EUV high volume manufacturing is finally a real phrase and not a wish. After more than a decade of slipping dates, ASML has pushed its source past the power production actually needs, and the first chipmakers are committing to run their next nodes on it. I did not expect to be writing this in 2017. For years I argued the light source would keep losing, and for years it did. Then it stopped. The machine that was a science project the last time I wrote about it is now penciled into fab construction plans.

250 watts. That is roughly the source power a fab needs to run EUV at a throughput worth having, and ASML is finally demonstrating it after years of falling short. Getting there took grinding on every part of the source at once: better droplet control, a stronger laser, a pre-pulse scheme that wrings more light out of each speck of tin, collector mirrors that survive the debris longer. No single breakthrough did it. It was hundreds of small fixes piling up until the power curve finally crossed the line a fab can use. With the watts there, throughput climbs into the range where EUV stops being a luxury and starts paying for itself.

Around 125 wafers an hour, more than 100 million dollars apiece: the NXE:3400B is ASML’s first EUV scanner built for production rather than learning, and that throughput would have looked like fantasy a few years back. It is slower than the immersion tools it sits beside, but for the most critical layers of the most advanced chips it removes the need for the towering multi-patterning stacks immersion now demands, the kind that piled onto the 16nm generation. One EUV exposure can replace several immersion passes. That is the whole pitch, and at 125 wafers an hour it finally holds up.

TSMC is designing EUV into its next nodes and aiming to have the tools in production within a couple of years, starting on a handful of layers and widening out, which makes it the most aggressive adopter by a wide margin. Samsung is close behind on its own advanced process. Intel is the quiet one, and that is the strange part: it put more money into this technology than anyone through the co-investment program, and now it says the least about when it will ship a product on EUV. That gap interests me. Intel bankrolled the rescue and looks set to watch TSMC light up EUV first.

Availability is the soft spot now, not capability. A tool that runs beautifully for one shift and then drops offline for maintenance is a real problem when a fab needs it producing around the clock, and EUV uptime is nowhere near immersion’s yet. The pellicle situation is better than it was and still not where anyone wants it. The cost keeps climbing, for the scanner and for everything built around it. So EUV can print production-grade chips, fine. Doing it cheaply and reliably enough to retire immersion across the board is a fight that runs for years, and it gets settled on the fab floor, not in a press release.

There is no second source. No machine to switch to if ASML stumbles or jacks up its prices, no rival scanner waiting in the wings. Nikon and Canon are nowhere close and have effectively conceded the leading edge, so ASML is the only company on earth that can build an EUV tool. For thirty years the quiet safety net in this industry was that you could always buy from the other guy, and at the leading edge that net is now gone. Whoever wants to make the most advanced chips in the world has exactly one supplier for the most important tool in the fab, and it sits in a single Dutch town.

I spent ten years skeptical about EUV, and I was right for most of them. The dates were pure fantasy. The source looked hopeless and the price tag absurd. What I underestimated was ASML’s willingness to keep grinding on an impossible problem long after a normal company would have folded and gone home. In 2017 the impossible problem is mostly solved, and the prize is a position no equipment maker has ever held: sole supplier of the one tool the whole future of computing runs through. I am not sure the industry has fully absorbed what that means. Ask me again the first time ASML raises the price and every customer pays it without blinking.