EUV
Corgi Lithography & Semiconductor Photonics ETF
Mentions (24Hr)
100.00% Today
Reddit Posts
ASML Q3 2026: record quarter, full-year guidance raised
Korean Chip Stocks Slide on China's DRAM Push
How today's highly valued AI stocks can turn into the next IBM like turn around
SK Hynix Rises Over 15% as Citi Says Chinese Memory Makers Won’t Impact Market in Short Term, Memory Tightness to Last Until at Least 2027
China’s CXMT Surges 466% in Asia’s Largest IPO of 2026
The ASML and Chip Stocks Sell-off is an Overreaction
ASML Q1 Results: Net Sales hit €8.8B, Net Profit at €2.8B, FY26 Outlook raised. With AI driving relentless lithography demand, is ASML the ultimate AI bottleneck?
Mentions
High NA-EUV makes for great pillow talk idk what you’re on abt
How is it that people assume that the most important CEO in America got hoodwinked, and what could he have leaked to China? People think what? That he has all the tech specs of the most advanced systems in his head and after he had sex his honey pot was like "Jensen you're so good in bed, I never came so hard, now tell me about Vera Rubin architecture in detail and how we can get our hands on some EUV machines? Also, please tell me about all the ways that the Pentagon is using Nvidia chips." And Jensen Huang was like "Yeah, I am good at sex. Sure, I'll tell you everything and give you total access to company and American security secrets because you're the only woman who I met with who wanted me even though I'm one of the wealthiest people in the world and a global hero. I'm also really low IQ, which is how I achieved all this to begin with." GTGOH
They're anything but helpless. This view is why the west is slowly losing to China in pretty much every field. Western automakers laughed off their Chinese counterparts 15 years ago, look where they are today. Same happened with western robotics firms. Sure they bought a few key players, but they have far surpassed western countries in robotics in the years since. Having physical access to a machine doesn't mean you can easily reverse engineer it. There's still a ton of software IP you can't get access to, which is really where the 20 years of R&D ASML spent are. Those machines need to be online and talk to ASML all the time, from ASML white-listed IP addresses. Despite what the US administration has said, ASML has repeatedly said all their EUV machines are accounted for, with known locations. So, even that claim that China has somehow smuggled a machine is questionable at best, for all the above reasons. You don't need to steal a machine to build a replica, assuming you have the supply chain to do it. That's not even a trade secret. ASML had been talking about these machines for literal decades before they got them to work. What China is actually doing is building a domestic supply-chain for EUV, everything from the precision optics (which Zeiss does for ASML, and Zeiss themselves rely on a deep supply chain in Germany for that), the nano-second pulsed lasers, to all the precision piezo actuators that need to handle the extreme accelerations and weights involved, while maintaining accuracy. China has already demonstrated a research-scale EUV light source, made with an entirely domestic supply chain. This is about where ASML was in 2012 or so. They have all the money they need to fast track development and close the gap in half the time or even less vs ASML.
Hahaha ... I like you. You know something. I chap and chappess that can actually reason and type more than two sentences. And we're pretty much on the same page. Unless you're a bot and you're blowing smoke up my ass. So easy to in this day and age to get into the psychosis sort of way of thinking, no? I would say, as I said before. Width is unknown. Good enough is good enough is good enough. Let's everyone sleep rather soundly if it's good enough. That's the interesting intellectual thing with all this shite. The deeper thought process, in the grand scheme of things. It's something I think about rather philosophically. I always think about this thought experiment of the paralegal fresh out of law school. Big bright lights beckoning. Pay of the student loads and all that. Cast law. Making mistakes. Seeing their name in bright lights eventually. Becoming a name in the practice. Eventually one of the names out of Prantagast Smith and X. Holding that thought experiment in my head is a good way to judge what's going on. Their possibility of getting to that point. I dunno. Personally, I think they're fucked. It's like a dipstick in the oil that you can check up on. Sad shit that like. For my sins I'm actually a book cover designer because I'm lazy fucking arty wank as well. Half work, half constantly learning and investigating shite. But nay fucker talks about ruben class at all. No one on here. Blinded by up-down bullshit versus the-light-and-the-way, polemic idiots emotive wank the lot of them. If I was a more cynical man though, I'd easily say. Every cunt is taking their eye off the ball. It's not exactly gonna replace the whole work force but their is a good argument in saying, it's probably smarter than most cunts fresh out collage right about now. Which is a sobering thought. Calls on instability moving forward. Sort of dig the face that China's EUV program is ahead of schedule. If you have an apple phone you're a chump because you can get that shite for €200. There is a strong reason behand that EUV shit. Scary and funny to my mind. But there again I'm a left leaning Brit and I adore my time more than I do money. Go figure. PS Fucking hate like bum-cancer-aids the fact that fucking no one bothers to even bother doing the minimum of research and just repeats shite they've heard. Lazy fucks. Crank the shaft and start up Three Island again. Now there's a clue for you to follow up on.
Been thinking about that although otoh Nvidia could follow more a path like Cisco. A lot more alternatives to Nvidia's chips now, Google making it's own TPU's, the big elephant in room with China making it's own EUV and GPU's just coming online, and then Samsung and the other's. Nvidia will still make good money but not printing like it was before, they can't charge those margins much longer. But doubt it would plummet super-far, just a question of the best price in proportion to a reduced but still healthy sales.
Great points here, been saying a lot of the same things. The circular financing and massive leverage was always going to be a disaster especially when bond yields and borrowing costs they went way up, like they started to do later in September. And then all the dumb hype and FOMO around the AI bubble, crazy promises the math would never support even with most over-optimistic assumptions and sales. And then how Amodei, Altman and Musk are so scummy in general with such shady IPO's. It's like they're not even trying to hide they're just looking for dumb retail bagholders for exit liquidity to dump their companies in massive pump and dump and rug-pulls. The SpaceX IPO was bad enough, but Dario and the Anthropic insiders basically set up the Anthropic IPO to openly fleece retail, getting cash out options on Day 1 and F-shares so retail doesn't even have much say in the company. At same time China and Europe (with Mistral) have [cheaper open weights models at](https://qz.com/ai-us-china-open-models-deepseek-qwen) same quality but [fraction of the cost](https://finance.yahoo.com/technology/ai/articles/businesses-experimenting-cheaper-chinese-ai-185931713.html) [and businesses already switching to](https://www.cnbc.com/2026/07/07/chinese-ai-models-costs-us-openai-anthropic.html). All while Treasury yields going up and debt gets so pricey CoreWeave and Oracle basically have junk bonds, and stupid valuations and PE ratio's all over. (Tesla still worth more than other car companies combined even though way down on the list of sales, and now BYD and soon Geely snatching away dominance in the EV business). So like you saying it was already a perfect storm before November. The harsh anti-AI backlash by the public was just iceing on that cake, but it gets played up because it gives Fox News and the usual idiots a convenient excuse and scapegoat to blame the Democrats for an AI bubble that was imploding anyway. And the hatred of AI data centers is uniquely bipartisan and all over the US, in red blue and purple states the same. [The harshest data center bans](https://www.toledoblade.com/local/politics/2026/11/04/data-centers-ohio-voters-reject-large/stories/20261104104) passed as ballot initiatives in red rural regions of Ohio on Tuesday, and even the few GOP governor candidates who did better read the room and went anti-data center themselves, especially Byron Donalds in Florida. (An easy call since it's so swampy, hit by sinkholes and massive storms the insurers wouldn't insure data centers there anyway). But those actual facts don't matter when Fox News now has the narrative to blame Democrats for a nasty economic recession that was coming anyway. And the worry is there might even be enough morons to fall for it in large numbers. The only thing I'd question is continuing to buy on the way down, at least in general, I'd say it maybe makes more sense do it very selectively. AI has a lot of good tech and value in the right places and applications, and a lot of the models help with things like summaries, translate and software but it's very costly, burns so much fuel and drives up power costs. Like with Internet and the dot-com bubble the tech will continue after the AI hype collapses but most companies won't. Even [Apollo had to admit](https://www.apollo.com/wealth/insights-news/insights/daily-spark/no-signs-of-ai-payoff-outside-the-tech-sector) the ROI is not that great and tokens costs way too much, and often even disrupts work more than helps. And now China is getting it's own EUV, GPU's and memory chips online so the AI models prices getting even cheaper, and the AI companies in hundreds of billions and debt and losses won't be able to raise their token prices, make a profit much less pay down that debt. Same goes for the hyperscalers and and neoclouds, even the shovel sellers got in too much debt at dumb valuations so they're falling too. Would make sense to buy selectively in small amounts but be careful, there's still a lot of leverage and excess has to unwind first.
Yeah there was never going to be a soft landing with that amount of extreme leverage and imbalance in AI and the neoclouds and hyperscalers, it was too dependent on extreme speculation and margin just like in 1929. And although it's true the Blue Tsunami puts in some tough anti-AI policy right away (Texas and Georgia are just tip of the iceberg with hard stops on data centers), it's much worse because the push against data centers it's bipartisan. Basically all of the data center ban initiatives passed and those take effect immediately, and many of the [harshest restrictions were actually passed in very Red areas like rural Ohio](https://www.toledoblade.com/local/politics/2026/11/04/data-centers-ohio-voters-reject-large/stories/20261104104) or [Indiana](https://www.ipm.org/news/2026-10-22/over-half-of-indiana-counties-now-restrict-data-centers). And the GOP state candidates like Donalds in Florida survived only because they read the room and went anti-AI bubble and opposed data centers themselves, and Florida with it's sinkholes hurricanes and storms would be the worst place to build them anyway, insurers won't cover them there. So there's not even any consolation prize and nowhere for the hyperscalers to run to. And now with China Europe and other places making their own cheaper frontier AI and chips (and China with it's own EUV, GPU's and chips supply coming online), the AI models already here can't raise their prices or they lose business. (Or bankrupt the customers they do have). So the shovel sellers take a hit too. This is why it's never a good idea to put so many economic eggs in one basket with so many companies priced to impossible perfection. Especially sector depending so much on massive debt and hype, and huge PE ratios and overvaluations even outside AI tech. Even with the massive falls today, TSLA still has a ridiculous valuation more than other car companies combined even though it's sales way below Toyota, VW and GM alone and it's lost the EV industry to BYD (and soon to Geely), way behind Waymo and even Mercedes in self-drive and Tesla not innovating as much. Much farther to fall, although in being fair they're hardly alone, Palantir has been same and even Nvidia's price for perfection had been ignoring cheaper GPU and TPU options, in general and now from China, no way to maintain those margins or prices. Heck even Apple couldn't sell enough for that kind of valuation, and Coreweave and Oracle have near junk bond status, just when debt re-finance gets pricier.
My day job fixing EUV masks agrees with his statement as well..
I’ve spent the last 20+ years watching Intel CEOs that were nothing but accountants. They may have started in stem, but they became accountants. Today’s ceo is a VC with a stem degree. I remember some statements from Craig Barrett 20+ years ago and I thought, “what a sh@tty thing to say about engineers in your own company.” About 10 years ago, I listened to the then Intel ceo say that they didn’t need EUV to hit 10microns, they spent years not being able to produce 10microns on that process, and finally having to give up to buy EUV machines from Asml. In basketball, a great coach can’t teach height. In vlsi design, you can’t teach small, it is that simple. You always need to be smaller and you always need to be on the absolute cutting edge of manufacturing. Intel’s manufacturing is second rate today. TSMC is the industry leader. I remember when TSMC was a joke. I remember “real men have fabs.” Intel took their lead I watched as Apple submitted bug reports after bug reports in skylake and Intel didn’t get them fixed. I watched Intel poopoo low power. No, mobile is not laptops with people dragging around power bricks and Ethernet cables. I remember shaking my head 20 years ago in this. Now, this was specifically not just Intel, but Intel was in that group. I remember when Intel said that the iPhone wouldn’t ship in enough quantity to justify producing for Apple. I knew in 2006 that was a horrible decision. Intel is just full of incompetence in my view.
I’ve spent the last 20+ years watching Intel CEOs that were nothing but accountants. They may have started in stem, but they became accountants. Today’s ceo is a VC with a stem degree. I remember some statements from Craig Barrett 20+ years ago and I thought, “what a sh@tty thing to say about engineers in your own company.” About 10 years ago, I listened to the then Intel ceo say that they didn’t need EUV to hit 10microns, they spent years not being able to produce 10microns on that process, and finally having to give up to buy EUV machines from Asml. In basketball, a great coach can’t teach height. In vlsi design, you can’t teach small, it is that simple. You always need to be smaller and you always need to be on the absolute cutting edge of manufacturing. Intel’s manufacturing is second rate today. TSMC is the industry leader. I remember when TSMC was a joke. I remember “real men have fabs.” Intel took their lead I watched as Apple submitted bug reports after bug reports in skylake and Intel didn’t get them fixed. I watched Intel poopoo low power. No, mobile is not laptops with people dragging around power bricks and Ethernet cables. I remember shaking my head 20 years ago in this. Now, this was specifically not just Intel, but Intel was in that group. I remember when Intel said that the iPhone wouldn’t ship in enough quantity to justify producing for Apple. I knew in 2006 that was a horrible decision. Intel is just full of incompetence in my view.
Corning is treated very unfairly. They are indispensable for EUV lithography and photonics, and they make pretty good plates too.
Absolutely this. AI is obviously an important tech that will stay around in some form, but that's a very different conversation than figuring out which specific companies or areas will actually profit or monetize from it. In a lot of ways that AI bubble even more dangerous for the US economy than the dot-com bubble it's not only due to sheer size, but because the competition is so much tougher and more advanced, and very hard to make a profit. Back in the 1990's China wasn't a factor in making Internet software or sites, now it's become the leader in research and innovation. This a factor often gets forgotten, China isn't just distilling Claude or gpt anymore to make the open weights models, they have their own advanced frontier and proprietary AI models and those source the new open weights. And they do it for 90 or 95% cheaper, so any business using the open weights has a huge competitive and cost advantage to the ones going broke from token-maxxing on the more expensive AI's, so there's no moat. And no way for Anthropic, SpaceX, openai to make a profit much less pay down debt esp as it gets more expensive, and that's taking down the also indebted hyperscalers and even the Big Tech and shovel sellers too. It's not even about the "just 6 months behind" meme anymore, now the open weights models are extremely good right out the gate, often even surpass the proprietary frontier ones at fraction of the cost. Not just in benchmarks, they[ satisfy actual business needs at much less cost for the businesses](https://qz.com/ai-us-china-open-models-deepseek-qwen), and a lot of businesses [already using them](https://finance.yahoo.com/technology/ai/articles/businesses-experimenting-cheaper-chinese-ai-185931713.html) and [making a full switch to more affordable open weights](https://www.cnbc.com/2026/07/07/chinese-ai-models-costs-us-openai-anthropic.html). Kimi K3.1 Pro open weights is just dropping and it's at or above the proprietary ones in benchmarks but more important, it handles all relevant business needs for 90% cheaper. And that's just tip of that iceberg (and with some non-Chinese models like Mistral also get very good), with China now bringing it's own EUV, GPU's and RAM online, the costs are getting even cheaper. So it's not about questioning the AI tech itself, that's a bit of a straw-man of course it's important and staying in some form. In fact if you look at it's history it's not even really new, machine learn and AI have been around for many decades and even the new generative focussed AI has been going on longer than normally realized, AI is already widely used as a tech. It's a question about the companies, investments and profitability, and issue is there's no realistic way for the AI companies and hyperscalers to be profitable with such massive costs and waste, much less pay down their debt. And so much funds got misallocated they're now even dragging down the shovel sellers, Big Tech and the whole US economy along.
" If someone figures out a way to do inference much more cost-efficiently than any of the current operators, and manages to keep it a secret, they will probably wipe out all the current industry leaders." That's looking more and more like some of the Chinese AI's right now (and even some non-Chinese like Mistral), especially Kimi, Qwen, Hunyuan and Deepseek. *Especially* Kimi. The Kimi K3.1 Pro is open weights but 95% cheaper than what Anthropic, SpaceX and openai putting out, and right at the level or above in the measurements. There actually was some secrecy around the release, along the side with the quiet development of their own EUV and GPU's but it's such a good model they couldn't keep it a secret for very long. This kind of thing puts a very low ceiling on what Claude, gpt or the other's can actually charge. No moat and no way to turn a profit much less pay back hundreds of billions in debt when cheaper better options with better inference and training are constantly around the corner. Even as it is companies [already switching to the open weights](https://www.cnbc.com/2026/07/07/chinese-ai-models-costs-us-openai-anthropic.html) otherwise they [go broke](https://finance.yahoo.com/technology/ai/articles/businesses-experimenting-cheaper-chinese-ai-185931713.html) from token-maxxing or their AI budgets eat up their whole overhead. It's why Anthropic rushing out the IPO now, before the era of companies all [clamping down hard on their AI budgets and](https://qz.com/ai-us-china-open-models-deepseek-qwen) switch more to open weights.
This, and too the Chinese AI's are no longer just distill or copy from Anthropic and openai, they're doing their own original research, making their own proprietary and frontier AI and making their open weights from that. 5 or 10 percent of the cost of Claude or GPT but same or even a bit better performance. Individual and business users aren't going to turn down that level of extra value. And now with China having it's own EUV, GPU's and RAM the costs are getting even cheaper for their AI's while quality goes up. Even some non-Chinese AI's (Mistral) getting good very fast but the Chinese AI's have the best quality for price levels and keep getting better. Anthropic, SpaceX and openai have just been burning cash right along with their hyperscalers and now even the shovel sellers with all the ridiculous circular financing. Making wild promises without a moat, Now debt is much more expensive to re-finance, they're well aware and Anthropic's IPO with it's bizarre terms (basically cash out on Day 1 for Dario and the insiders, and F-shares to lock retail out of decisions) is basically screaming it's a rug-pull desperate for retail regards to pile in and become bagholders for exit liquidity at their expense.
Yeah the danger is mainly how AI is bringing down even the oldest once sure things of Big Tech and even the hyperscalers and shovel sellers. I mean Oracle has been around since the 1970's and it's bonds are basically junk bond level due to all the disasters with their AI spending. Not only Meta but Google Alphabet now has negative cash flow and it's hurting it's ad and search business, Amazon is getting hurt and even Microsoft is starting to see heavier financial burdens, even their bonds are going way up in cost, to say nothing of the nasty outlook for Coreweave and SpaceX. Nvidia got entangled in the circular financing and pie in the sky profit estimates they can't fulfill, GPU's are sitting and rotting becoming obsolete by data centers never being built or utilized, at same time Google and other's are moving away into their own GPU or TPU supply and China now with their own EUV, GPU and RAM options come online. Even Apple, while less exposed is taking sales hits from the crazy memory costs now (may have contributed to fiasco of their latest roll-out). So directly and indirectly, the entire US tech sector is being brought down by the AI bubble. It's another reminder why all the hype is bad, even for a good new tech the allocation has to be proportional to what it can actually bring in.
Again that isn't true anymore and about a year outdated, the Chinese models especially are doing thier own R&D and have the most innovative labs, the big and even smaller players there make their own frontier and proprietary and then use that for open weights. They haven't relied on Claude for a longtime and their own work surpasses it, and now with their own cheap EUV, GPU's and RAM coming online they can make things even cheaper but higher quality. Not trying to say it's just China, there's a lot of non-Chinese AI models innovation too like Mistral that got surprising good fast, and even places like Japan, Switzerland, Korea and other places are making their own custom fitted models for their own countries that are quite good. But it's not a matter of just scrape or distill Anthropic anymore, that hasn't happened for a while there's lots of other innovation centers doing it well or even better.
Yep, no moat. And to make it even worse, the ultra-cheap Chinese models are even getting smarter but staying just as cheap or even cheaper and they and Baba have their own proprietary frontier models, and now new EUV and GPU tech to keep the prices down and the token price ceiling even lower. And then surprises like Mistral improving fast and getting business too, and other open-weights competing for now cost-conscious business AI token users. Literally no way for Anthropic to become profitable much less begin to pay down it's still huge growing debts.
This. It isn't just the lack of a moat, it's lack of even a partial, leaking moat when Anthropic **would need $800 billion in revenue just to break even**, the circular financing mass fraud and the most painfully obvious pump and dump and rug-pull schemes blatantly structured into the IPO, It's such fantastically stupid pie in the sky math to begin with, even if they had a moat there's not enough free cash in the economy to ever make them profitable, the AGI and ASI talk are pure BS (and not controllable and useless to businesses if they even were doable), even the AI boosters Apollo admitted [the ROI sucks](https://www.apollo.com/wealth/insights-news/insights/daily-spark/no-signs-of-ai-payoff-outside-the-tech-sector), [Americans overwhelmingly hate AI tech shoved down their throats](https://futurism.com/artificial-intelligence/americans-turn-against-ai-poll) and pushback gets infinitely worse, and then the whole [circular financing](https://www.bloomberg.com/graphics/2026-ai-circular-deals/) circle jerk that's basically[ Enron fraud mixed with Bernie Madoff](https://www.openmarketsinstitute.org/publications/brace-for-impact-report-release) and the [whole 2008 financial crisis](https://www.thetimes.com/business/technology/article/imf-warns-technology-circular-deals-systemic-risk-global-financing-d9v2wxsms) derivatives mess [on steroid](https://informedclearly.com/en/ai/62307/bis-circular-ai-financing-systemic-risk-2026)s. All this while debt servicing and financing are soaring indefinitely into the future with over $40 trillion and soon $50 trillion in public debt and almost $20 trillion in household private debt in the US, which means more and more pricey bond costs and the junk or near-junk ratings for Anthropic, Oracle and Coreweave are only going to get massively more expensive and crushing. **And Dario and the other insiders clearly know it**, that's why the almost stupidly insulting terms of the IPO that practically scream "come and be a bagholder for us to cash out at your expense, retail dumbass" And here's the thing, as bad as that (still horrendous) "best case scenario" would be, it's even worse because Anthropic has no moat. Open-weights generally, not even just from China and Europe (maybe soon even Japan and Korea too) can just be the mole in whack-a-mole, [pop up perpetually](https://qz.com/ai-us-china-open-models-deepseek-qwen) [with cheaper](https://finance.yahoo.com/technology/ai/articles/businesses-experimenting-cheaper-chinese-ai-185931713.html) [and](https://enlyft.com/tech/products/mistral-ai) often [better AI options](https://www.cnbc.com/2026/07/07/chinese-ai-models-costs-us-openai-anthropic.html) that any business would have to use to survive or they get token-maxxed into [bankruptcy or huge overhead costs](https://imfounder.com/science-tech/ai/ai-token-maxing-crisis/), like [has already happened](https://businessmodelanalyst.com/ai-token-costs-tokenomics-foundation-enterprise-spending/)\--prompting companies to introduce strict token use restrictions, more reason Anthropic's revenue fantasies are mathematically impossible. And despite all the copium, China isn't just copying or distill frontier AI's, they're doing a lot of the original research now, with more efficient and working AI's and their own EUV and GPU supply chain and cheaper conditions in general, means a low ceiling forever for AI token usage and no moat or profitability. Which Dario and the Anthropic club clearly know, thus the bizarre terms basically allow them to cash out on Day 1, and get F-shares so retail shareholders are even more powerless except to be regarded exit liquidity. The AI bubble is even worse than a bubble because it's dragged otherwise solid Big Tech into the bloodbath and even the hyperscalers and shovel-sellers, who can never meet these insane revenue much less profit fantasies either. Anthropic, openAI and SpaceX have basically lit the biggest cash-burning bonfire in history, and it's sucking the whole US economy into it unless it bursts before causing even more damage. Even the dumbest degenerate regard should do a double-take and stay away from this money-burning IPO fire, and the whole AI hype and any associated sectors until
The value of Intel is their EUV/DUV machines, and their x86 licenses. One quarter of low retail cpu sales because of high memory prices won't change the fundamental valuation of them for the long term.
Decades? With an S? Fucking lol. CXMT did it in less than 10 years without ASML giving them an EUV machine. Now imagine a new company in Europe or America with funding and no regulations. God you people sound dumber and dumber with every comment. Just fucking LOL. Hahahahaha
Lol, you actually think some random big company can just stroll into advanced DRAM and HBM in 3-5 years and start eating Samsung and SK hynix’s margins? That’s adorable. This shit takes decades of process know-how, tens of billions in fabs, EUV mastery, and yields that don’t suck—plus customers who won’t even talk to you until you’ve proven you can ship real volume without blowing up their servers. High margins exist exactly because it’s a brutal, specialized game that most “we want a piece” companies keep failing at. Keep dreaming though.
the purpose of the EUV machines in question is to produce chips, therefore no chips means no working EUV machine for the purpose of making chips
Wrong, while the prototype can generate EUV light, the machine hasn’t produced a single chip because China still hasn't solved the ultra-precision optics bottleneck. Which further proves my point that it takes years for one to be a competitor even if they have lots of capital. Nobody can just show up and become a competitor like you were suggesting
Those machines are made by ASML and are run by installed by ASML. ASML doesn’t want to sell to China because they will steal their IP. And to be clear, China stole/piecemealed an EUV machine back together and hired former ASML to help them reverse engineer/build it. So yes, China does in fact have an EUV machine. Again, please do two seconds of research before talking. You sound stupid.
SK hynix has EUV machines while China doesn’t, china wants to build their own but couldn’t, even when they have massive amounts of capital. It’s not just about capital in this case like you suggest.
Please tell me how SK Hynix hasn’t produced their own EUV machine yet then. Oh wait, that would ruin your whole argument. Fucking clown show around here.
Then could you please explain why China hasn’t mass-produced their own EUV machines yet? Clearly the state has enormous capital that can be used for anything.
Then could you please explain why China hasn’t mass-produced their own EUV machines yet? Clearly the state has enormous capital that can be used for anything.
Yes it may happen that more companies substitute with open source. However this has been said for years now and never been done. The point I made is that these commitments are largely compute commitments and are also being payed by their revenue and not solely by reserves. Then whether that growth will keep happening is speculation however as of now Anthropic is still growing rapidly. A different point is that China only has a fraction of the HPC AI Compute as they either have to get it “illegally” or use more inefficient chips. They don’t have any access to EUV Chips so own domestic chips are much much more inefficient.
They are limited by number of customers, cannot sell to china and their uber expensive EUV machines have limited production volume. I dont think they are cheap in any stretch despite trading this low. Memory makers on the other hand. We treat them as cyclical while they have presold inventory/locked prices for years into future and are trading for single digits P/E looking into the future. They have wider customer base and their TAM is only increasing with Robotics requiring huge amount of memory as well along with AI and even Autos.
I hold ASML since 700$ because it was defensive play, after jump to 1100$ and back to 500$ i was bit depressed. But bought more. Now I sold at 1300$ because of concentration risk but the stock is solid compounder for years, IMO you cant go wrong with it. But it's cyclical, let's assume CAPEX stays or goes lower - whos gonna ramp up production? Current spike is 100% due to memory / tsmc. This might be a supercycle, but there will be peak, anyone who think this is straight up line for 10 years must see beyond initial AI buildout phase. My theory could be raise of robotics in 30's, probably all robots, all autonomous vehicles need chips, but do they need EUV? Not sure. Perhaps? Anyway Im fine sticking around 10% of my porftolio, would trim further if its back to 15% again.
Korean exports this year is almost equal to Japan even when you exclude semiconductors. China is still 3 or so years behind in HBM tech, and there's still a question whether they can match EUV scale base/logic die performances wit DUV technology. Given that we have just entered the era of HBM4 (which uses EUV-scale nodes for logic dies), the gap there might widen even further. It's also something to note that Samsung has their own foundry. If the rumors are true that Samsung will mass produce Exynos 2700 and the 5G modem on the 2nm node, they might pull up 2nm yields dramatically like they did with the 4nm node, catching up to TSMC. When we say "Korean shipbuilding" we really mean VLGCs and LNGCs. There's such a huge LNGC backlog around the world right now that Korean shpbuilders aren't worried for the next 5 years. Even then Canada is increasing their LNG output to the Pacific over the next few years which means that the demand will keep going
The three aren't one trade. The dispersion inside that 50% is about China exposure and memory/logic mix, not about AI beta. AMAT, LRCX and KLAC sell into the same WFE pool but their geographic revenue split and their memory vs logic exposure differ a lot. A 50% drawdown is the market pricing a 2027 WFE cut. What actually decides whether this is the bottom is (a) how much of the 2025-26 AI capex was pull-forward from 2027 rather than new capacity, (b) whether NAND capex restarts, and (c) China domestic tool share continuing to climb. ASML down only ~22% is the tell, and not because of valuation: EUV backlog gives visibility that deposition/etch/inspection orders don't have. That's an order-book difference, not a multiple difference. What I'd check before calling a bottom: each company's latest 10-K geographic revenue split and the deferred revenue trend. If China share is still in the high-20s or above and trending down, this is a repricing, not a discount. (No position. I keep a WFE tracker for my own notes.)
Chinese can talk all kinda shit but the fact to the matter is without EUV. huawei chip is stuck at 5 or 7nm with very bad yield and performance. instead keep wasting money, buying Nvidia chip can at least get Chinese LLM model on par with US SOTA and maybe made money. money talk, loser walk
Strong quarter, no argument there. My problem is the price. I ran a few DCF models and the middle case comes out near $700 a share, against about $1,650 for the ADR on October 15, and even the 90th percentile of my simulations gets to roughly $1,175. I might be too harsh on how long the EUV lead lasts. What revenue are you assuming for 2030 at this price? Either way... i would still wait for a pullback.
Did China break ASML’s EUV monopoly or are you talking out your ass? Because only one of those can be true.
It's a lot of money when a lot of the projects conducted at this 'Trinity Fab' will go through trial and error producing no profit. The whole point of this is to domesticate photoresists, EUV masks, and domestic testing and verification equipment after Japan tried to ban exports of those materials in 2018 using their 97% market share leverage in hydrogen fluoride (HF) and 90% market share in photoresists. Currently Korea's domestication rate of semiconductor chemicals and equipment is 30%. The goal is to raise it to 50%
Not OP, but thought I would chime in some color commentary on what I believe is more relevant today regarding "nationality" and governance, because ASML EUV is already governed by the USA, which is restricting the sales of EUV to China, etc. Another risk to growth today for ASML is the MATCH Act proposed by the USA, which would potentially lead to further restriction of maintenance and tooling sales to China, etc. The spike in (DUV tooling) sales in Q3 to China is a signal that China is preparing for the MATCH Act to pass as a real possibility (however low % of a chance). The progression of the MATCH Act is very important as an ASML investor or potential investor.
As someone in the semi industry, the most volatile portion of the business is the tool. They are the most expensive part in running a fab and it's the first thing the business cuts down on buying whenever there are stormy waters ahead. The other thing that needs to be looked at is TSMC has said it doesn't need the latest and greatest EUV tool from ASML for their next process nodes. Additionally, Nikon and Canon are making real progress on nano imprint lithography tools would could become legit competitors to the EUV platform in 5-10 years.
I don't think it's going to be a threat to the mainstream chipmakers like Nvidia (and maybe, just maybe AMD can elevate themselves into the same room). What you need to imagine is the availability of Huawei Ascend chips which China domestic models are going to be trained on (rather than for pure inference) with domestic HBM. Sure they are stuck 2-3 generations behind with far lower yields due to a reliance on DUV, but you'd bet on China doing something creative to get the parts for high NA EUV put together in the next couple of years
I’m abundantly aware about branding modules, bot. And no, CXMT hasn’t surpassed Samsung. You have to be a bot or literally retarded. Can CXMT make HBM4? GDDR7? DDR5-9600MTs? No it can’t. CXMT can make average speed consumer ddr5. That’s a sliver of the current market. They don’t have EUV lithography so they won’t be advancing much further.
the Q4 guide adds up to \~€44-45B for the year, the top end of what they promised in july. "we can't build enough through 2028" is a hell of a flex when you're the only company on earth making EUV machines
Yeah, pretty sure I said nothing of the sort. I literally stated China has a ton of Engineers. They are great at copying and then mass producing. Shouldn't be shocking considering 1.4 Billion people and their students aren't encouraged to get degrees in Art-History, French Poetry, or Women's studies. All I'm saying is that some shit is hard as hell to copy or everyone would do it. You don't think American companies haven't thought about getting in on the EUV game? All you have to do is license 1000 patents from the Netherlands, Japan, and the US (assuming you aren't going to commit IP theft...), train your engineers on tech that they have literally never seen before, and then mass produce it. [https://turbli.com/blog/why-is-it-so-hard-for-china-to-develop-jet-engines/](https://turbli.com/blog/why-is-it-so-hard-for-china-to-develop-jet-engines/)
lol, based on Huawei's AI chip roadmap, CXMT is shipping HBM3e next year and hbm4 by 2028. CXMT's 4F is an improvement to current 6F dram cell architecture, beating SK and Samsung by 1 year. 4F require advanced hybrid bonding. Huawei's logic folding Kirin 9050Pro SOC also requires advanced hybrid bonding. This shows Chinese semi tools and fab companies are quite advanced in hybrid bonding and 3D chips, equal to what TSMC, SK and Samsung can do. And hybrid bonding is fundamental to HBM. But I agree CXMT will not materially affect Mu's earnings in the next 3 years because of 1) CXMT new volume will be fully taken by domestic Chinese industries, 2) unlike conventional wisdom Chinese companies will sell at lower prices, CXMT is pricing the same as international players, even higher, they like to make money too, 3) demand of US based companies still far outstrip supply in the next 2-3 years, and 4) US government protectionism. A reasonable guess is China will have production EUV machines by 2031. After that, no one can estimate the supply and margins. But I would argue the memory industry will maintain healthy margins going forward.
Oh, China has tons of engineers and have the advantage of being able to copy. They are not inventing the wheel. But there's a reason why they still need Boeing and Russia for Jet engines. FFS, not that long ago, they couldn't make pens... Attempting to copy an EUV machine? That shit will take a decade, minimum and 10s of Billions, if not 100s. It's been properly compared to sinking a hole in one.... from the Moon. And China isn't exactly great at allocating capital or even raising it. It's not a shock tha their stock market is at 2008 levels...Even worse, it's up a whopping 8% over the past 5 years. That while the entire world's markets were mooning off the COVID lows. [https://www.washingtonpost.com/news/worldviews/wp/2017/01/18/finally-china-manufactures-a-ballpoint-pen-all-by-itself/](https://www.washingtonpost.com/news/worldviews/wp/2017/01/18/finally-china-manufactures-a-ballpoint-pen-all-by-itself/)
“with no customers and struggled to survive. It has now built an almost unassailable position in a leading global industry. Jamie Ward reports Inside a “clean room” in Veldhoven in the southern Netherlands, technicians assemble a machine that sits at the heart of the modern semiconductor industry. An extreme ultraviolet (EUV) lithography system is roughly the size of a double-decker bus, weighs roughly 180 tonnes and costs more than €180m. It fires lasers at tens of thousands of microscopic droplets of molten tin every second, creating a plasma that emits light at a wavelength of 13.5 nanometres. That light is used to print the extraordinarily fine patterns from which the most advanced computer chips are made.”
i think the drop of quarterly bookings pushes the bar up a bit – you lose that independent check, so it feels more like a free‑hand number. but the big price hikes and the backlog for High‑NA/EUV builds give some cushion – not a surprise slash in the first half. in short, i trust the guide grudgingly, not as hard as the 5.6% option move would suggest.
Without bookings, I'd lean on backlog commentary and how many High-NA and EUV systems they say shipped versus were recognized, since revenue recognition timing alone could explain a lot of that implied Q4 jump. Honestly, losing the bookings line makes me trust the guide a bit less, because a wide €43-45B range with no order book to cross-check leaves management a lot of room to steer the narrative.
The entire semiconductor industry is screwed. Everything has to work together, so everyone has to work together to build stuff that works together. That means ANY change has to be done by the ENTIRE industry. Chip fabs, their suppliers and the suppliers of those suppliers, and so on. The switch to EUV is one example, currently they are working on switching the mask size, which is gonna take years and needs the majority of the industry on board to even attempt. So what TSMC is getting out of this is cooperation with the new kid in town. A kid that is known to just do stuff itself whenever nessesary, and you really wanna make sure that kid works on the stuff everyone is working on, or you end up with a split industry, and that WILL result in the eventual death of one arms of that industry. Cant afford that to happen, so nip it in the butt from the very start. Literally everyone wins.
> Intel famously did not adopt low NA tools and instead opted for quad patterning which caused process issues, not because they struggled with next gen litho tools. And Intel did that mainly because EUV was simply not ready for the planed launch. 10nm (now Intel 7) was supposed to be out in 2015/2016 already. When TSMC launched their 7nm in late 2017 and later volume production in 2018 it was using DUV as well, for the same reason. TSMC only slowly started using EUV for key layers with further iterations of 7nm.
"Flooded" int he world of EUV is funny.
Ah, okay, that's right, I forgot, China is bad and stupid. And can only copy. And copy badly. Got it. Sometimes I think yanks having got past the 'Deadwood' stage of thinking of the Chinese. Something something "manifest destiny" something something something "exceptionalism" something something. Whatevs. Me personally, I'd love to see a few more countries making chips, might bring down the price. Fuck me the RTX cards are expensive now. Also might relieve some of the of the Taiwan pressure, innit. Here's hoping they get the EUV cracked. Unless you're holding such stocks long? Share the info. Get the stuff made.
Yea, so basically what they are doing is using DUV machine’s (Old generation tech) and to achieve smaller patterns they do multiple runs. But that technique won’t work at scale & reliability as it for one takes more time per chip & Yield(Working vs defective) is lower. ASML has made a machine that does both HIGH tech with an impressive EUV light source(Which is one of the most important parts) that allows for smaller patterns & makes it economically scalable at cost. The complexity of the EUV machines with the mirrors from Zeiss(Only manufacturer in the world based in Germany) & the light source(Just look it up on youtube and just imagine the precision and BIG complicated & hard to develop systems that requires), makes me quite confident they won’t catch up as easily. And if they do its prolly due to Spying(Which they have/tried many times) and on that front its also getting harder for them since the government is cracking down on Chinese PHD students being allowed for that research & background checks getting Spy levels of security.
I'm an employee. Yes we stopped at 12nm, EUV was too expensive and we diversified in 12/14nm nodes. We have numerous customers that play in this area for low power electronics. But no, we do not produce CPUs/GPUs but if you look at the content in your phone, we do produce more of the chip area. Also take a look at what is to come. Photonics and Quantum do not require sub 10nm printing to produce a QPU or Photonic modulators.
How do we feel about EUV vs LYTE or FOTO, besides the fact that the latter two are more expensive?
I listened to a podcast on this from actual experts. "Most scientific papers" =/= better. I can produce a billion scientific papers on whatever subject and claim to be the best. Meanwhile, Joe can produce 2 scientific papers that completely revolutionizes technology as we know it. And this is *exactly* how China tries to manipulate perception on the global stage. They commonly use the tactic of "Most = Better". If they are so great, why haven't they cracked EUV? Or High-NA? With all those "most scientific papers", you'd think they'd have cracked EUV by now. The technology is already pretty old.
Calling it advanced money destroyer isn't serious analysis, that's meaningless. The fundamentals also changed for AMD when people saw the growing importance of CPU's for AI data centers. AI cap ex has also far exceeded expectations. Even just this year, ASML's projected EUV machine totals for 2028 went from 70 to 90 to 110 and now they are trying to figure out if they can build more than 110 because demand is that high. You haven't show that the fundamentals were there for AMD in 2025 but it was irrationally mispriced.
They do use EUV for the best chips, but that's irrelevant to China's independence from the one single company in the entire world.
I didn’t say they don’t use DUV anymore I was under the impression that they use EUV for the best chips. But there still is a difference between Chinese DUV and ASML’s DUV. That’s why I sad they might have a chance. And yes I know it’s the US the reasoning is geopolitical.
Again, use some common sense, EUV exists, but it's sold out like you mentioned. On top of that, they'll just stop using the DUV machines even when they can still make a fuck ton of money off of it? How does that make sense to you? ASML can't sell to China because the US tells it that it can't, look it up.
I didn’t know Micron uses DUV as a cost cutting tactic. I assumed because the two Korean players are on EUV. Maybe CXMT has a chance on their own + Chinese DUVs. And ASML can’t sell to China because of geopolitical reasons, look it up.
Micron still uses DUV for its HBM production. EUV is nice, but again, DUV isn't obsolete. EUV is not easy to make, that's for sure. But it's not something that has a progress bar that advances constantly at the same rate. A breakthrough can happen that accelerates it to just 3 years, or even 3 months away. Or that breakthrough never happens. We all know that it's possible, because it exists. And China has the largest number of engineers in the world. CXMT wasn't on the radar at all even last year, but right now it's being talked about a lot. ASML is not an US company, so while the US has a ton of power over the Dutch company, it's still a Dutch company. So I don't see how the US has permanent grip on it to have the decision to not sell China EUV permanent like you claimed.
Xi Jinping. I believe SKHY and Samsung resumed production after cancelling and I believe Micron is still not making any consumer memory atm following their Dec 2025 announcement to stop and focus on HBM. HBM is gold right now. I know AI demand won’t last forever but as long as it does China isn’t getting any EUVs for HBM because of restrictions and ASML is sold out thru 2027. And when the slow down happens which it will just don’ know when, China might have their chance to get their hands on EUV but it wouldn’t really matter anymore. And if China decides to build I don’t think this AI build out will last 20 years. There is a reason ASML is the only player and they are the only ones to figure out such tech. It ain’t easy even if you have all the money to build.
Calling memory easy to make is wild. Samsung, SK Hynix, and Micron control \~95% of global DRAM/HBM capacity—you can’t just 'choose' to build an HBM fab overnight without $15B+ and ASML EUV tools. But you're 100% right on the software moat. CUDA locks developers into Nvidia, whereas standard DRAM is interchangeable. Though with HBM4 using custom logic base dies, even memory is getting physically locked into specific GPU designs now. HBM is not interchangeable as it used to be. They’re locking more and more.
CXMT is catching up quickly. I know you want to claim EUV as the ultimate machine for all other DRAM makers. That's true, CXMT can't make EUV yet, and so is Samsung and every single other chip maker. No one except for 1 single company can make EUV, the only problem with it is that CXMT can't buy it. So, what if CXMT gets the green light to buy it just like everyone else?
No, they are just slightly behind much like their LLMs. EUV is irrelanent in this case.
You don't need EUV for consumer grade dram
CXMT is years behind. They still don’t have EUV machines.
Where did you get that they were at micron capacity? And if they are, which i doubt, since they don’t use EUV, for them to get the same feature size, they have to get smart with the way they pattern their wafers, which causes much lower yield. Plus new advancements are always harder fort hem due to them not having EUV
The money is in advanxed / leading edge … I just looked up that $$$’s, 90% of profits are based on EUV manufactured processors because that’s where the margins are. So while GFS may be stable they have low margins and many competitors they have no real MOAT and plenty of competitors doing the same. We saw this in DRAM in that while it was low margins 3 manufacturers control more than 90% of the global supply … so prices sky rocketed when demand increased there is a natural bottle neck and cartel like control on a business level Even if OPs idea would hold , they have no moat and their are 7 or 8 other companies globally that support the same fab processes. There are many other smaller but atleast 7 or 8 of equal size to them.
Physical AI will need local CPU's and inferencing. The main reason would be to break dependency on network connectivity and network latency. Imagine an autonomous vehicle needing to send what it "sees" in realtime to a back end network to process "what to do" and that also needs to get relayed back. It doesn't work in realtime applications. The bulk of chips that GFS fabs are simple, single purpose; for example the chip that controls power management on an electronics device or the chip for radio frequency for communications. Some 50-60% of their sales are for mobile phones and autos - but in the phones for example, they aren't the main fancy chip from a big chip designer. Sure GFS market will grow some. But unlike cutting edge, it's low cost to produce, higher yield, more competition and much easier for a larger chip designer to self manufacture - that would not be the case for EUV. You mention the likes of NBIS and CRWV which are growing revenue at over 3 digits percentage points. Yet GFS revenue has been more or less flat or yo-yo over past several years - and this would have been the timeframe for digital transformation and IoT explosion, yet as of today they have no sustained growth to show for it. So that's why some are "hot" and some are not.
The numbers aren't wrong but they are a severe outlier, Micron's own history shows that. Commodity memory has never sustained anything over 60% for long because the capex cycle always kicks in. Someone builds more fabs, supply catches up, margins compress back to earth. The only precedent I can think of that even approaches this is ASML during the EUV monopoly window, and even they couldn't hold 85% indefinitely because their customer base is too concentrated. Memory is way more cyclical than lithography tools though, so I'd be shocked if this lasts past mid-2027
Quick history lesson on GF- It was founded as the conglomeration of AMD’s Dresden Fab, some IBM fabs in upstate NY and Vermont, and Chartered Semiconductor in Singapore as AMD went fabless. This was accomplished via the relevant business units being purchased by Mubadala, the UAE’s investment arm. First few years AMD remained the majority of the revenue but it quickly became a struggle as AMD wanted to compete at low tech nodes with Nvidia and Intel, but GF couldn’t afford the insane R&D cost that would come with that race, so in 2017 they announced a pivot away from EUV technology to focus on “essential” chips. This means things like radio modems in smartphones, car sensors, power switches, etc. - all the stuff that is needed for that 2nm chip to actually do what it needs to do. This has been a heavily questioned move - many felt like giving up on EUV was essentially giving up on being a market competitor. But it’s proved to at the very least be a stabilizing move, allowing GF to spend the last few years slowly modernizing their internal business processes, something that has shown in the fairly steady margin growth from roughly 25 to 30% As you mentioned, they’ve begun investing heavily into new non-EUV tech; silicon photonics has been a boon to their revenue as the smartphone market has plateaued, and there are legitimate reasons to be excited about GF’s placement in the quantum computing space - their 22FDX platform shows some promise for control modules at cryogenic temperatures for example. Now, do I think that there’s reason to think gf is undervalued in the mid 40s? Absolutely. But I think the “right” price for the stock is probably in the mid 50s to low 60s for the next couple of quarters. If they can start showing some major demand and capacity ramps in the coming quarters then they could shoot up, but I doubt you’re gonna 5x your investment anytime soon. It’s still a very immature company (in the grand scheme of things) that’s gotta get through some growing pains and find the market they can really own.
They are not the same use case. intel is 1 of 3 leading edge companies in the world and the only American company that can do what they do. Intel has a massive already client processer business, Intel also has a massive IP portfolio and is intrical to national security, GFS is no that, they don’t have that type of moat nor importance to the global market or the US plans. An intel buy last year was a totally different thesis and not just well this company could be more important , intel was already important and its concerns were foundry investment which is huge for high-na EUV -‘d leading edge.
No, not likely as that’s not their stated goal. The marker they play, they focus on trailing edge nodes 180nm to be 20nm I believe they can do 12nm today … but these sizes are used for standard electronics and microprocessors. They gave up on leading edge and EUV in 2017 not having the capital to play in that domain.. so advanced CPUs, GPUs are not in their realm of capability and also the domain they do play in has plenty of other players. This is why Samsung, TSMC and Intel are so important they are the only leading edge foundries in the world.
As long as the EUV machines keep shipping!
There is another part of this discussion that makes the whole “Chinese auto industry vs American auto industry” framing much less clean than it sounds. Tesla itself is a good example. Tesla is American, but its battery supply chain has never been purely American. Panasonic was crucial to Tesla’s early scale, and Tesla later diversified its battery sourcing to companies including LG Energy Solution and CATL. So even the most successful American pure-EV manufacturer was built around a global, heavily Asian battery supply chain. Brazil makes this even more interesting because we are watching several versions of this industrial integration happen at the same time. BYD took over Ford’s former industrial complex in Camaçari and is progressively localizing production. GWM took over Mercedes-Benz’s former factory in Iracemápolis. Renault and Geely are expanding their industrial relationship in Brazil. GM itself is now assembling Chinese-developed Chevrolet EVs in Ceará. The Spark EUV and Captiva EV come from GM’s Chinese industrial ecosystem with SAIC and Wuling and are being assembled locally. Toyota and BYD have a 50/50 EV R&D joint venture. Nissan has had a major industrial relationship with Dongfeng in China for decades. At some point, asking whether a technology is simply “Chinese” or “Western” stops having an easy answer. And Brazil gives us another interesting experiment: range. If you look only at Brazilian homologation figures, some EVs can look surprisingly short-legged. Cars with batteries in the 50-60 kWh range can receive official Inmetro/PBEV range figures around 300 km. But the Brazilian number is deliberately conservative. Inmetro does not simply publish the raw laboratory result. Adjustment factors are applied to produce a more conservative real-world reference. The problem is that a lot of automotive discussion here then treats that number almost as the maximum distance the EV can realistically travel. Owner experience often looks very different. We now have EV owners driving these cars across Brazil, including mountainous areas and routes that climb from the coast onto the plateau. There are owners and independent tests substantially exceeding the official Inmetro range, and under favorable conditions some results get much closer to WLTP. Larger-battery EVs can exceed 400 km in real use even when their Brazilian homologated number looks much less impressive. Obviously that does not mean WLTP is guaranteed real-world range. Drive at 120 km/h, climb continuously, add headwind, temperature changes or heavy HVAC use and consumption changes dramatically. But this exposes an interesting asymmetry in how cars are discussed here. When a small 1.0-liter ICE car achieves an exceptionally good km/l result, automotive enthusiasts and media are perfectly happy to show what the car can achieve under favorable real-world conditions. With EVs, I often see the opposite. The conservative Inmetro figure gets repeated as the defining range of the vehicle, while owner consumption and independent road results receive much less attention. A much better way to discuss EV range is to show the homologation numbers and then show actual energy consumption. If an EV does 12, 15, 18 or 22 kWh/100 km, anyone can understand what a 40, 60 or 90 kWh battery means under different conditions. That's much more informative than saying “this is a 300 km car” because one homologation system printed 300 km on the label. And all of this comes back to the original Detroit question. Brazil is becoming a useful real-world laboratory because Chinese manufacturers are no longer simply shipping Chinese-built cars here. They are buying former Western factories, hiring Brazilian workers and progressively localizing production. At the same time, established American, European and Japanese manufacturers are increasingly using Chinese partners, platforms, batteries, engineering or complete vehicles. Europe is moving in a similar direction as Chinese manufacturers localize more production there. That gives us an opportunity to separate two things that are constantly mixed together in this discussion. If Chinese manufacturers lose most of their price advantage when they manufacture in Brazil, Europe or eventually the US with local workers, then Chinese wages, subsidies and domestic production conditions were clearly doing much of the work. But if a substantial advantage survives localization, then wages cannot be the entire explanation. Battery costs, vertical integration, EV-specific platforms, supplier organization, procurement, manufacturing scale, automation, development cycles and margins all have to enter the discussion. That's why I would actually find an American-built Chinese EV much more interesting than another imported Chinese EV. Require American production. American wages. American safety and environmental standards. Apply the same rules to everyone. Then compare the products. If an American-built BYD ends up costing roughly the same as an equivalent American-built GM or Ford, we learn something important about where the original Chinese cost advantage came from. But if it can still compete aggressively on price, equipment and efficiency while paying American production costs, then tariffs didn't answer the underlying competitiveness question. They only delayed the experiment.
One thing I think this thread is showing is that the question is becoming bigger than simply "Chinese cars are cheap because Chinese workers are cheap." Wages and subsidies matter. But Chinese manufacturers are increasingly moving production outside China, which gives us a chance to test that explanation in the real world. Brazil is a particularly interesting example. For decades, the market was dominated by established American, European, Japanese and Korean manufacturers with local factories, suppliers, dealerships and huge brand recognition. Chevrolet was enormously strong, and the Onix spent years as the country's best-selling car. Now the structure is changing. BYD took over Ford's former industrial complex in Camaçari and is progressively localizing production. GWM took over Mercedes-Benz's former factory in Iracemápolis. Renault and Geely are expanding their industrial partnership and investing together in Brazil. And it gets stranger than simply "Chinese companies versus legacy automakers." GM itself is assembling Chinese-developed Chevrolet EVs in Brazil. The Spark EUV and Captiva EV are being assembled in Ceará using products originating from GM's Chinese ecosystem with SAIC and Wuling. Toyota has a 50/50 EV R&D joint venture with BYD. Nissan has a long industrial relationship with Dongfeng in China. Renault is partnering with Geely. So the borders between "Chinese" and "traditional" manufacturers are becoming increasingly blurry. Europe may become an even better test. Chinese manufacturers are actively looking for existing European factories rather than simply exporting everything from China. BYD says that, longer term, it expects to need three vehicle assembly plants and a battery plant in Europe. That is why I don't think the wage argument settles this. If BYD builds cars with Brazilian or European workers and most of its price advantage disappears, then labor costs, subsidies and producing in China were obviously doing a huge amount of the work. But if a substantial advantage remains, we have to ask what else explains it: battery costs, vertical integration, platform design, automation, supplier organization, scale, development cycles, margins, or some combination of them. And this is where I come back to Detroit. The US doesn't have to allow unlimited Chinese imports to test this. Require local production. Require American wages. Require US safety and environmental standards. Apply trade safeguards. Then let the products compete. Protection can give an industry time to adjust. But the important question is what Detroit does with that time. Because Chinese manufacturers aren't standing still outside the US. They're localizing production, buying or reusing factories, forming partnerships with established manufacturers and becoming part of the same global supply chains that legacy automakers use. Brazil is already experiencing that transition. Europe increasingly is too. Renault and Geely, for example, just announced another €319 million investment in their Brazilian partnership. If an American-built Chinese EV eventually costs roughly the same as an American-built competitor, we'll have learned something important about the original Chinese cost advantage. But if it can still compete aggressively on price and equipment while paying American production costs, then keeping the imported version out didn't solve Detroit's underlying competitiveness problem. It just postponed the test.
TSMC only really holds up as long as it takes CXMT to fully ramp up. At that point most of the world's advanced semis are going to switch to China purely for cost reasons. So TSMC isn't going to have the same EUV lead they previously enjoyed. In fact Intel has a new fab up right now that can match it and they're looking for customers.
No US/Netherlands has export controls on EUV machines. Only the old DUV is allowed
It should be able to design and construct a cheaper version of ASML's EUV machines first, that's the real bottlebeck.
Nike going all in on selling through their website and ending direct sales on Amazon and Amazon owned Zappos really cost them. It wasn't just Amazon though. Nike intentionally cut ties with roughly 50% of its wholesale retail partners globally. The goal was to eliminate "undifferentiated" stores, cut out the middleman, and force shoppers to buy directly from Nike’s own website, apps, and flagship stores to increase profit margins. They thought they could do it as they were big enough. What the management did not realize was they were just selling shoes not EUV machines. It allowed breathing room for unknown smaller running shoes brands. People started to realize there are other companies making similar or better products. They know it was a massive blunder because they are now back in Amazon. But might be a little too late as there is already blood in the water. They will still be dominant but their best days might be behind them.
Even without EUV? You think AI algorithms get that much better that they can achieve the same levels with limited compute? Or do you think China cracks EUV on a scalable basis in the next 5ish years?
[https://hc2025.hotchips.org/assets/program/conference/day2/40\_Marvell\_Kuemerle\_final.pdf](https://hc2025.hotchips.org/assets/program/conference/day2/40_Marvell_Kuemerle_final.pdf) Marvell® dense SRAM in 2nm Marvell IP -> but they goto TSMC Intel Foundry + Intel DCG -> products with stack SRAM (i.e. Logic on too of SRAM) Intel Xeon 6+ clearwater forest -> this is launched and should be shipping to hyperscaler. Intel Xeon 7 diamond rapids -> product announced but shipping early-mid 2027 If SRAM actual silicon, that is the logic foundry players. If SRAM IP, then Marvell and Broadcom (maybe MediaTek) some players that benefits: ASML -> because you need EUV for SRAM at those advanced nodes. Foundry -> Intel and TSMC for fabrication and advanced packaging.
There's a nike exec reading this thread right now asking Claude "How do I pivot my company toward selling EUV machines in the next quarter?".
Nike going all in on selling through their website and ending direct sales on Amazon and Amazon owned Zappos really cost them. It wasn't just Amazon though. Nike intentionally cut ties with roughly 50% of its wholesale retail partners globally. The goal was to eliminate "undifferentiated" stores, cut out the middleman, and force shoppers to buy directly from Nike’s own website, apps, and flagship stores to increase profit margins.They thought they could do it as they were big enough. What the management did not realize was they were just selling shoes not EUV machines. It allowed breathing room for unknown smaller running shoes brands. People started to realize there are other companies making similar or better products. They know it was massive blunder because they are now back in Amazon.
That will be funny but it won’t be because of these GPUs It will be because of new hardware that makes AMD and Nvidia irrelevant EUV machines and fabs built in mainland China fabs in the US all making Taiwan reunification not a US problem stock market will hate it though
I’ve saved your comment to see if China will be able to manufacture EUV within the next 20 years. Of course, if China really does succeed, some people have already written the ending to the story: “China stole it.”
It’s all going to be moot anyway. We will be using photonic chips shrunk already 10,000x for training & inference for AI. One optical processing unit (OPU) from Neurophos (8 chips) is equal to a 3,000 lb server GPU rack while drawing 100x less power. Light doesn’t generate heat nor does it require the enormous amount of power that transistors do. We are the wall of what is possible with shrinking transistors on silicon. We are stacking them, AMD did chiplets when single chips no longer could keep up and NVDA is investing very heavily into $LITE, $GLW and $COHR bc they know if they surround their shitty power hungry GPUs with photonic components it’ll make it appear like they are making efficiency gains. The only moat is CUDA & Q.ANT (German startup) already has proven in real life AI datacenters that they can work beside GPUs and pull 30x less power while doing certain tasks like inference 50x faster. NVIDIA knows this but I would be investing in all things optical unless NVIDIA has a few tricks up their sleeve. Buy INTC, LITE, COHR and GLW to name a few NOT AAOI. The ETF EUV has a nice balance but I would wait for the 3rd week in September when the great ETF, hedge fund, index fund rebalancing occurs & all those chip stocks get kicked to lower concentrations. Especially considering the money in momentum funds currently like SPMO.
Very interesting. Yeah I mean if they can do it I'm in full support. High yield advanced node fab accessible to everyone is a good thing for the world. I'm very proud of my Chinese heritage and am not one to criticize China specifically or for the sake of any East vs West argument. It just seems so difficult for them, or anyone, to create high NA EUV alone. I've seen their progress in deposition which, though extremely impressive, is still lagging a fair bit to current technology. I also just don't know how they would be able to mass produce good tools for each step and then maintain high enough yields to be successful. You can see how terrible Intel and Samsung yields are even with full access to the most advanced tools.
I was in China for about a month in February. I completely understand how competent the talent in China is. But a lot of the chip industry is not tied to talent alone. I mentioned how the US or any country alone would not be able to catch up in this industry. Each EUV, deposition, etch tool alone is a feat to create. ASML sold about 50 tools total in 2025 which comprised the majority of their $37B revenue. Companies would not be paying hundreds of millions per tool (~$200m each) if they were able to create their own. Chip fab is literally the most complex and advance form of technology we humans have ever made and it took 10 or so of the largest countries multiple decades to get to the point we are at. It just is not possible for a single country to get to that point in a fraction of the time.
batery tech is not that simple but obviously EUV is magnitude harder than battery because EUV involves nano scale and super high precision and a host of very difficult engineering problems
Oh really? Using what EUV machines?
Yeah China has definitely caught up (if not advanced) in EVs/batteries, renewables, and LLMs. I'm definitely not disputing this. But I also mentioned this in my previous comment. EVs, batteries, renewables are far simpler technologies to copy and reverse engineer. Same with ai models where they can use a lot of the pre-existing training. I would argue that just one single component of chip fab alone (EUV lithography) is far far more complex than any of the listed technologies. There's a reaaon that Nvidia is the largest company in the world and every aspect of chip fab functions as essentially a monopoly. There's a reason that China has had access to DUV tools for decades now and were only able to replicate it this year. And DUV to EUV is like the same level of magnitude as creating an optical microscope to a transmission electron microscope. I am pretty confident that nobody, outside of ASML, will be able to create high NA EUV in the next 20 years, let alone every other component in dep, etch tools and chambers.
They are working on it China appears closer to EUV capability without having obtained ASML's production-ready machine. The US allegation that an ASML EUV machine reached China remains unverified, and ASML says Washington has provided it no supporting evidence. China has, however, reached a different milestone by building its own crude EUV prototype in Shenzhen, reportedly with former ASML engineers and parts sourced from older equipment.
As a Chinese American semi HW engineer, I've heard a ton of headline propaganda from my mom/uncles about how China's tool/chip manufacturing has caught up. How China is as capable as TSMC, ASML, etc. I give them a pass because they know literally nothing about the semiconductor industry and assume stuff like DUV vs EUV litho is basically the same thing. Definitely think China has innovated a ton in spaces like batteries and renewables especially. But these are fairly simple technologies compared to chip manufacturing. Unless they are able to continuously replicate and mass produce all the best litho, dep, etch, etc tools, I genuinely don't think they will ever fully catch up. I don't think the US alone would be able to either.
Won't happen unless they somehow pull of the miracle of mass producing their own EUV in the next years
I need you to go ahead and research how they are pursuing EUV. They have a machine it's the size of a fricken warehouse it's completely different technology and 30% as reliable. And they literally have 1. Just... Do the bare minimum research. Meanwhile our memory manufacturing facilities house 10-100 higher precision higher yield EUV with much greater capabilities. Cxmt is not even pursuing HBM. Do research dude. Easiest sector to gain some knowledge on.
You think CXMT can't get EUV machines in 2 years? You are not realistic.