Hacker News Reader

Comment by darkamaul | Hacker News Reader

I don’t really get why ASML is putting money into Mistral AI. ASML is specialized in lithography machines. Mistral, on the other hand, is yet another LLM startup.

What’s the actual synergy here? The closest angle I can imagine is that AI workloads drive demand for more chips, but I believe ASML is already selling everything it can make.

View on HN

Past Mistral investors: JC Decaux (urban advertizing), CMA CGM CEO (maritime logistics), Iliad CEO (Internet service provider), Salesforce (client relation management), Samsung (electronics), Cisco (network hardware), NVIDIA (chips designer)[0]. I agree ASML is a surprising choice, but I guess investments are not necessarily directly connected to the company purpose.

BTW, I generated that list by asking my default search engine, which is Mistral Le Chat: indeed, using Cerebras chips, the responses are so fast that it became competitive with asking Google Search. A lot of comments claim it is worse, but in my experience it is the fastest, and for all but very advanced mathematical questions, it has similar quality to its best competitors. Even LMArena’s Elo indicates it wins 46% of the time against ChatGPT.

[0]: https://mistral.ai/fr/news/mistral-ai-raises-1-7-b-to-accele...

The list seems to be missing a couple of other notable investors: Eric Schmidt (former Google CEO), Andreessen Horowitz, Lightspeed Venture Partners, General Catalyst and Microsoft (only $16M).

I didn't realize Mistral was A16z's pony in the race unless the splashed across the board(?)

Quite complex algorithms are used to compensate and tune for pattern clarity and focus in high end semiconductor production.

Its a field that has used neutral networks before. (As people pushed down the size pre-EUV, apparently alot of wierd techniques were layered to produce features at or smaller than the wavelength)

But mistral just makes llms. There is no reason to believe experts in llm would be at all competent at quantom scale physics simulation and prediction.

It feels more logical to invest on the existing researchers and companies in the nanotechnology design field to adapt newer AI techniques.

Multi patterning to get effective smaller wavelengths has been around a while. It’s cheaper to reuse machines you already own, but slows down production.

OpenAI does more than LLMs, they have bio ML research etc. and Google has AlphaFold. It would not surprise me if Mistral had an ML team on physics related to work that ASML could use.

I suppose, but i don't feel like that makes mistral special enough to excuse this amount of funding. They would need clever researchers with resources to do research. The kind of AI we're talking about would likely not benefit from data-center scale training either. So why the 1.7B euro? That amount of money could fund multiple small dedicated research labs for exactly the domain ASML is interested in.

I don't think it adds up if this is truly for multi-patterning or pattern exposure correction technology.

As others mention it could be for entering and grabbing some value from down-stream technologies (actual investment expecting return of some sort) but it's odd how they skip over like 200 steps between their industry and the industry they invest in. Its like iron ore mine investing in precision screws. Its down the value added chain but such a massive leap that it makes me scratch my head.

The most strategic move ASML can make is change its licensing structure such that Apple will have to pay 30% of their revenue for using their Fab platform.

Probably a bad advise. ASML can’t make or export products without American licenses:

https://www.reuters.com/technology/new-us-rule-foreign-chip-...

This.

Absolutely majority of IP in this field belongs to intel, IBM, KLA and Lam research. Everyone else is a licensee. This is one of the reasons us and allies are desperate to keep bailing out intel, or get it acquired by another american company.

The linked article clearly states the export ban is the result of the US convincing other countries, such as ASML's host country the Netherlands, to join their export controls.

I'm not sure why European countries signed this deal, given that the US still started trade wars a few months later. Maybe they had more faith in the American electorate than they should've had.

There is no need to convince. It's probably diplomatic language/misunderstanding by the journalist.

U.S export controls have a foreign direct product rule. If your product directly depends on US IP/tech then US export controls apply to you too. ASML's EUV machines depend on a lot of U.S ip and tech. These range from patents from the EUV consortium days, all the way to today where a lot of the components are designed or patented by the US (intel, ibm, applied materials, lam, KLA, [1]). Thus, export controls apply whether netherlands likes it or not.

In EUV litho, the strategic I.P - and what IP is a strategic chokepoint is subjective, so IMHO - is 50% EU, 40% U.S and 10% Japan. More than enough for FDPR.

See ECCN 3B090 (2022) for lots of extra restrictions placed beyond this, specifically targeting semicon manufacturing exports to PRC, such as the presumption of denial clause.

[1] and cymer. Altho they were acquired by asml their IP remains san diego based aiui.

Do Intel, IBM etc. and their shareholders not want to extract more money from Apple (through ASML)?

I wonder what it would cost Apple to recreate ASML.

Ask China, they've been trying for a decade.

Give them enough time and they will. EUV will hit limits anyway in a decade.

For china it's DUV+packaging for now, NIL/DSA mid-term, and MoS₂/2D chips long term. But wafer scale, defect free 2D logic is 20–30 yrs out, so no EUV shortcut anytime soon

Thanks to a combination of espionage and homegrown Chinese technology advancements, they went from "decades behind" to "years behind" quite rapidly on several critical parts of the chip manufacturing process.

China isn't quite there yet, but they will catch up. The question then becomes whether China can surpass the west or if they're stuck in lock-step behind us.

Yeah but China wasn't (and won't be) given the tech. The fastest path for Apple would be to get POTUS a gold iPad in exchange for the US removing exclusivity terms for the EUV tech they gave ASML.

And SMIC is a decade or less behind without any of that.

It's very difficult. It took ASML 20 years, and Apple has none of the core competences to make this happen, like optical lithography, EUV optics, plasma physics, vacuum, laser, sources...and then they would have to catch-up to the other tech. For example, today's top end ASML stages accelerate with >10g while still having nm position accuracy.

I can't find the article, but there was an estimate to catch up to ASML would cost between $100 - $200 Billion. You'd also be competing with ASML for a very small talent pool the whole time. See the $100 million payouts for AI researchers, for instance.

This is something the Chinese government is actively trying to do, it's not theory. I'm interested to see what the results are, because they are absolutely not competing for the same talent pool as ASML, they're attempting to create an entirely Chinese supply chain and talent pool.

I don't know enough about chips to say whether any of these numbers make sense.

China is decades behind the West in EUV technology. The attempt to create an independent supply chain is also a forced choice since all the EUV supply chain and knowledge pool are heavily protected by the West and are so complex and big that China cannot sidestep it even with a lot of resources.

Those numbers are realistic. EUV is the most complex machine ever built by humans

There's no way to catch up really - if they keep innovating like they are it's not possible to bridge that gap.

Also, it's probably a patent minefield.

Having directly worked on fab process engineering typically if its patented it’s going to get copied.

I think even then patents can be prohibitive or else https://news.ycombinator.com/item?id=45181828 does not mean anything.

Years of time and an organizational distraction more importantly than money IMO. Then the same for TSMC if the goal was autonomy.

While I associate Mistral with LLMs, the electric design automation software used for planning and designing chips already uses machine learning/reinforcement learning for some approaches. AI could play an even greater role in chip design in the future.

Llms are fundamentally different algorithms and problem space to IC design and production. Why would mistral be helpful?

I dont see how even the algorithms involved translate well. IC design is closer to a physics simulator connected to a heuristic optimizer. Mabie some ideas from alfageometry or alfafold could be applied, but thats not the kind of research mistral is doing.

And there are big players with existing expertise in the IC design space. Why not just fund them to do more research?

Mistral needs a looooooot of GPUs. GPUs are made by Nvidia. Nvidia asks TSMC to make more. TSMC needs new lines to produce. TSMC acquires more machines to make new lines. TSMC buys from the only monopoly that has those machines. ASML. Now ASML happy, each machine costs 100s of millions, ASML makes back money.

Maybe something like https://www.cadence.com/en_US/home/ai/overview.html ?

Everybody in that space does it, or at least tries to.

Maybe it is about defending their business by going vertical? Others are too, right? Like OpenAI is partnering with Broadcom. Google and Amazon have their own chips. Nvidia will probably need to do more than build AI training chips as well.

AI is being used more and more in chip design

I don't believe there are any "LLM"-style AI being used for chip design yet (if ever). It is a different problem space and the current RL and ML practices are still state of the art.

LLMs too? (outside of educating new engineers)

transformer expertise has a tendency to transfer. these are curve fitting machines par exemplar and there are many curves in chip design

truly. if a company can’t find a way to reinvest money in its core business, it should return it to stockholders (or worse case, invest it in public markets) rather than trying to become a stock picker for a different industry.

it speaks to the likely regulatory overheads in returning money to investors that they choose this route

I also believe Trump has a big part of this. Exporting will be less profitable. So business will rather invest in EU than in an American AI company. This to increase local demand for their hardware.

Reply on news.ycombinator.com