In mid-August, process engineers at Samsung Electronics Co. (SSNLF) stood on stage in Seoul and said the industry’s most advanced chipmaking machine still was not ready for their factories.

In the week of Sept. 7, Samsung and Taiwan Semiconductor Manufacturing (TSM) signed up for it anyway. That kind of reversal rarely happens this fast in an industry that plans in decades, not quarters.

ASML Holding (ASML), the only company in the world that builds extreme ultraviolet lithography machines, confirmed the commitments in a joint statement with TSMC published September 8.

Samsung will use the newest High-NA EUV systems for memory chips starting in 2028. TSMC will follow for logic chips in 2030, according to Bloomberg, joining Intel Corporation (INTC) as the third confirmed customer.

Each machine costs roughly $400 million, according to CNBC. That is the exact price Samsung and TSMC cited in August, when both said they would hold the technology back until nodes near 2030. Three weeks later, they signed a formal joint commitment instead.

Samsung’s timeline matters beyond the lithography roadmap. The company will bring High-NA EUV into memory production as a global shortage of memory chips, driven by AI data center demand, pushes prices higher across electronics, according to Bloomberg.

That shortage gave Samsung a reason to formalize a date instead of leaving it open-ended.

Related: US, Japan $550 billion deal leaves AI chip stocks guessing

The objection was never about the technology

The delay was never about whether High-NA machines work. Intel already settled that question by running the systems in production, according to CNBC.

The real question was whether the economics justified the price, since TSMC and Samsung compete on cost per chip, not on owning the newest tool.

That is why the second half of the announcement matters more than the headline. ASML, TSMC and Samsung also agreed to shift from six-inch photomasks, the stencils that print circuit patterns onto silicon, to a larger 12-inch format.

12-inch masks enable greater scanner productivity and allow the industry to meet the demand for smaller, faster, and more energy-efficient chips.

ASML Chief Technology Officer Marco Pieters noted that the bigger masks could lift machine output by 40%. That means fewer machines are needed to produce the same number of chips.

A $400 million machine looks very different once it can print 40% more wafers. That figure, not the sticker price, is what determines what a chip actually costs to make.

The mask initiative, more than the lithography tool itself, appears to be what moved TSMC and Samsung off the sidelines.

The near-term roadmap, as confirmed in the companies’ announcements, breaks down like this:

  • High-NA EUV systems are already in production at Intel, which took delivery of the equipment before either rival, according to CNBC.
  • DRAM manufacturing with High-NA EUV is set to begin at Samsung by 2028.
  • TSMC is targeting logic chip production with High-NA EUV by 2030, with a 12-inch mask pilot line to follow by 2031.
The shift to larger 12-inch photomasks allows chipmakers to increase wafer output by 40%, fundamentally changing the economics of $400 million EUV machines.

Oleh Stefaniak / Getty Images

Intel is finally ahead instead of behind

Intel’s position here inverts its own history. The company delayed adopting ASML’s original EUV machines in the 2010s, a decision that let TSMC pull ahead and eventually forced Intel to outsource its most advanced chips.

This time, Intel moved first and is the only company already running High-NA systems in volume production.

For a foundry business still trying to rebuild, being first matters. Intel is using its High-NA lead to court outside customers who need leading-edge capacity.

It is one of the few technical advantages Intel can currently claim over TSMC.

TSMC’s exposure to ASML is bigger than it looks

ASML trades on Nasdaq and Euronext Amsterdam under the ticker ASML, and the stock has climbed roughly 120% over the past year, according to CNBC.

TSMC alone accounts for about 16% of ASML’s revenue, according to data compiled by Bloomberg. That concentration cuts in both directions.

More Tech:

It gives ASML unusual visibility into future demand whenever a customer of TSMC’s size signs a multiyear commitment. It also means ASML’s growth is tied to the capital spending decisions of three companies, since Samsung and Intel round out its largest customers.

A slowdown at any one of them would show up quickly in ASML’s order book.

AI chip capacity is a decade-long project

The most overlooked detail in this week’s announcement is the timeline. ASML and TSMC are targeting a pilot line for 12-inch masks by 2031, with full production readiness by 2033, according to the companies’ press release. That is seven years from now.

For investors who treat AI chip shortages as a problem that more capital can fix within a year or two, that timeline is a useful correction.

The equipment that will make next-generation AI chips affordable is still being designed, and the companies that build it are only now agreeing on its format.

The bottleneck in AI hardware is not closing soon. It is being built for the 2030s, one commitment at a time.

Related: Why is Taiwan hiding the backers of its $20B US pledge?