Five machines this year. Twenty next year. That is the entire initial order book for China’s first homegrown immersion DUV lithography tool, according to an Information investigation that Reuters confirmed on Monday, and it is somehow both the least impressive chip news of the week and the most important.

I have spent a lot of time on this blog arguing that SMIC’s N+3 node proves multi-patterning can squeeze 7nm-class density out of old DUV steppers without EUV. That was always the software half of China’s workaround: cleverness applied to hardware it was allowed to buy. What changes this week is the hardware half. A state-backed manufacturer, still unnamed in the reporting, is now building the steppers themselves, and the prospective customer list reads like a who’s who of the companies I already cover here: SMIC, Hua Hong, and ChangXin Memory Technologies, the same CXMT that just posted a 466 percent pop on its Shanghai debut and briefly became China’s most valuable listed company.

ASML declined to comment, which is corporate-speak for “we saw the same report you did.” Its shares fell more than 7 percent on the news, and the damage wasn’t contained to Veldhoven: BE Semiconductor Industries, Soitec and Infineon all dropped too. That is a strange reaction to five machines. Nobody thinks five tools replace ASML’s global installed base overnight, and the reporting is upfront that the Chinese system still lags in performance and reliability and needs further validation before anyone calls it production-ready.

But I keep coming back to the number that actually matters here, and it isn’t five or twenty. It’s 6,000. That’s how many wafers per day ASML says its NXT immersion tools can push through, and immersion DUV is the part of ASML’s business that generates the volume, not the headlines. The company recognized revenue on 279 DUV systems in 2025, 47 percent of them immersion, worth 12 billion euros in system sales alone. Layer on the 8.2 billion euros ASML pulled from its installed base last year (the parts, the upgrades, the field engineers who fly out when a stepper hiccups) and you start to see why this story isn’t really about EUV at all. EUV remains untouchable under export controls and will stay that way for years. DUV is the workhorse tier that actually runs the volume for cars, industrial gear, communications chips and the less glamorous logic and memory layers, and it’s the tier China just decided it doesn’t need to keep buying forever.

The Dutch government tightened its licensing regime on this exact category back in January 2025, moving to a case-by-case authorization requirement for anything leaving the EU. I wrote at the time that restricting High-NA EUV access would just accelerate whatever domestic alternative China was already building, and I’ll admit I expected that to show up first as a multi-patterning trick, not an actual competing machine. Both happened. That’s the part export-control advocates tend to undersell: the policy doesn’t freeze engineering knowledge, it just changes who has the incentive to replace the thing being restricted. Twenty machines running through 2027, even mediocre ones, generate operating data, reveal component failures, and train a domestic engineering bench. SMIC, Hua Hong and CXMT stop being customers in that process and start being co-developers, whether or not this specific tool ever threatens ASML on a spec sheet.

None of this is the sudden collapse of ASML’s monopoly, and I’m not going to pretend a five-unit pilot run is. What it changes is the clock. ASML’s real moat was never just the optics, it was the decades of service relationships and the fact that nobody else had bothered building an alternative worth maintaining. China just decided that dependency itself is worth paying to end, even at a quality discount, and it’s doing it in the same month CXMT proved the memory side of that same independence push already has real commercial legs. I don’t know yet whether the Chinese tool ever closes the reliability gap enough to dent ASML’s installed base revenue. I do know the five-machines-is-nothing framing making the rounds this week is measuring the wrong year.

Sources