chinese-owned semiconductor fabs reached 343 units by the first quarter of 2026.
- Why it matters: The bulk of them, the NXT:1980i series, can also be used in 7nm-class manufacturing, leaving China room to expand advanced chip production even without access to EUV.
- What to watch next: How many tools are actually operating and how they are allocated across processes, equipment needed beyond lithography, HBM stacking and advanced packaging capacity, and access to spare parts and maintenance.
On September 26, the US policy research institute Center for Technology & Statecraft (CTS) released a report estimating that cumulative imports of ASML immersion DUV lithography tools for Chinese-owned semiconductor fabs reached 343 units by the first quarter of 2026.
The main group, the NXT:1980i series, which can also be used in 7nm-class manufacturing, accounts for an estimated 270 of them. Because China cannot obtain EUV lithography tools, the authors argue that immersion DUV could still let it expand advanced chip production, and they recommend a complete halt to immersion DUV exports to China.
However, the figure of 343 is not a measurement of tools currently in operation. Nor can it be converted directly into AI chip output. Multiple conditions must be met, including which fabs and processes the tools are assigned to, as well as memory manufacturing, advanced packaging and maintenance.
Breaking down the 343 tools, and what the estimate excludes
CTS counted ASML immersion deep ultraviolet (DUV) lithography tools imported for fabs owned by Chinese companies from 2012 through the first quarter of 2026.
The count covers immersion lithography, in which the space between the lens and the wafer is filled with water to improve resolution when transferring circuit patterns onto wafers with light. Dry DUV tools and Nikon's immersion lithography tools are not included in the 343.
Appendix E of the report estimates median cumulative imports at 396 units including foreign-owned fabs in China, of which 343 went to fabs owned by Chinese companies.
For the Chinese-owned estimate, the central 80% of the distribution falls between 312 and 383 units. This is not a measurement error from counting tools on site. It is a range showing how much the result moves when assumptions used in the estimate, such as tool prices and import classifications, are changed.
| Model group (Chinese-owned fabs) | Median cumulative imports | Main use assessed by CTS |
|---|---|---|
| NXT:1980i series | 270 | Can support 7nm-class logic and DRAM for HBM2e |
| NXT:2050i | 22 | Usable for advanced processes requiring higher precision |
| NXT:2100i | 6 | Usable for the most demanding layers in advanced processes |
| NXT:1970i and earlier | 45 | Older-generation logic and memory |
| Total | 343 | Not a count of tools confirmed to be operating |
Source: CTS, September 2026 report, Appendix E. The count covers ASML immersion DUV tools for fabs owned by Chinese companies from 2012 through Q1 2026. Figures are medians for each category.
Adding up CTS's medians for NXT:1980i and later models gives 298 units (270 + 22 + 6), about 87% of the 343. The ratio is 298 ÷ 343 × 100, rounded to an integer.
It is clear that a considerable number of tools capable of advanced processes have accumulated, but that does not mean all 298 are running in 7nm production lines.
Appendix E itself states that these figures should be treated as "cumulative imports," not "tools currently in operation." Retired and re-exported tools are not removed from the model, and imports before 2012 are not included. Placement at individual fabs is not fully tracked either.
The 343 figure indicates the approximate scale of immersion DUV tools China has acquired to date.
Estimating 343 from earnings reports, customs data and a US House investigation
CTS combined ASML's financial disclosures, Chinese customs statistics and shipment information published by the US House Select Committee on China to estimate cumulative imports.
Not every year was calculated the same way, though. For 2021–2024 it used model-level shipment data obtained by the House committee, while for some periods before and after it estimated from ASML earnings materials and Chinese customs figures.
The House report Selling the Forges of the Future is an investigation compiled from materials obtained from five semiconductor equipment makers, including ASML.
The model-level graph on page 29 of that report shows that shipments of the higher-performance NXT:2050i and NXT:2100i to China fell to zero in 2024, while shipments of the NXT:1980i series surged. A table on page 30 states that 70% of the immersion DUV tools ASML sold in 2024 went to China.
This 70% is not a share of ASML's total equipment sales revenue but China's share of immersion DUV unit sales.
Chinese customs statistics, meanwhile, do not record detailed lithography tool models. Immersion DUV, dry DUV, i-line tools and others fall under the same category, so CTS broke down import volumes and declared values using assumptions about unit prices.
In addition, the province where the customs consignee is registered does not necessarily match where the tool is actually installed. The estimate changes depending on tool prices, item classification and how deliveries to foreign-owned fabs are treated.
The report puts NXT:1980i series acquisitions at about 90 units in 2024 and about 89 in 2025, and estimates spending over those two years at more than $13 billion.
But while 2024 benefits from model-level shipment data, 2025 is an estimate combining financial information and customs statistics. The overall trend of surging purchases should be considered separately from whether individual annual figures can be treated as exact counts.
The House committee also states explicitly that its investigation does not determine whether any laws were violated. Even if sales to restricted entities or equipment imports after export controls took effect were identified, that alone does not mean the exports were illegal.
Why 38nm resolution can still produce 7nm chips
ASML's published NXT:1980Di specifications list a 193nm ArF light source, resolution down to 38nm depending on conditions, and tool-to-tool overlay accuracy of 2.5nm or better.
Because CTS's "NXT:1980i" category groups multiple generations of the series, it would be inappropriate to assume all 270 estimated tools match the NXT:1980Di's performance. Even so, ASML's description confirms that this series supports volume production processes that use multi-patterning.
In immersion lithography, filling the space between lens and wafer with water improves the optical system's resolving power.
Multi-patterning, in turn, forms patterns too fine to draw in a single exposure by splitting them across multiple exposures and processing steps. By combining deposition and etching and aligning each step with high precision, it creates structures finer than a single exposure can achieve.
The name "7nm" is also a label for a generation of manufacturing technology, not a statement that every structure on the chip measures 7nm.
There is precedent for mass-producing 7nm-class semiconductors with DUV.
TSMC began volume production of N7 in 2018, and N7+, which uses EUV, entered volume production in 2019. In its October 7, 2019 announcement, TSMC described N7+ as its first commercial EUV process and said the conventional N7 had already been in volume production for more than a year.
In other words, making 7nm-class chips without EUV is not merely a theoretical possibility but a technique the semiconductor industry has actually used in volume production.
However, the more exposures a layer is split into, the more times the same wafer must pass through the tool, with deposition, etching and alignment repeated each time.
The number of wafers a lithography tool can process per hour cannot simply be counted as finished wafer output. More process steps mean longer processing times and greater difficulty in holding down defects and maintaining high yield.
One advantage of using EUV is not only that it can draw finer patterns, but that it can form complex layers that DUV must split into multiple steps with fewer steps.
TSMC's production record shows only that 7nm-class manufacturing with DUV is possible. It is not evidence that buying the same lithography tool models would let Chinese fabs achieve the same yield or productivity.
Only by putting surrounding processes such as deposition, etching and inspection in place, and continually optimizing manufacturing conditions, can a tool's capability be tied to actual products.
The "100,000 a year" estimate also depends on HBM and advanced packaging
CTS calculates that, in a Chinese fab with sufficient equipment beyond lithography, the exposure capacity of one NXT:1980i could support production of more than 100,000 "Ascend 910C-equivalent" products per year.
Reading this as "one lithography tool can make 100,000 finished AI chips a year" would misrepresent the manufacturing process, however.
The 910C-equivalent product assumed in the Appendix D model combines two 7nm logic dies with eight HBM2e memory stacks.
For HBM, the model assumes eight stacks, each made of eight stacked DRAM dies, so each finished product requires a total of 64 DRAM dies.
Logic and DRAM differ in the number of exposures required, the number of dies obtained from each wafer, and yield. For HBM, the yield of the process of stacking multiple DRAM dies adds a further factor.
Taking these differences into account, CTS divided the number of wafers a lithography tool can process by the number of passes each process requires, then multiplied by dies per wafer and yield to calculate logic and memory output separately.
It then determined how allocating limited lithography tools between logic and memory would produce the most 910C-equivalent products, and also factored in advanced packaging yield.
The figure of "100,000 per tool per year" is therefore a converted value worked backward from the total exposure capacity supporting multiple manufacturing processes.
The biggest assumption is that front-end equipment other than lithography and advanced packaging capacity are sufficient.
The report points out that shortages of etching and deposition equipment are also a constraint in China, and explains that unless these bottlenecks are resolved, adding immersion DUV tools may not increase 910C output.
Being able to expose DRAM circuits is a different capability from stacking them as HBM at high yield, combining them with logic dies and shipping finished products.
Appendix F, which compares production capacity through 2035, also states explicitly that China's figures are an "upper limit on capacity from the lithography tools' standpoint," assuming other process constraints are resolved, and are not a forecast of actual output.
Holding many immersion DUV tools is an important asset supporting room for future production growth. But capacity calculated under conditions cannot be read as the output current fabs are actually achieving.
CTS recommends moving from licensing to a full export ban
The Netherlands introduced export controls on advanced lithography tools in September 2023 and, on September 7, 2024, extended licensing requirements to the NXT:1970i and NXT:1980i.
In a statement on September 6 that year, ASML explained that exports of these tools would require license applications to the Dutch government rather than the US government. Tools from the NXT:2000i onward had already been subject to Dutch export licensing before then.
However, the Dutch government's explanation states that this is a system of case-by-case review for exports outside the EU, not a blanket export ban.
In its September 2026 report, CTS analyzes that a considerable number of NXT:1980i shipments to China continued to be approved even after licensing was introduced.
Whether individual current transactions are permitted cannot be judged from past descriptions of the system alone, but "an export license became necessary" is not the same as "all exports stopped."
What CTS now proposes is to change this licensing system into a policy that in principle halts immersion DUV exports to all of China.
It also calls for restricting maintenance, the supply of spare parts, and the relocation or repurposing of tools already installed at advanced fabs.
This is a policy recommendation from a research institute; publication of the September report did not bring a new full export ban into effect.
The authors' concern is that even tools imported for mature manufacturing processes could later be repurposed for advanced chip production.
The report lists moving tools or key components to other fabs, upgrading manufacturing processes at existing fabs, and moving in-process wafers between fabs as possible routes.
These, however, are analyses of diversion risks that export controls should consider, not evidence that all 343 imported tools were actually diverted to advanced processes.
What makes export controls difficult is that the same lithography tool can be used across multiple generations of chipmaking.
That a model can be used for 7nm-class manufacturing does not mean the buyer is using it for that purpose today. A system that permits exports based on end use would require checking not only declarations at the time of purchase but also where tools are placed after installation and which processes they are used in.
ASML's customer support includes maintenance and upgrades aligned with production schedules, diagnostics that reduce unexpected downtime, and a logistics system that places spare parts close to the tools. When a tool is moved to another fab, planning, materials and support from skilled engineers are also provided.
In other words, a lithography tool is not simple equipment that maintains the same performance for a long time without support once purchased.
Policies restricting shipments of new tools and policies restricting parts and maintenance support for existing tools therefore affect different things.
The former limits lithography capacity added in the future. The latter affects how long, and with how stable a performance, tools already installed can be kept running.
The fact that China has acquired a cumulative 343 tools alone does not allow a judgment that it can maintain the same production capacity going forward without outside support.
To gauge China's capacity to expand advanced chip production, it is necessary to track which fabs the immersion DUV tools actually in operation are placed in, and whether they are used for logic or DRAM.
It also matters whether front-end equipment such as deposition and etching, HBM stacking and advanced packaging capacity are in balance with lithography capacity.
Only when China can secure spare parts and maintenance, and keep shipping products continuously while holding high yield across the entire manufacturing process, will it become clear how far the cumulative 343 imported tools can be tied to actual AI chip supply capacity.