TSMC's refusal of ASML's expensive High-NA EUV equipment, explained

[user nl]

ASML has launched its 0.55 High Numerical Aperture Extreme Ultraviolet (High-NA EUV) in an effort to extend Moore's Law. The market had originally expected TSMC to adopt it first, but the company has held back. TSMC Senior Vice President of Global Business Kevin Zhang stated at the North America Technology Symposium that there are currently no plans to introduce High-NA EUV before 2029, mainly because "it's too expensive!" This decision also reflects how TSMC is shifting competition focus from equipment to process integration and cost efficiency.


ASML's next-generation High-NA EUV equipment was unexpectedly first adopted by Intel, while Samsung Electronics has also expanded procurement, yet TSMC(2330.TW) has taken the opposite approach. According to plans made between TSMC and ASML several years ago, High-NA EUV was originally expected to be introduced in the A14 node around 2028. However, after multiple rounds of discussions over the past two years, TSMC Chairman C.C. Wei led senior executives to visit ASML's headquarters in the Netherlands in mid-2024, with both sides even publicly sharing photos of the visit.

Half a year later, ASML's CEO personally visited Taiwan to meet TSMC, but pricing negotiations still failed. Each High-NA EUV system costs as much as EUR350 million (approx. US$410.3 million). Given the high R&D costs, if ASML were to concede on pricing, other customers would likely demand similar discounts.

.....................................................................................

The third factor is TSMC's dominance in the foundry industry, including its customer base, capacity, and profitability. As competitive barriers rise, Samsung and Intel have invested heavily in High-NA EUV to surpass competitors through advanced equipment. However, this strategy may instead weigh on their profitability. Without sufficient order volumes to support these investments, the massive capital expenditures for new fabs and equipment depreciation may take a long time to recover.

If yields do not improve, adopting High-NA EUV could further strain the financial positions of Samsung and Intel. TSMC's wait-and-see approach has ultimately become its greatest advantage.

The pressure now shifts back to ASML. Although ASML remains confident about the future, it will lack major High-NA EUV orders from TSMC over the next three to four years, while the Chinese market continues to shrink. For now, orders from logic and memory customers expanding production are still providing support. However, if the memory industry stabilizes and Samsung and Intel also reassess and follow TSMC in delaying High-NA EUV purchases, ASML's operations could face a significant impact.

https://www.digitimes.com/news/a20260427PD204/tsmc-asml-high-na-euv-equipment.html

 

[Xebec]

Hmm, if TSMC actually has plans to introduce High NA EUV in 2029, shouldn't they already be buying machines?

 

[user nl]

Hmm, if TSMC actually has plans to introduce High NA EUV in 2029, shouldn't they already be buying machines?

I think ASML managed to reduce the leadtime to 12-18 months. Even though they did not deliver more than 45 low NA EUV annually in 2024-2025, ASML kept producing more(modules) so that they can reduce the lead time. Remember many modules of the low NA and High NA are merging to common shaired modules, so that around 2029-2030 only the mirror module is different between low and high-NA.

My feeling is that TSMC has everything under control now up to 13A. TSMC seems to be quite happy with their EUV pellicle R&D progress. I think TSMC goes all in for maximum wafer output in all their new and old fabs to satisfy the huge demand for 2nm and 3nm wafers till 2030.

 

[hist78]

EUV tools and all the fabs’ land, buildings, and machines are part of Property, Plant, and Equipment (PP&E) in accounting. Here I compare PP&E between Intel and TSMC. Although both have similar PP&E, the results they deliver are very different.

TSMC’s performance has been very consistent over this 10‑year period (2016–2025), while Intel is facing a difficult decline. Intel must improve its cost structure and efficiency with rigorous financial discipline as soon as possible. The challenge is how to do this under so many existing or historical high‑cost conditions that have been in place for many years.

Intel in blue color, TSMC in green color

Net PP&E (USD billions)

Revenue per $1 of net PP&E

Net profit per $1 of net PP&E

 

[user nl]

TSMC’s performance has been very consistent over this 10‑year period (2016–2025), while Intel is facing a difficult decline. Intel must improve its cost structure and efficiency with rigorous financial discipline as soon as possible. The challenge is how to do this under so many existing or historical high‑cost conditions that have been in place for many years.

My feeling is that INTEL will only invest in new tools when they see that profitability is there. It seems they have Fab shell clean room space enough for the coming 3-4 years. They will slowly buy tools and slowly ramp 18A and probably 14A in a few years, while "milking" their CPU products as much as possible to generate cash. And IFS will probably buy more low-NA-EUV tools compared to high-NA-EUV tools, as low NA-EUV has become so much more profitable tool-investments with the rapid improvement of source power towards 800-1000 Watt the coming 3-4 years.

I do not hold Intel Foundry Services at the same performance level as a business unit like TSMC. IFS has INTEL products to balance P&L and provide cash for operations. SAMSUNG Foundry has SAMSUNG Memory products to subsidies operations of Foundry and provide cash.

And everyone IFS, SAMSUNG Foundry-TAYLOR and TSMC-ARIZONA benefit from the 35% prepayment cash checks provided by Uncle Sam / Main Street.

IFS is a US National Security Asset so will obey USG/Congress policies and strategy. USG/Congress will never let IFS go down the drain. Just my feeling.

 

[Jim351]

I saw X-Light present on April 16 at the TechInsights Critical Materials Conference in Portland. Their concept is really only interesting if one were building a clean sheet, large scale fab with many EUV tools. There is only one project like that that I know of; TeraFab. Their EUV 13.5nm (or even higher frequency) utility-scale photon source will also require cooperation with ASML since the only way this makes sense is to buy the tools stripped of their individual, still underpowered Tin laser sources. This looks to be the play for Elon....

 

[hist78]

I saw X-Light present on April 16 at the TechInsights Critical Materials Conference in Portland. Their concept is really only interesting if one were building a clean sheet, large scale fab with many EUV tools. There is only one project like that that I know of; TeraFab. Their EUV 13.5nm (or even higher frequency) utility-scale photon source will also require cooperation with ASML since the only way this makes sense is to buy the tools stripped of their individual, still underpowered Tin laser sources. This looks to be the play for Elon....

xLight is looking for its first commercial system operational by 2029. It's a very aggressive and optimistic timeline.

 

[Daniel Nenni]

ASML has launched its 0.55 High Numerical Aperture Extreme Ultraviolet (High-NA EUV) in an effort to extend Moore's Law. The market had originally expected TSMC to adopt it first, but the company has held back. TSMC Senior Vice President of Global Business Kevin Zhang stated at the North America Technology Symposium that there are currently no plans to introduce High-NA EUV before 2029, mainly because "it's too expensive!" This decision also reflects how TSMC is shifting competition focus from equipment to process integration and cost efficiency.


ASML's next-generation High-NA EUV equipment was unexpectedly first adopted by Intel, while Samsung Electronics has also expanded procurement, yet TSMC(2330.TW) has taken the opposite approach. According to plans made between TSMC and ASML several years ago, High-NA EUV was originally expected to be introduced in the A14 node around 2028. However, after multiple rounds of discussions over the past two years, TSMC Chairman C.C. Wei led senior executives to visit ASML's headquarters in the Netherlands in mid-2024, with both sides even publicly sharing photos of the visit.

Half a year later, ASML's CEO personally visited Taiwan to meet TSMC, but pricing negotiations still failed. Each High-NA EUV system costs as much as EUR350 million (approx. US$410.3 million). Given the high R&D costs, if ASML were to concede on pricing, other customers would likely demand similar discounts.

.....................................................................................

The third factor is TSMC's dominance in the foundry industry, including its customer base, capacity, and profitability. As competitive barriers rise, Samsung and Intel have invested heavily in High-NA EUV to surpass competitors through advanced equipment. However, this strategy may instead weigh on their profitability. Without sufficient order volumes to support these investments, the massive capital expenditures for new fabs and equipment depreciation may take a long time to recover.

If yields do not improve, adopting High-NA EUV could further strain the financial positions of Samsung and Intel. TSMC's wait-and-see approach has ultimately become its greatest advantage.

The pressure now shifts back to ASML. Although ASML remains confident about the future, it will lack major High-NA EUV orders from TSMC over the next three to four years, while the Chinese market continues to shrink. For now, orders from logic and memory customers expanding production are still providing support. However, if the memory industry stabilizes and Samsung and Intel also reassess and follow TSMC in delaying High-NA EUV purchases, ASML's operations could face a significant impact.

https://www.digitimes.com/news/a20260427PD204/tsmc-asml-high-na-euv-equipment.html

TSMC asked customers and they said no hurry on HNA-UEV. The risk/reward is not there yet. Simple as that. This topic came up in the press Q&A. I knew it would cause a media bubble. I'm wondering how may of the medias friends and family shorted ASML stock?

Things may change when Intel 14A is deemed the best PPA of the class but today that is not the case and the capacity/yield certainly will not be there.

Even though, it will take years for Intel or Samsung to ramp up HNA-EUV capacity and yield required for the AI surge so I see little risk here for TSMC to lose the lead and more importantly the margins.

As it stands, TSMC won 3nm, 2nm, and now they are looking good at 14. Lets see what A12 brings us. Will 12 be the HNA-EUV sweet spot? ASML will be the truth bearer on this one. How many TWINSCAN EXE:5200 system can ASML deploy by 2030?

 

[Fred Chen]

Their EUV 13.5nm (or even higher frequency) utility-scale photon source will also require cooperation with ASML since the only way this makes sense is to buy the tools stripped of their individual, still underpowered Tin laser sources.

That's right, xLight really needs ASML buyoff.

 

[user nl]

I saw X-Light present on April 16 at the TechInsights Critical Materials Conference in Portland. Their concept is really only interesting if one were building a clean sheet, large scale fab with many EUV tools. There is only one project like that that I know of; TeraFab. Their EUV 13.5nm (or even higher frequency) utility-scale photon source will also require cooperation with ASML since the only way this makes sense is to buy the tools stripped of their individual, still underpowered Tin laser sources. This looks to be the play for Elon....

ASML has been following the FEL technology for at least 2 decades or so I think. ASML follows many novel EUV source technologies like HHG and others.

My feeling is that the rapid improvement at ASML of the Tin LPP source from around 100 Watt to some 1000 Watt, has made the FEL EUV Technology somewhat "too late". Indeed the EUV source is "just one part/module" of the EUV lithography tool.

As a fist thought, if you have a single FEL beam line serve multiple FABs, all EUV tools will go down if that single FEL beam line is off line for maintenance / failure. They try to build/have a second FEL beam line to mitigate that, but it means you build half of your investment just for redundancy. For me too much of a single source of failure for running 4 Fabs with say 20-30 EUV tools.

And why would ASML cooperate with this new project? ASML is too busy scaling its production to 80-100 EUV tools/year production/shipment.

Seems too much a "Make America Great Again" project supervised by PG:
https://www.xlight.com/about#executive-leadership

Just my feeling looking at the Physics/Technology and the Business implementation by the major global EUV-Fab users. Maybe Musk can burn some billions on this FEL-project together with PG.

 

[Xebec]

TSMC asked customers and they said no hurry on HNA-UEV. The risk/reward is not there yet. Simple as that. This topic came up in the press Q&A. I knew it would cause a media bubble. I'm wondering how may of the medias friends and family shorted ASML stock?

Things may change when Intel 14A is deemed the best PPA of the class but today that is not the case and the capacity/yield certainly will not be there.

Even though, it will take years for Intel or Samsung to ramp up HNA-EUV capacity and yield required for the AI surge so I see little risk here for TSMC to lose the lead and more importantly the margins.

As it stands, TSMC won 3nm, 2nm, and now they are looking good at 14. Lets see what A12 brings us. Will 12 be the HNA-EUV sweet spot? ASML will be the truth bearer on this one. How many TWINSCAN EXE:5200 system can ASML deploy by 2030?

Isn't 14A still "vanilla" EUV? I thought it was 10A targeting High NA EUV.