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.
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.
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. 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.
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.
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.
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.
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.
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).
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.
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.
Does this mean they do high-resolution designs in the center of the lens and low-resolution designs in the periphery?
Crescent island is air cooled and likely fast enough for a group of users running something like GLM 5.2 or Kimi K3.
I have no explanation other than people, researchers are comfortable with it and don't want to switch.
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...
Why is the Netherlands government even allowing this?
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.
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.
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.
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
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.
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.
It's a fantastic machine but the bulk of the achievement is clearly a joint European-American effort.
Weirdly enough it never seems to work that way in discourse.
The companies job is to hire well and create strong teams that work together.
https://www.techrepublic.com/article/news-tsmc-us-investment...
Intel staying strong is still vital for competitiveness though
https://www.cnbc.com/2021/04/21/foxconn-mostly-abandons-10-b...
> As of 2025, the first fab has been completed and is producing four-nanometer (nm) chips.
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).
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.
In a the best of timelines, intel, tsmc and samsung (and glofo maybe?) Would all have competitive fab nodes.
they might have floundered in the past.
panther lakes I hear are pretty competitive with some of the apple silicone.