I believe Intel has a winner with the upcoming Crescent Island GPUs with up to 480GB of LPDDR5 memory.
A lot of SMEs will need local AI with large open weights models (i.e. >1.5TB of VRAM), but they don't need a watercooled 8x Nvidia GB300 box. Besides, it's difficult to deploy and very expensive.
Crescent island is air cooled and likely fast enough for a group of users running something like GLM 5.2 or Kimi K3.
My general take is all of The GPU design efforts are going toward Enterprise / datacenter graphics cards. They are just skipping a generation or two of consumer.
Druid, the 4th generation, has been shelved entirely, and all consumer DGPU products, possibly _all_ DGPU products, have been killed for Celestial.
Its likely the only products shipping with Celestial will be IGPUs for the next generation and a half, until Serpent Lake comes out with RTX graphics tiles from Nvidia, and then never another Xe product ever again.
To Nvidia the biggest threat is a capex cut from big tech not another company in the same domain, maybe some Chinese companies but people aren't going to be moving off of Nvidia for large deployments in the short term even if some TPU/Tranium etc is used it's not close to Nvidia.
I have no explanation other than people, researchers are comfortable with it and don't want to switch.
Nvidia's software support remains "best in class" so they've become the default for a lot of this stuff. To overcome that I think you'd have to get a lot of hobbyists and by extension interested young people using your hardware instead of Nvidia. So you'd have to have good software support, decent hardware, and low hardware cost. People started with Nvidia because most machines had an Nvidia card in them, and Nvidia's software support was carried-over momentum from how they've supported their GPUs with PC gaming. I've tried AMD, for example, and the compatibility matrices and bugs and just general lacking software support for ML on their hardware makes it a total non-starter. It may be a lot better with current-generation stuff but it's going to take a lot for me to trust AMD for ML.
True, Given I am young, hobbyist, and have written a lot of AI libraries for myself and small companies for local deployments, and can't afford any serious Nvidia hardware I will switch to whoever can get me a system with enough VRAM for a decent price under 1000$...
My best GPU is 5070Ti on which the best model I can run is Laguna S 2.1 118B with NVFP4 and offloading most experts on CPU with an expert router layer and 128k context.
I could kill for a 1000$ AMD or Intel card with 32 or 48GB of VRAM even GDDR6 around a 1000$ range but it feels like no one even cares.
With ~4 48GB cards I could seriously locally deploy most open models models with some SSD offloading and good 4bit quantizations with ok context sizes. But at current prices.. sigh... I tried begging Intel to maybe it's cards available in my region but alas no such luck especially not at reasonable prices.
Honestly getting an AI subscription feels cheap atm and working hard at solving and home labing stuff is insane.
Even in research no one cares about anything but Nvidia, because at this point they are expensive but they seem to care, I just don't see others caring though I have worked with Intel gpu and compute teams on implementing this stuff and they are very enthusiastic but well the companies themselves aren't solving anything for me as an individual, but they are begging for me to add support for their stack in software I maintain. The Irony is unreal, why should I even add support when no one can even use it.
Only company that had decent stuff was surprisingly Apple but their stuff is now too expensive as well. .sigh.
This should have been obvious to me before but for whatever reason reading your comment made me realize something.
The hardware that nvidia is selling can be used against them because it can generate the code that is lacking for hardware from other companies.
So there’s a world out there where Intel or AMD just generate the needed software with compute from rented Nvidia hardware or the end consumer does the same, akin to opensource volunteers spending their free time reverse engineering undocumented hardware for the last few decades.
That’s if the claim that other companies produce acceptable hardware that is mostly just kneecapped by poor software.
If that’s the case there’s a distinct chance that we’ll see feature parity between the different vendors soon enough.
Unless the models that run on nvidia hardware are inadequate for this task, but that sort of raises a catch-22 — if the models aren’t good enough to generate drivers and CUDA type software what are they good enough for?
> To Nvidia the biggest threat is a capex cut from big tech
Not the biggest. Maybe top 3.
The bigger issue right now is data center expansion speed. You have SpaceX turning up random generators and doing whatever to get it over the line but that's the exception. There are likely stockpiles of GPUs and racks of AI clusters waiting to go online. It's only going to get worse.
It doesn't matter anymore. Nvidia's core product isn't "GPUs" anymore. It's a whole AI rack. AMD also announced their own of the same. CPU, GPU, network, etc is all bundled.
A 4% stake doesn't even make Nvidia one of the top 3 holders of Intel stock... They'd be crucified for changing strategy for a 4% stake.
Also Nvidia arguably needs Intel more than vice versa, Intel now has SOTA fabs on the newest EUV technology, TSMC is at max capacity and they need to fight Apple and others for allocation...
Amazing looking machine! $380 million a pop. Must surely be one of the most sophisticated machines humanity has ever built? A great achievement of European science and technology (ducks).
> A great achievement of European science and technology
It's an amazing piece of technology but this is... wildly wrong. ASML is a multi-national company that licenses IP largely from USA and Japan, but also Taiwan and Germany. The actual EUV light source is developed and produced in California by Cymer, which ASML acquired in 2013. But ASML was only permitted to acquire the company under a strict technology sharing and export control agreement with the US government. Additionally, a huge portion of the photolithography research is directly developed (and owned) by US companies and research organizations such as IBM, Albany NanoTech, and SEMATECH.
There is a reason why ASML's next-generation research photolithography machine is currently being installed and developed in upstate New York, and not somewhere in the Netherlands. The same reason that Cymer is still in San Diego instead of being relocated to Europe. Because it's American technology.
> There is a reason why ASML's next-generation research photolithography machine is currently being installed and developed in upstate New York, and not somewhere in the Netherlands.
This is a wildly ignorant comment. The R&D lab in NY State if anything is an effort to copy IMEC, the world's leading research organization when it comes to semiconductor R&D. They essentially set and define the roadmap for node advancements and then the industry follows by actually implementing their innovations at scale in their fabs. I guess they are technically not in the Netherlands, being located about 50 miles south of the Dutch border in Belgium. And they have in fact have an EXE system of their own, and have had access to ASML's high-na tools since before they were even commercially available
Agreed. Berkeley / CXRO is the American beyond-NA research center not Albany NanoTech which doesn’t have a charter AFAIK. They are just beginning to install the EXE:5200 in Albany now..
They do have the xLight program which doesn’t have a true European equivalent but still is a strategic partner of ASML who have experimented with FEL in the past.
> There is a reason why ASML's next-generation research photolithography machine is currently being installed and developed in upstate New York, and not somewhere in the Netherlands. The same reason that Cymer is still in San Diego instead of being relocated to Europe.
Why is the Netherlands government even allowing this?
The supply chain both in parts for their machines, but also in terms of research and technology for new machines is very deep and very international. If the US states that a technology can only be used by ASML if they keep R&D in the US, then there is no easy alternative for ASML.
That is true not only for ASML, but for the semiconductor industry in general. The number of smaller, specialized companies for certain tools, chemicals and software is staggering. And they sit all over the US, Europe, Japan, Taiwan and Korea.
The entire multinational effort is in US-Europe-E.Asian countries which are a coordinated allied bloc. The Netherlands allows it because they are entirely aligned.
Strong mutual defense agreements cover four out of the five crucial countries here (NATO, Anpo, MDT US/ROK) and the last (Taiwan) has arms sales assurances.
It’s not like these are strongly opposed nations or anything. Part of the same fleet in some sense.
Russia invaded in 2022, Trump was elected in late 2024 and took office in 2025. The out of context quote that kicked all this off was in 2017 or 2018. Somehow you picked the one year out of the last 10 or so that didn't matter for any of this.
The really wild thing about this whole narrative is how unlikely this entire sequence of events was. Let's review:
1) Trump gives some rambling answer to an Estonian journalist in 2017 (or 2018) in which he says about 4 different things, all of which conflict with each other.
2) That journalist clips the answer to make it sound like Trump wants to leave NATO.
3) European leaders lose their mind and for the first time in 35 years actually up their military spending. Something that every US president since Bush Sr has made a major foreign policy goal that ever single administration failed at.
4) So completely by mistake, a journalist nobody has ever heard of, handed Trump the biggest foreign policy win for any US president since Bush Sr.
5) Russia invades Ukraine
6) Sweden and Finland join NATO
7) Trump wins the 2024 election and cuts off some of the Ukrainian aid (but not intelligence sharing).
8) The removal of US limitations on Ukraine allows them to strike Russian oil infrastructure.
9) Russia can no longer afford to fund their war in the way they had been.
10) Ukraine is now actually winning.
So somehow, Trump has gotten a huge foreign policy win and helped turn around the war in Ukraine completely by accident. Its like watching a cow talk.
Not by accident. By the immense sacrifice of the Ukrainian people and it's armed forces coupled with US goals just happening to align on Iran. I somehow doubt the Ukrainians mind very much that shahed production is now pivoted towards targets closer to Iran than to Ukraine.
I mean Trump didn't intend to help. But somehow NATO is in a better state and Ukraine is winning. Its only in the largely imaginary political narrative where NATO is somehow weaker. What is weaker is the credibility of several European leaders. And nobody here thinks that this was some master plan or 10D chess by Trump.
Because ASML acquired the US company that created this tech and the agreement for allowing Netherlands to do so was that the tech and know how remain in the US.
The EUV lithography process and other pieces that comprise what ASML sells was developed with US government funding, and it remains encumbered by that. The US government thus has veto power over the tech, and thus over ASML.
> The actual EUV light source is developed and produced in California by Cymer, which ASML acquired in 2013.
Cymer was one of multiple vendors (Gigaphoton, XTREME) providing this sort of light source, and ASML acquired it because it couldn't achieve what they wanted (they needed 100+W and Cymer could only achieve 15W). ASML needed to control and fund the company, and add engineering talent, to achieve what they actually needed.
Oh and ASML bought Cymer because Gigaphoton (another company at basically the same stage of development as Cymer), being a Japanese company, objected to foreign control. Though if ASML knew the protectionist, insular direction the US was going to go they probably would rethink their past decisions.
So this retconning that Cymer is really the secret magic of ASML and the Netherlands is just stealing US ingenuity is hilarious stuff.
And for that matter, it is amazing how so often we hear about "American" tech that is actually overwhelmingly combining acquisitions and manpower from abroad. Like is an iPhone an "American" creation? LOL, no.
> the secret magic of ASML and the Netherlands is just stealing US ingenuity is hilarious stuff.
I never said anything about ASML stealing technology or ingenuity. The point is that EUV photolithography is an international effort. The machines don't exist at all without about 6 different countries contributing technology and expertise.
ASML has certainly proven itself to be a very capable integrator of many international technologies and processes.
reason why US didn't produce most of these component because US gov didn't continue investment so privately owned enterprise take over from there namely ASML
that also reason why US can restrict EUV sales to china
isnt DUV and EUV technologies discovered and developed by US national labs (LLNL, LBNL, Sandia) in cooperation with EUV Consortsium funded by US Chipmakers (Intel, AMD, Motorola) and licenced to Cymer?
EUV was "invented" (demonstrated) in Japan first, though those US labs did a lot of work on the overall lithography system and reflective optics to put it into practice. The US DOE along with the manufacturers made a consortium (EUV LLC consortium) to commercialize/scale that Japanese technology because they could obviously see the promise, and Cymer is a member, as was ASML already via SVG.
Simultaneously Japan created its own initiative (EUVA), and Gigaphoton didn't rely upon the US initiative at all. And again in an alternate universe -- one where Japan isn't protectionist and Cymer wasn't out of financial runway and desperate -- we'd be talking about how ASML is really Japanese tech, etc.
Also Japan has a monopoly on essential EUV materials like photoresists and mask blanks and pellicles. Japan is also the only country in the world that can produce a number of highly specialized ancillary machinery like the actinic light systems used to inspect EUV photomasks. ASML also acquired its e-beam inspection tech from a Taiwanese company. There's also no way ASML could run anything at production if it wasn't for TSMC and Samsung.
Not to mention the critical role of the US in its development. The core EUV light source was invented in San Diego!
LOL, no Mention of IMEC(Interuniversity Microelectronics Centre) in the EUV development process and the only reason that the US can restrict the ASML EUV equipment is the EUV Light source and mirror optics for that is partially US produced and the IP there falls under US IP/Export restrictions!
From ASML:
"ASML’s extreme ultraviolet (EUV) light source was developed and is manufactured at its facility in San Diego, California. Acquired from the US firm Cymer in 2013, the R&D team here developed the laser-produced plasma (LPP) system that fires lasers at tin droplets 50,000 times per second to generate EUV light."
As with most of this type of technology, most of the stuff you see is basically a giant refrigerator and giant vacuum cleaner. The actual meat is much smaller and hidden from view.
ASMLs machines can easily be "one of the most sophisticated" without stepping on LHC or ITERs toes. I'm not sure ISS qualifies anymore. I would have went with JWST.
neither of the observations gp made is dependent upon the definition of machine, nor is it dependent on the inclusion of the examples you brought up in that definition.
Not even that, the company happens to be Dutch but it's employees are from everywhere. A company doesn't make anything, the employees do, especially in a sector like this.
The companies job is to hire well and create strong teams that work together.
That's what makes this hard to interpret. The whole strategy of intel is that they have their own version of this machine, part of their "18A" design. The only reason they need the ASML machine is because of the "optics" (between quotes because at the frequencies they work at no optics work - only (bad) mirrors do. However you need make the mirrors accurate enough - we're talking far more accurate than the chips, as in sub-nanometer max deviation, they can function as optics with high losses, but low enough that exposure works). No American company can achieve even 10 times lower accuracy at the moment.
So is this giving up on their own version? That would mean their stock price is absolutely not justified at $110, and it should be more like $60-70 if not lower. Or is this "we have technical reasons to use this. We're still going to do 18A and it's still going to beat 2nm", which would mean their stock price is easily justified at $200 and it's high time to buy.
Intel does not have their own version of this machine. This ASML machine is a component in a very large system of machines from many manufacturers that runs wafers automatically from machine to machine until a month later a chip comes out the other end. That system is “18A”. Intel designs, builds and operates that system. Intel does not make the machines, they buy them and figure out how to configure and operate them as a part of the system.
As I understand it, the machine also needs lots of 'normal parts' and can be designed differently to fit the factory and the automation. Like if you need a pipe, instead of buying it from ASML, you can buy it from a local hardware vendor. And parts of the ASML also break, and those parts would be directly shipped from the vendor that makes the part. ASML manages the integration between these parts and each part can have multiple vendors that makes them according to the specs.
Only the optics is an ... interesting way of phrasing it.
There was a great Economist article a while ago that talks about this. ASMLs moat isn't just the know-how to build the machines themselves. It's a network of suppliers all across Europe, specialized businesses doing one little part better than anyone else in the world.
This is the type of thing that Europe is still leading in, and it'll be incredibly hard to reproduce elsewhere.
You don't just need to figure out how to build an High-NA EUV machine. You need to figure out how to build or procure all of the parts that go into it. Many of these suppliers have exclusive contracts with ASML.
Yeah. ASML doesn't make the mirrors, but they have exclusive sales rights, exactly as you say. And yes, positioning them, for example, both in the fixed sense (mounting the mirrors in the machine) and in the dynamic sense (rapidly positioning wafers to nanometer-or-better accuracy below the beam, I forgot how quick exactly but it's 10s to 100s of times per second to maintain production rates) is very much not easy either. But it's "been done" (by teams of phds). Producing the mirrors has not been duplicated.
Of course, producing the EUV beam in the first place has also not been duplicated and it's Californian technology.
Intel 18A has a different strategy all together. I mean, it's not like they're saying much about it, but from the little we know, it's very different.
No the beam source we know how to create; it's just that the way to do so requires a factory sized machine that better be feeding 10~30 steppers or a pair for uptime/maintenance windows and then probably more like 50 steppers.
Free electron lasers are not hard at "mere EUV", and the accelerators to feed aren't either, it's just that an efficient setup requires recycling the beam which means bending it back which means a large (factory hall sized) accelerator at the beam energies needed for good EUV light.
Basically most of the technology was US govt funded at some point. Now the Dutch can just leech off of that. I guess that’s what they’re good at considering their long history with VOC and “mercantilism”.
I was just reading about this in a tech magazine, couple of startups also trying other techniques of interest, one of them using helium atoms for the lithography. This uses pretty close to x-ray wavelengths of light. Amazing feats of engineering taking place in this crazy world of ours.
Sibling has a good comment about silicon. Another reason you're printing something like 100B transistors. So that's at least 100B raster steps. It would just take too long.
Gotta start somewhere. Plus foxconn only abandoned 1 DC build out during Covid, a bit different from TSMC investing billions to build multiple, of which they already built 1.
IMO it's competition that makes better products in the end. If either party becomes a full monopolist the progress will stall (as already happened with Intel I guess)
In which sense? Intel is neither a full monopolist (AMD, ARM) nor has progress stalled (we're still getting a regular cycle of new generation CPUs which improve on the previous).
Intel had a monopoly on bleeding edge cpu for a long time until they got stuck on 14nm for years, allowing TSMC to catch up and providing AMD the means to compete.
In a the best of timelines, intel, tsmc and samsung (and glofo maybe?) Would all have competitive fab nodes.
It was a pretty bad time - ARM wasn't really encroaching on any of their markets in the way they are now & AMD was struggling. Intel basically just left the engineering to rot & the business people to try and gouge consumers with things like the Intel Upgrade Service, low power CPUs that were essentially just normal cores but less rather than optimized, inflated prices etc.
It also allowed AMD to completely blindside them with Zen - so we're in a situation where AMD now dominates consumer desktop and arguably a lot of enterprise/DC instead of it balancing towards a more competitive centre.
Eh, xeon 2690 was released 2012q1 at 8 cores, 2690v4 was 2016q1 at 14 cores. Same suggested customer price. Clock speed dropped a bit, but IPC was way up (AES-NI had a big throughput increase among other things). IIRC, desktop core counts were pretty stagnant, but a Haswell/Broadwell desktop chip at the same corecount and similar clocks was much faster than a Ivy Bridge/Sandy Bridge.
During 2012-2016 AMD was firmly in the Bulldozer disaster.
In 2017, Zen1 came out, and Intel's tick/tock mostly stopped. Zen1 was better than Bulldozer, but not better than Intel. Zen2 in 2019 was the big moment, and Intel hadn't done much. I recall an interview where the AMD person (might have been Jim Keller) said they had been targetting Zen2 to be competitive with what they thought Intel would have when it came out. Since Intel was in their pipeline bubble, Zen2 was a big win.
Intel had something very close to a monopoly for a long time. ARM wasn't taken seriously for desktop or server applications, and AMD was unable to compete with Intel's top-end CPUs until Zen (and with it Ryzen and Epyc) came around.
Intel Panther Lake is a serious improvement over the previous Intel CPUs, both in energy efficiency and in GPU performance. The good energy efficiency is a good sign for Intel's 18A CMOS fabrication process.
Nevertheless, it is worrying that the computers with Panther Lake CPUs have become much more expensive in comparison with their equivalents of last year with Arrow Lake, and this before adding the hugely increased costs for DRAM and SSDs.
It is also confusing for buyers that the 300-series Panther Lake part numbers that appear to be replacements for the previous 200-series Arrow Lake in most cases correspond to lower technical specifications, i.e. a much smaller GPU and lower clock frequencies.
The Panther Lake models that have the new big and very good GPU are extremely expensive and also hard to find, despite the fact that the GPU continues to be produced by TSMC. Perhaps TSMC has raised its price and prioritizes other customers, so Intel cannot get enough big GPUs, but even the Panther Lake models with the very small GPU (much smaller than Intel has used in many years for its mobile CPUs) have become far too expensive.
Because Panther Lake introduced a few important new features that could have been interesting for experiments in software development (e.g. FRED), I intended to buy some mini-PC with Panther Lake, but I gave up because even one that would cost double in comparison with the mini-PC with Arrow Lake H from last year that I have (i.e. around $1000 as barebone, instead of around $500) would be a downgrade in specs, instead of being an upgrade.
I believe Intel has a winner with the upcoming Crescent Island GPUs with up to 480GB of LPDDR5 memory. A lot of SMEs will need local AI with large open weights models (i.e. >1.5TB of VRAM), but they don't need a watercooled 8x Nvidia GB300 box. Besides, it's difficult to deploy and very expensive.
Crescent island is air cooled and likely fast enough for a group of users running something like GLM 5.2 or Kimi K3.
I mean, I sure as heck can't buy a nVidia one, at least not the founder editions which are the one that fit in my case...
I thought Nvidia after they invested in Intel killed Intel’s GPU efforts as it could be a threat to Nvidia’s GPUs
My general take is all of The GPU design efforts are going toward Enterprise / datacenter graphics cards. They are just skipping a generation or two of consumer.
There will be no future Xe GPUs.
Druid, the 4th generation, has been shelved entirely, and all consumer DGPU products, possibly _all_ DGPU products, have been killed for Celestial.
Its likely the only products shipping with Celestial will be IGPUs for the next generation and a half, until Serpent Lake comes out with RTX graphics tiles from Nvidia, and then never another Xe product ever again.
Didn’t somebody claim Battlemage was canceled? But they shipped it?
To Nvidia the biggest threat is a capex cut from big tech not another company in the same domain, maybe some Chinese companies but people aren't going to be moving off of Nvidia for large deployments in the short term even if some TPU/Tranium etc is used it's not close to Nvidia.
I have no explanation other than people, researchers are comfortable with it and don't want to switch.
Nvidia's software support remains "best in class" so they've become the default for a lot of this stuff. To overcome that I think you'd have to get a lot of hobbyists and by extension interested young people using your hardware instead of Nvidia. So you'd have to have good software support, decent hardware, and low hardware cost. People started with Nvidia because most machines had an Nvidia card in them, and Nvidia's software support was carried-over momentum from how they've supported their GPUs with PC gaming. I've tried AMD, for example, and the compatibility matrices and bugs and just general lacking software support for ML on their hardware makes it a total non-starter. It may be a lot better with current-generation stuff but it's going to take a lot for me to trust AMD for ML.
It's sad that this is true and yet their software is filled with bugs...
True, Given I am young, hobbyist, and have written a lot of AI libraries for myself and small companies for local deployments, and can't afford any serious Nvidia hardware I will switch to whoever can get me a system with enough VRAM for a decent price under 1000$...
My best GPU is 5070Ti on which the best model I can run is Laguna S 2.1 118B with NVFP4 and offloading most experts on CPU with an expert router layer and 128k context.
I could kill for a 1000$ AMD or Intel card with 32 or 48GB of VRAM even GDDR6 around a 1000$ range but it feels like no one even cares.
With ~4 48GB cards I could seriously locally deploy most open models models with some SSD offloading and good 4bit quantizations with ok context sizes. But at current prices.. sigh... I tried begging Intel to maybe it's cards available in my region but alas no such luck especially not at reasonable prices.
Honestly getting an AI subscription feels cheap atm and working hard at solving and home labing stuff is insane.
Even in research no one cares about anything but Nvidia, because at this point they are expensive but they seem to care, I just don't see others caring though I have worked with Intel gpu and compute teams on implementing this stuff and they are very enthusiastic but well the companies themselves aren't solving anything for me as an individual, but they are begging for me to add support for their stack in software I maintain. The Irony is unreal, why should I even add support when no one can even use it.
Only company that had decent stuff was surprisingly Apple but their stuff is now too expensive as well. .sigh.
This should have been obvious to me before but for whatever reason reading your comment made me realize something.
The hardware that nvidia is selling can be used against them because it can generate the code that is lacking for hardware from other companies.
So there’s a world out there where Intel or AMD just generate the needed software with compute from rented Nvidia hardware or the end consumer does the same, akin to opensource volunteers spending their free time reverse engineering undocumented hardware for the last few decades.
That’s if the claim that other companies produce acceptable hardware that is mostly just kneecapped by poor software.
If that’s the case there’s a distinct chance that we’ll see feature parity between the different vendors soon enough.
Unless the models that run on nvidia hardware are inadequate for this task, but that sort of raises a catch-22 — if the models aren’t good enough to generate drivers and CUDA type software what are they good enough for?
> To Nvidia the biggest threat is a capex cut from big tech
Not the biggest. Maybe top 3.
The bigger issue right now is data center expansion speed. You have SpaceX turning up random generators and doing whatever to get it over the line but that's the exception. There are likely stockpiles of GPUs and racks of AI clusters waiting to go online. It's only going to get worse.
> killed Intel’s GPU efforts
It doesn't matter anymore. Nvidia's core product isn't "GPUs" anymore. It's a whole AI rack. AMD also announced their own of the same. CPU, GPU, network, etc is all bundled.
A 4% stake doesn't even make Nvidia one of the top 3 holders of Intel stock... They'd be crucified for changing strategy for a 4% stake.
Also Nvidia arguably needs Intel more than vice versa, Intel now has SOTA fabs on the newest EUV technology, TSMC is at max capacity and they need to fight Apple and others for allocation...
> TSMC is at max capacity and they need to fight Apple and others for allocation...
Nvidia has (or almost) eclipsed Apple to be the largest customer and definitely has reserved capacity.
> Intel now has SOTA fabs on the newest EUV technology
That's still early. Hard to say Intel can consistently actually produce something good in large volumes yet.
Amazing looking machine! $380 million a pop. Must surely be one of the most sophisticated machines humanity has ever built? A great achievement of European science and technology (ducks).
> A great achievement of European science and technology
It's an amazing piece of technology but this is... wildly wrong. ASML is a multi-national company that licenses IP largely from USA and Japan, but also Taiwan and Germany. The actual EUV light source is developed and produced in California by Cymer, which ASML acquired in 2013. But ASML was only permitted to acquire the company under a strict technology sharing and export control agreement with the US government. Additionally, a huge portion of the photolithography research is directly developed (and owned) by US companies and research organizations such as IBM, Albany NanoTech, and SEMATECH.
There is a reason why ASML's next-generation research photolithography machine is currently being installed and developed in upstate New York, and not somewhere in the Netherlands. The same reason that Cymer is still in San Diego instead of being relocated to Europe. Because it's American technology.
But Europe certainly plays a huge role in it.
https://www.governor.ny.gov/news/governor-hochul-announces-n...
https://research.ibm.com/blog/euv-center-albany-nstc
https://www.eetimes.com/asml-sematech-team-on-manufacturing-...
https://www.eetimes.com/asml-to-build-400-million-us-researc...
> There is a reason why ASML's next-generation research photolithography machine is currently being installed and developed in upstate New York, and not somewhere in the Netherlands.
This is a wildly ignorant comment. The R&D lab in NY State if anything is an effort to copy IMEC, the world's leading research organization when it comes to semiconductor R&D. They essentially set and define the roadmap for node advancements and then the industry follows by actually implementing their innovations at scale in their fabs. I guess they are technically not in the Netherlands, being located about 50 miles south of the Dutch border in Belgium. And they have in fact have an EXE system of their own, and have had access to ASML's high-na tools since before they were even commercially available
Agreed. Berkeley / CXRO is the American beyond-NA research center not Albany NanoTech which doesn’t have a charter AFAIK. They are just beginning to install the EXE:5200 in Albany now..
They do have the xLight program which doesn’t have a true European equivalent but still is a strategic partner of ASML who have experimented with FEL in the past.
> There is a reason why ASML's next-generation research photolithography machine is currently being installed and developed in upstate New York, and not somewhere in the Netherlands. The same reason that Cymer is still in San Diego instead of being relocated to Europe.
Why is the Netherlands government even allowing this?
The supply chain both in parts for their machines, but also in terms of research and technology for new machines is very deep and very international. If the US states that a technology can only be used by ASML if they keep R&D in the US, then there is no easy alternative for ASML.
That is true not only for ASML, but for the semiconductor industry in general. The number of smaller, specialized companies for certain tools, chemicals and software is staggering. And they sit all over the US, Europe, Japan, Taiwan and Korea.
One of the best examples of this is how TOTO the toilet company makes the electrostatic ceramic wafer chucks that hold wafers during processing.
https://www.youtube.com/watch?v=CIB49e_r1OI
The entire multinational effort is in US-Europe-E.Asian countries which are a coordinated allied bloc. The Netherlands allows it because they are entirely aligned.
Strong mutual defense agreements cover four out of the five crucial countries here (NATO, Anpo, MDT US/ROK) and the last (Taiwan) has arms sales assurances.
It’s not like these are strongly opposed nations or anything. Part of the same fleet in some sense.
2023 called...
Russia invaded in 2022, Trump was elected in late 2024 and took office in 2025. The out of context quote that kicked all this off was in 2017 or 2018. Somehow you picked the one year out of the last 10 or so that didn't matter for any of this.
The really wild thing about this whole narrative is how unlikely this entire sequence of events was. Let's review:
1) Trump gives some rambling answer to an Estonian journalist in 2017 (or 2018) in which he says about 4 different things, all of which conflict with each other.
2) That journalist clips the answer to make it sound like Trump wants to leave NATO.
3) European leaders lose their mind and for the first time in 35 years actually up their military spending. Something that every US president since Bush Sr has made a major foreign policy goal that ever single administration failed at.
4) So completely by mistake, a journalist nobody has ever heard of, handed Trump the biggest foreign policy win for any US president since Bush Sr.
5) Russia invades Ukraine
6) Sweden and Finland join NATO
7) Trump wins the 2024 election and cuts off some of the Ukrainian aid (but not intelligence sharing).
8) The removal of US limitations on Ukraine allows them to strike Russian oil infrastructure.
9) Russia can no longer afford to fund their war in the way they had been.
10) Ukraine is now actually winning.
So somehow, Trump has gotten a huge foreign policy win and helped turn around the war in Ukraine completely by accident. Its like watching a cow talk.
Not by accident. By the immense sacrifice of the Ukrainian people and it's armed forces coupled with US goals just happening to align on Iran. I somehow doubt the Ukrainians mind very much that shahed production is now pivoted towards targets closer to Iran than to Ukraine.
I mean Trump didn't intend to help. But somehow NATO is in a better state and Ukraine is winning. Its only in the largely imaginary political narrative where NATO is somehow weaker. What is weaker is the credibility of several European leaders. And nobody here thinks that this was some master plan or 10D chess by Trump.
Can't undo a 2013 tech sharing agreement because you got a call in 2023...
Technically NATO is much harder to undo, and yet there we go.
It's much easier to discontinue military (and other) cooperation than it is to unshare knowledge.
Because ASML acquired the US company that created this tech and the agreement for allowing Netherlands to do so was that the tech and know how remain in the US.
The EUV lithography process and other pieces that comprise what ASML sells was developed with US government funding, and it remains encumbered by that. The US government thus has veto power over the tech, and thus over ASML.
Because the EUV light source is an invention originated in the USA and therefore it is under US export controls. No light source, no machine.
It wasn't just invented in the US, it was co-invented by the US government itself.
https://www.osti.gov/biblio/1413999
if they dont allow it ASML will move away and they have nothing.
The US is currently aggravating practically every other country on the planet and writing like this only pisses everyone else involved all the more.
I'm not sure if it's wise to minimise the contributions of everyone else in the current climate tbh.
As opposed to the parent post that left the US out entirely?
> The actual EUV light source is developed and produced in California by Cymer, which ASML acquired in 2013.
Cymer was one of multiple vendors (Gigaphoton, XTREME) providing this sort of light source, and ASML acquired it because it couldn't achieve what they wanted (they needed 100+W and Cymer could only achieve 15W). ASML needed to control and fund the company, and add engineering talent, to achieve what they actually needed.
Oh and ASML bought Cymer because Gigaphoton (another company at basically the same stage of development as Cymer), being a Japanese company, objected to foreign control. Though if ASML knew the protectionist, insular direction the US was going to go they probably would rethink their past decisions.
So this retconning that Cymer is really the secret magic of ASML and the Netherlands is just stealing US ingenuity is hilarious stuff.
And for that matter, it is amazing how so often we hear about "American" tech that is actually overwhelmingly combining acquisitions and manpower from abroad. Like is an iPhone an "American" creation? LOL, no.
> the secret magic of ASML and the Netherlands is just stealing US ingenuity is hilarious stuff.
I never said anything about ASML stealing technology or ingenuity. The point is that EUV photolithography is an international effort. The machines don't exist at all without about 6 different countries contributing technology and expertise.
ASML has certainly proven itself to be a very capable integrator of many international technologies and processes.
Most of those technologies being European, including the most difficult to make components in the system.
Why isn't there an American company doing this? Why isn't an American company making the lasers, mirrors, metrology, vacuum handling tech, etc?
I’d say Japan is a more important source than sources in Europe.
they are licensed to European company yes
reason why US didn't produce most of these component because US gov didn't continue investment so privately owned enterprise take over from there namely ASML
that also reason why US can restrict EUV sales to china
isnt DUV and EUV technologies discovered and developed by US national labs (LLNL, LBNL, Sandia) in cooperation with EUV Consortsium funded by US Chipmakers (Intel, AMD, Motorola) and licenced to Cymer?
EUV was "invented" (demonstrated) in Japan first, though those US labs did a lot of work on the overall lithography system and reflective optics to put it into practice. The US DOE along with the manufacturers made a consortium (EUV LLC consortium) to commercialize/scale that Japanese technology because they could obviously see the promise, and Cymer is a member, as was ASML already via SVG.
Simultaneously Japan created its own initiative (EUVA), and Gigaphoton didn't rely upon the US initiative at all. And again in an alternate universe -- one where Japan isn't protectionist and Cymer wasn't out of financial runway and desperate -- we'd be talking about how ASML is really Japanese tech, etc.
This video:
https://youtu.be/MiUHjLxm3V0
Has a lot of the history
Bingo!
Yes the OP is completely biased.
The most absurdly biased and inaccurate take on ASML I've ever seen, congrats.
If the machines are so easy to build and integrate, why isn't it happening in America?
Can't build shit in America. Not even data centers.
ASML is the only real tech company in the EU. Just let them have this one :)
EUV was developed in a joint venture with the Department of Energy and ASML, Intel, etc called EUV Corp. It’s why the US can export control it.
Also Japan has a monopoly on essential EUV materials like photoresists and mask blanks and pellicles. Japan is also the only country in the world that can produce a number of highly specialized ancillary machinery like the actinic light systems used to inspect EUV photomasks. ASML also acquired its e-beam inspection tech from a Taiwanese company. There's also no way ASML could run anything at production if it wasn't for TSMC and Samsung.
Not to mention the critical role of the US in its development. The core EUV light source was invented in San Diego!
LOL, no Mention of IMEC(Interuniversity Microelectronics Centre) in the EUV development process and the only reason that the US can restrict the ASML EUV equipment is the EUV Light source and mirror optics for that is partially US produced and the IP there falls under US IP/Export restrictions!
From ASML:
"ASML’s extreme ultraviolet (EUV) light source was developed and is manufactured at its facility in San Diego, California. Acquired from the US firm Cymer in 2013, the R&D team here developed the laser-produced plasma (LPP) system that fires lasers at tin droplets 50,000 times per second to generate EUV light."
Yes, and that changes what?
> Amazing looking machine!
As with most of this type of technology, most of the stuff you see is basically a giant refrigerator and giant vacuum cleaner. The actual meat is much smaller and hidden from view.
Depends on definition of machine. Do ISS, LHC or ITER count?
ASMLs machines can easily be "one of the most sophisticated" without stepping on LHC or ITERs toes. I'm not sure ISS qualifies anymore. I would have went with JWST.
Can’t buy and ship those.
I love how ASML machines are shaped to the contours of a jet fuselage for this reason!
no.
neither of the observations gp made is dependent upon the definition of machine, nor is it dependent on the inclusion of the examples you brought up in that definition.
The lasers are made in San Diego, the lenses made in Germany.
> A great achievement of European science and technology (ducks).
It's a fantastic machine but the bulk of the achievement is clearly a joint European-American effort.
Then Apple M series is also a joint European-American effort, no? Since ARM is European?
Weirdly enough it never seems to work that way in discourse.
Should add China to the list of pretty much all of them as well. It’s a silly take.
In that case the internet is mostly due to the Brits (invention of packet switching at the NPL by Donald Davies). Cheers!
Not even that, the company happens to be Dutch but it's employees are from everywhere. A company doesn't make anything, the employees do, especially in a sector like this.
The companies job is to hire well and create strong teams that work together.
That's what makes this hard to interpret. The whole strategy of intel is that they have their own version of this machine, part of their "18A" design. The only reason they need the ASML machine is because of the "optics" (between quotes because at the frequencies they work at no optics work - only (bad) mirrors do. However you need make the mirrors accurate enough - we're talking far more accurate than the chips, as in sub-nanometer max deviation, they can function as optics with high losses, but low enough that exposure works). No American company can achieve even 10 times lower accuracy at the moment.
So is this giving up on their own version? That would mean their stock price is absolutely not justified at $110, and it should be more like $60-70 if not lower. Or is this "we have technical reasons to use this. We're still going to do 18A and it's still going to beat 2nm", which would mean their stock price is easily justified at $200 and it's high time to buy.
Intel does not have their own version of this machine. This ASML machine is a component in a very large system of machines from many manufacturers that runs wafers automatically from machine to machine until a month later a chip comes out the other end. That system is “18A”. Intel designs, builds and operates that system. Intel does not make the machines, they buy them and figure out how to configure and operate them as a part of the system.
As I understand it, the machine also needs lots of 'normal parts' and can be designed differently to fit the factory and the automation. Like if you need a pipe, instead of buying it from ASML, you can buy it from a local hardware vendor. And parts of the ASML also break, and those parts would be directly shipped from the vendor that makes the part. ASML manages the integration between these parts and each part can have multiple vendors that makes them according to the specs.
Only the optics is an ... interesting way of phrasing it.
There was a great Economist article a while ago that talks about this. ASMLs moat isn't just the know-how to build the machines themselves. It's a network of suppliers all across Europe, specialized businesses doing one little part better than anyone else in the world.
This is the type of thing that Europe is still leading in, and it'll be incredibly hard to reproduce elsewhere.
You don't just need to figure out how to build an High-NA EUV machine. You need to figure out how to build or procure all of the parts that go into it. Many of these suppliers have exclusive contracts with ASML.
Here's the article, de-paywalled: https://archive.ph/jP1HX
Yeah. ASML doesn't make the mirrors, but they have exclusive sales rights, exactly as you say. And yes, positioning them, for example, both in the fixed sense (mounting the mirrors in the machine) and in the dynamic sense (rapidly positioning wafers to nanometer-or-better accuracy below the beam, I forgot how quick exactly but it's 10s to 100s of times per second to maintain production rates) is very much not easy either. But it's "been done" (by teams of phds). Producing the mirrors has not been duplicated.
Of course, producing the EUV beam in the first place has also not been duplicated and it's Californian technology.
Intel 18A has a different strategy all together. I mean, it's not like they're saying much about it, but from the little we know, it's very different.
No the beam source we know how to create; it's just that the way to do so requires a factory sized machine that better be feeding 10~30 steppers or a pair for uptime/maintenance windows and then probably more like 50 steppers.
Free electron lasers are not hard at "mere EUV", and the accelerators to feed aren't either, it's just that an efficient setup requires recycling the beam which means bending it back which means a large (factory hall sized) accelerator at the beam energies needed for good EUV light.
Uhh, you do not need an accelerator to generate EUV. It's literally just molten tin hit with a laser.
Weiss makes the mirrors.
Basically most of the technology was US govt funded at some point. Now the Dutch can just leech off of that. I guess that’s what they’re good at considering their long history with VOC and “mercantilism”.
I was just reading about this in a tech magazine, couple of startups also trying other techniques of interest, one of them using helium atoms for the lithography. This uses pretty close to x-ray wavelengths of light. Amazing feats of engineering taking place in this crazy world of ours.
most importantly, is Ian Cutress' blog basically the new Anandtech? I miss that site...
Is it so hard to explain what the term "High-NA" is supposed to mean at the start of your article?
> ASML moved to camera-style anamorphic optics that magnify the mask by different amounts along the two axes
Does this mean they do high-resolution designs in the center of the lens and low-resolution designs in the periphery?
How come we don’t make chips with really small 3d printers?
Sibling has a good comment about silicon. Another reason you're printing something like 100B transistors. So that's at least 100B raster steps. It would just take too long.
The closest thing there is is e-beam lithography. It isn't really practical.
Isnt that smaller? I was under the impression e-beam is one of the (possible) next generation chip technologies.
Getting it to scale is the issue
AFAIK you can't really print crystalline silicon so they have to start with a wafer and dope/etch into it to form transistors.
Im rooting for Intel. I hate the current paradigm where all the best chips are made in one factory in Asia.
Tsmc are opening 10 factories in USA at cost of $265bn
https://www.techrepublic.com/article/news-tsmc-us-investment...
Intel staying strong is still vital for competitiveness though
Until they are built, it's all talk.
https://www.cnbc.com/2021/04/21/foxconn-mostly-abandons-10-b...
They are built, at least one in Phoenix:
> As of 2025, the first fab has been completed and is producing four-nanometer (nm) chips.
One is built, OP said 10.
Gotta start somewhere. Plus foxconn only abandoned 1 DC build out during Covid, a bit different from TSMC investing billions to build multiple, of which they already built 1.
TSMC also AFAIK is forbidden from setting up bleeding-edge fabs outside of Taiwan?
Intel latest is the bleeding edge and more advanced than TSMCs.
Four more 2nm (or below) fabs in Arizona. That's bleeding-edge or very close to it.
By end of decade, no? Seems like domestic fab will always be two steps behind. I assume by design.
They're building 1.4nm in Taiwan
Will they remain bleeding edge by the time they are operational?
IMO it's competition that makes better products in the end. If either party becomes a full monopolist the progress will stall (as already happened with Intel I guess)
> as already happened with Intel I guess
In which sense? Intel is neither a full monopolist (AMD, ARM) nor has progress stalled (we're still getting a regular cycle of new generation CPUs which improve on the previous).
Intel had a monopoly on bleeding edge cpu for a long time until they got stuck on 14nm for years, allowing TSMC to catch up and providing AMD the means to compete.
In a the best of timelines, intel, tsmc and samsung (and glofo maybe?) Would all have competitive fab nodes.
Intel didn't do much to improve their CPUs while AMD was out of the competition with the Bulldozer disaster.
It was a pretty bad time - ARM wasn't really encroaching on any of their markets in the way they are now & AMD was struggling. Intel basically just left the engineering to rot & the business people to try and gouge consumers with things like the Intel Upgrade Service, low power CPUs that were essentially just normal cores but less rather than optimized, inflated prices etc.
It also allowed AMD to completely blindside them with Zen - so we're in a situation where AMD now dominates consumer desktop and arguably a lot of enterprise/DC instead of it balancing towards a more competitive centre.
Eh, xeon 2690 was released 2012q1 at 8 cores, 2690v4 was 2016q1 at 14 cores. Same suggested customer price. Clock speed dropped a bit, but IPC was way up (AES-NI had a big throughput increase among other things). IIRC, desktop core counts were pretty stagnant, but a Haswell/Broadwell desktop chip at the same corecount and similar clocks was much faster than a Ivy Bridge/Sandy Bridge.
During 2012-2016 AMD was firmly in the Bulldozer disaster.
In 2017, Zen1 came out, and Intel's tick/tock mostly stopped. Zen1 was better than Bulldozer, but not better than Intel. Zen2 in 2019 was the big moment, and Intel hadn't done much. I recall an interview where the AMD person (might have been Jim Keller) said they had been targetting Zen2 to be competitive with what they thought Intel would have when it came out. Since Intel was in their pipeline bubble, Zen2 was a big win.
Intel had something very close to a monopoly for a long time. ARM wasn't taken seriously for desktop or server applications, and AMD was unable to compete with Intel's top-end CPUs until Zen (and with it Ryzen and Epyc) came around.
yeah - I wouldn't bet against Intel.
they might have floundered in the past.
panther lakes I hear are pretty competitive with some of the apple silicone.
Intel Panther Lake is a serious improvement over the previous Intel CPUs, both in energy efficiency and in GPU performance. The good energy efficiency is a good sign for Intel's 18A CMOS fabrication process.
Nevertheless, it is worrying that the computers with Panther Lake CPUs have become much more expensive in comparison with their equivalents of last year with Arrow Lake, and this before adding the hugely increased costs for DRAM and SSDs.
It is also confusing for buyers that the 300-series Panther Lake part numbers that appear to be replacements for the previous 200-series Arrow Lake in most cases correspond to lower technical specifications, i.e. a much smaller GPU and lower clock frequencies.
The Panther Lake models that have the new big and very good GPU are extremely expensive and also hard to find, despite the fact that the GPU continues to be produced by TSMC. Perhaps TSMC has raised its price and prioritizes other customers, so Intel cannot get enough big GPUs, but even the Panther Lake models with the very small GPU (much smaller than Intel has used in many years for its mobile CPUs) have become far too expensive.
Because Panther Lake introduced a few important new features that could have been interesting for experiments in software development (e.g. FRED), I intended to buy some mini-PC with Panther Lake, but I gave up because even one that would cost double in comparison with the mini-PC with Arrow Lake H from last year that I have (i.e. around $1000 as barebone, instead of around $500) would be a downgrade in specs, instead of being an upgrade.
Intel is doing well. They have 18A yields up to 85% at Fab 52. That's huge for the foundry business: the best foundries are not all in Asia now.
Rumor is they're moving Nova Lake production mostly in-house due to 18A success.
I wish that, but for high performance RISC-V micro-achitectures.
https://www.eetimes.com/barcelona-bsc-celebrates-tape-out-of...