HN comments for: Japan bets $67B to become a global chip powerhouse once again

China bet almost three times as much. Rhetoric in the West is starting to change as they make progress. Used to be that people would laugh and say it’ll never happen. Now there’s talk of “what we’re going to do when they flood the market with cheap silicon”. All thanks to largely self-defeating sanctions. Same thing will happen with AI acceleration hardware. In fact it’s already happening.

They were always going to try to in-house chip making - it's a huge point of geopolitical risk for them. Sanctions were not self-defeating, if anything, they were effective in their goals and slowed China down.

Sanctions on letting them buy chips 1. reduced available R&D money for US competitors by reducing their revenue and 2. raised the local price to monopoly prices, creating a great reward for the first local manufacturer to achieve parity.

If you want to see effective de-industrialization, look at what happens to US industry when China applies the opposite of sanctions (subsidies).

Adam Smith figured out that mercantilism is inherently self-defeating over two hundred years ago. Maybe nation-states will catch up in the next few centuries.

China is still a few decades from commercially viable high end semiconductors. Yes, they can subsidize poor yields, but it doesn’t really bring them closer to figuring out the problem (they basically need to experiment a lot, and that will take time even if the money is there). China was always going to do this, they announced they were going to pour billions into semiconductors and modern jet turbines more than a decade ago. The trick is ensuring the money isn’t siphoned off to corruption.

Out of curiosity what’s stopping them cloning existing processes and technology like they do for everything else?

The processes being hard to just clone, involving a lot of detailed expertise and being extremely sensitive to minute differences.

The West is able to lean on existing knowledge Intel has from running its own fabs in the US and the knowledge of ASML in the Netherlands who makes the machines.

Basically, they could get a machine and take it apart and figure out how it works, but that's meaningless if they don't understand how to make the stuff or how to keep it working well.

You don't think China already has folks working in TSMC fabs stealing IP?

If you believe that every Taiwanese engineer is rabidly anti-China then I have a vaccine I want to sell you.

It is my belief that China will steal what it can, corrupt who it can, threaten who it can and most importantly, re-invent what it is missing (they have a bigger STEM education pipeline that most Western countries combined) until they no longer have this achilles heel.

The only unknown is if there is will be hot war before they reach that inflection point.

China will steal what it can, corrupt who it can, threaten who it can and most importantly, re-invent what it is missing (they have a bigger STEM education pipeline that most Western countries combined) until they no longer have this achilles heel.

https://en.wikipedia.org/wiki/Semiconductor_Manufacturing_In...

China has done what you said since 2005. it've been 20 years. is China leading the semis industry?

And finally, they will force their citizens to be happy.

It is my belief that China will steal what it can, corrupt who it can, threaten who it can and most importantly, re-invent what it is missing

So, from 10,000ft I see what you're saying, and might even agree.

But what possesses someone to describe a people or country like this? We do the same with Russia; "other" them, like they have inherently "evil" traits. As if the US isn't above or involved in espionage.

I guess I'm confused how otherwise seemingly intelligent people can look at another country and think their people are fundamentally different. But god forbid you point that lens at the wrong group.

TSMC itself relies on lithography tooling coming from the west. I get people want the western world to be wrong here but they have the edge and China doesn’t. It’s going to take them awhile to catch up and by that point the west will have innovated in new ways. China overplayed its hand way to early and will suffer technologically for this.

The proof is in the pudding: they haven’t been able to replicate the process.

ASML tools are step 432 of 1234. There’s a whole fab built around it and a whole supplier ecosystem.

They will eventually succeed. That is obvious. High end chips have become existential in the modern world, especially now with AI, but have already been before that. When will they succeed is an open question as stealing the blueprint is absolutely not enough.

it's not just ASML but its suppliers, like Zeiss that make some of the key components of chip-making tools that are extremely difficult to replicate

1) Industrial processes are stupidly hard to duplicate.

It doesn't even have to be something as advanced as semiconductors, duplicating a line that manufactures something as prosaic as paint is an exercise in patience and frustration. It's like building software from source using documentation, you may think you wrote down all the steps, but you find all the crap you missed when you actually try to execute the task.

2) Top-down command is anathema to engineering progress

American corporations are bad enough for only reporting good news up the chain (see: Intel and deep UV lithography). A dictatorship like China is going to be ridiculous. You will deliver good news to Dear Leader, and you will deliver it when expected, or you will find yourself in the doghouse. So, especially if your task fails, you will make bloody sure that it either gets reported as successful or that you leave someone else hanging with the consequences (see: China and water and rocket fuel). This slows the engineering process to a crawl as nobody can trust anything delivered from anybody else.

What’s the difference between American Corporations and International Dictators? Serious question. I’ve always been under the assumption they operate similarly.

As an employee of a corporation, I often speak my mind. The most I risk is being fired, and I believe I can find another job relatively easy. I sometimes do keep my mouth shut, because saying the truth would get me some sour looks, and I don't feel like dealing with that right now.

I lived for a few years behind the iron curtain, in an actual honest to god dictatorship. The fear pervading and perverting society was palpable. Say the wrong thing at the wrong time and you might die, and your family will suffer too. Not to mention destroying your career and social status. Best to lie most of the time, even to realtives. You never know who to trust, and the price of a mistake is enormous.

From the way you phrased your question, I get the feeling you have no idea what a dictatorship is like. The two are only very superficially similar. Look at how Navalny ended up vs Ilya Sutskever.

Well there are a lot more corporations than dictators, and usually shit doesn't go sideways when a corporation falls.

Corporations usually can't just murder people in the streets if they don't like them.

A not insubstantial part of China's economy involves private corporations that compete with each other. The image of China as a command economy hasn't been true since Deng's reforms of the early 1980's. At the higher levels, sure, a lot of executives will be beholden to the CCP, but there are few places China places a single bet on a single state venture. Even their fully government-owned ventures are often companies that compete with each other or have corporate subsidiaries that compete with each other, and so while you might be tempted to just deliver good news, you face the risk that your competitors will deliver better news backed with results.

This is not an attempt to suggest China's government is good, because it's not, but it's also not a carbon copy of the worst sides of the Soviet Union - for all of Deng's brutal authoritarianism, he did recognise and address a lot of the worst mistakes of Mao and the Soviets in terms of the economy.

It also still doesn't necessarily mean employees in a private company in China will be as open to reporting issues as they might have been elsewhere, but it does mean there are incentives in play that at least make many executives want to put effort into identifying and addressing issues.

The tooling is ultra specialized and requires whole teams of highly skilled techs to assemble and calibrate. That's like half the cost involved, and the crews running the lines don't know much about how it works, they don't need to. If something goes wrong, they have service contracts.

So China can glean some of the process, but, because the design, assembly, and setup of the machines is done by western contractors, they're effectively locked out of 'the hard part'. Not to say they can't learn it, it's just not nearly as trivial as the rest of the things they clone.

Also worth noting that, many of the chinese clones are made in the same plants as the orignals, sometimes in parallel to the originals as to benefit from existing successful processes that were already in place, but using cheaper materials, less QC, different branding, etc.

It's actually quite impressive engineering really, they can often reduce part count and cost significantly while shipping a functional, albeit inferior, product. Chip fabs don't work in that model though, so, that presents an additional difficulty.

Reading Chris Miller's Chip War allowed me to realise how mind-bogglingly complicated the chip making process is.

China is still a few decades from commercially viable high end semiconductors.

I'm going to set a reminder to check in on this comment in three years to see how it has aged. My guess is, very poorly.

I heard that five and ten years ago so I’m not really worried (or if you mean china will always be three years away from commercially viable high end chips?).

2. raised the local price to monopoly prices, creating a great reward for the first local manufacturer to achieve parity.

The government was both going to demand, and fund local companies creating a competing product and was already doing so long before sanctions arrived. Nothing about the sanctions sped up the process in the least. The Western sanctions had almost 0 bearing on accelerating local chip production, but it did do a GREAT job of limiting their ability to buy the hardware required for advanced fabrication.

You also seem to have a belief that a "monopoly" in China is anything remotely like the West. It's not a free market, if a company in China tries to gouge the market or take it in a direction that's not approved at the expense of what the party wants, their executive leadership will quickly find themselves on the outside looking in (if they're lucky) or sitting in a jail cell if they aren't. See: Jack Ma

Semiconductors were core to the "Made in China" strategy the party was pushing in 2014...

https://www.mckinsey.com/industries/semiconductors/our-insig...

I have some good news - you suggest that the Party will step in if there is gouging or an independent direction being taken - that is to the advantage of China's competition.

One of the reasons groups like the USSR or Serious Communist China ended with everyone starving is that there is literally no known method to determine the correct price without mass signalling through a free market. Ditto what direction "should" be taken - the only known method is 10s of companies (ideally with funding weighted by what clever investors think is likely to work) where 9s of them fail and a selection succeed.

China isn't getting where they are because the party is helping. If anything it is the opposite, the party must have some capitalist incubators where they are holding themselves back from interfering all that much, otherwise the whole thing will collapse at some point because some bureaucrat refuses to believe prices were set too low until the shortages starve millions to death.

Serious Communist China ended with everyone starving

This is certainly a claim, let's see how it stands up against the historical record (as from wiki): 1961: Last famine in the PRC 1966: Start of the Cultural Revolution 1976: Mao dies

I'm having a bit of trouble reconciling these, but if you've got some arguments explaining why the Cultural Revolution was after the end of "Serious Communist China" I'd be interested.

I agree. Communes were only privatised in the early 1980s, and I think that is a better marker for the end of communism in China.

https://en.wikipedia.org/wiki/Household_responsibility_syste...

One of the reasons groups like the USSR

The 'USSR' starved in the 90s after the fall of communism.

Serious Communist China ended with everyone starving

Oh this nonsense. Serious Communist China had to deal with nuclear threats from both the US and the USSR. Serious communist china tok a back water colonized nation to a nuclear power within a single generation.

there is literally no known method to determine the correct price without mass signalling through a free market.

You have it backwards. 'Free markets' led to 'starvation'. That's why we, in the US, have farm subsidies. A form of 'price control' along with many other measures. There is no nation ( especially a major ) on earth that has a 'free market'. There was also never a time when we had 'free trade' either. I'll let you figure that one out by yourself.

Every industry, from housing to oil, has price controls on them. It's all a matter of the degree and levers used.

The same idiots who praise government subsidies to build up our own chip industry are whining about 'price controls'.

The government was both going to demand, and fund local companies creating a competing product and was already doing so long before sanctions arrived. Nothing about the sanctions sped up the process in the least.

Whatever encouragement or pressure the Chinese government was applying was insufficient to get manufacturers like Huawei to switch to domestic suppliers. Chinese companies still worked with established companies like TSMC, until forced by US sanctions to switch to worse domestic alternatives.

If sanctions are so effective the CCP could just do the same thing to themselves via a trade ban on imported goods.

CCP could just do the same thing to themselves via a trade ban on imported goods.

They could, but a domestic trade sanction, just like foreign sanctions, would guarantee at least near-term damage to Chinese companies.

Domestically, Beijing would be seen as the responsible party for the resulting economic damage, causing resentment towards the government, with no actual guarantee of self sufficiency. Companies, in general, do not care about having a domestic supply chain; they want to increase profits.

Internationally, other countries would also see China as the initiator of the conflict, and those that are negatively affected may want to respond with sanctions of their own, escalating it beyond what Beijing intended.

The risk analysis is different between foreign sanctions and domestic sanctions. In the end, the Chinese government was not prepared to take that risk, but the American government was.

See, policy like this I wish we had in the U.S., where the government controls industry, not the other way around.

The inherently parasitic, extractive elite in every empire so far loved mercantilism though, it appeals to the centralization focused, extractive, landlording mindset. China is trying to be as mercantilistic as possible (buy raw goods everywhere, ship only final products) and has none of the free trade idealism of the us. Which is one of the reasons the belt and road initivative is already collapsing.

I'm starting to think the US has little free trade idealism left.

The empire, the external face-hugger to any country, eats it all after a while, turning it into giant estates ala rome, with slaves and no middle class - with hired mercenaries for troops.

In fairness, a lack of sanctions an also accelerate the industrial progress of rivals.

Nazi Germany and the Soviet Union had close partnerships with Detroit auto makers that helped their industry.

I am not seeing it. Either I am blind or we are seeing two different things? China sky-rocketed since the pandemic. They had a rough lock-down but there is something strange brewing there: Their tech/quality sector is releasing stuff at record speed (cars, phones, chips, etc..), their exports have boomed and even in this recent downturn seems to have stabilized, their "green" transition is happening at an unbelievable speed (look for China solar installation for last year), they continue to pump highways, HS rail, metros, airports at mind-boggling quantities, etc...

I am afraid the train left the station, or more like the rocket has launched already. I don't think the sanctions are going to make that big of an effect and if anything, they might come to bite the US back than anything else.

The infrastructure spending is indeed mind boggling, but I have the feeling similar to their real estate glut that they are building too low quality and are making a grave mistake in this because they will be stuck with this infrastructure for the next decades (comparable with how the US is stuck with its aging airports, highways and congress centers).

The public infrastructure (especially the newer one) is actually quite decent even by Western standards. The private buildings are not, at least the old ones (actually they are quite bad). The Chinese government recently banned high-rises because of this. Many of these buildings are now a liability to the public.

The sanctions are making a huge impact - chip making is more complex than almost any other current industry. I don't think sanctions will stop them from achieving their goal, but it will push them back from their goal of independence by about a decade.

Who knows if it comes back to bite the US. If I were China, I'd remember how the US treated them. I don't know.

I agree they were always going to try it, but the point is prioritization. All the sanctions did is bump it up on the prioritization list alot. So for instance in 2011 we passed an act banning China from the ISS. The argument was that they had nothing to offer, and implicitly that they'd just steal everything. 10 years later they've created, launched and manned their own space station, have carried out the only sample return from the Moon since 1972, and in general have one of the most advanced space programs on the planet.

Had we continued to collaborate with China in space, it's likely that most of these developments would have happened at a much slower pace, simply because they'd have had a lower overall priority. Sanctions, in general, are just a very poor weapon that increasingly obviously seem at least as likely to hurt the sanctioner as the sanctioned.

do you know what caused golden space age in 60's ??? because American don't want lose to USSR

the world need strong rival like China to challenge western hegemony. that's why we have Artemis now, if China still in backwater like in 60's, we would not have project Artemis

Had we continued to collaborate

And the West would have had access to the same tech. The issue with putting a wall between yourself and the biggest manufacturing hub on the planet is that the well manufactured stuff might not be on your side of the wall.

they were effective in their goals and slowed China down

I believe that, unfortunately, the US has a much lesser grip on the global economy nowadays. So, these kinds of sanctions only slow them down (the US), while other economies benefit immensely, even China itself.

i.e. there will always be a buyer for a billion surplus chips in the market.

They always wanted to, but Chinese companies didn't cooperate because market incentives were lacking. Chinese fabs wanted to buy the best international equipment and shunned Chinese toolmakers. Chinese toolmakers never had a chance to break through to large market share. Until now: the sanctions gave toolmakers a captive market.

Chinese toolmakers sucked because they didn't have market share and lacked practice. With a captive market, they finally have customers to practice on. This learning-by-doing cycle is what allows them to rapidly improve their quality, something they couldn't do before.

The sanctions hurt short-term, but also gave them a long-term opportunity which they didn't have before.

Sanctions were not self-defeating, if anything, they were effective in their goals and slowed China down.

What goal is that again? China is down to 5-7nm chips in practically no time which is more than enough for virtually every industrial and military application (the latter being the ostensible target of sanctions).

Huawei is even shipping phones now again on domestic supply chains. Pretty much the only thing the sanction have achieved is what, that Chinese gamers can't play Cyberpunk 2077 at max graphic settings? The irony is that cutting edge chips are predominantly used in sectors that China itself regulates to hell already. If you're concerned with China's generative AI advances you could have spared yourself the sanctions because so is the Chinese government

It seems quite early to evaluate if they are making any progress, if the goal is to catch up to, say, ASML.

They’re at 7nm now. Working towards 5nm. Yields leave much to be desired, but in the absence of other options cost doesn’t really matter. They’ll just have to make more wafers.

Where did you get the 7nm from?

Huawei apparently is making 7nm chips for their Mate 60 phone, but it’s apparently terrible, overheating issues, slow, high power consumption.

I think the power issue is because the camera and mic are on 24/7, they showed QR codes being read while the screen is off.

Gotta start somewhere. Everyone sucks at some process until they don't.

Sucking at a process is part of learning to do it better. But only if it is on a path of learning; practicing the skills that will be useful later.

They’ve stuck with DUV, which everybody else in the world decided was a dead end. Sticking with DUV may cause further progress to be impossible for all we know (maybe somebody should ask Intel). Going to EUV will require a different set of skills. They’ll have to replicate ASML’s machines to do so. Nobody else has figured it out. Their process on figuring out DUV tricks doesn’t tell us anything about their progress making that other type of machine.

Their process on figuring out DUV tricks doesn’t tell us anything about their progress making that other type of machine.

Yet you have no problem proclaiming it's leading them to a dead end.

It is a dead end path. They probably are exploring multiple paths in parallel. ein0p claimed that their progress down the dead-end path is evidence that they are making good progress down the productive path. It isn’t. That’s all I’m pointing out.

Don’t read more into the post than that. I’m not going to defend an affirmative claim that they aren’t making progress. They have a ton of engineers and are willing to invest in the field, so it is entirely plausible to believe that they’ll catch up.

I'm laughing at your comment because you seem to assume that China is doing nothing in EUV. I have somewhat inside knowledge and I can tell you that they have multiple teams working on multiple approaches.From SSMB to LPP, they are leaving nothing to chance.

I have been repeatedly pointing out that we don’t know, not claiming that they aren’t making progress. Do you deliberately misread your insider information as well?

Also worth pointing out the Overton Window has shifted from "China can't silicon." to "China's doing crappy 7nm." in a span of only a few years.

The copium levels in the atmosphere here are reaching critical levels.

That is just not true. The Huawei Mate 60 is selling about as much as the iPhone in China atm - it really is a breakthrough product.

Google “SMIC 7nm”.

Ultimately chokepoint still around creating litho/entire semi supply chain. which TBF is coming along. Big Fund was wasetful but did produce a lot of uncoordinated domestic players, US sanctions now forcing them to coordinate. PRC probably only actor with entire indigenous semi supply chain on the horizon. Fabbing leading edge chips was always the "easy" part, there's a reason TSMC/Samsung were willing to open fabs on mainland before US sanctions/CHIPs act without massive subsidies. Other east asians know Chinese work culture is compatible, they can work how they work and make what they make. Hence SMIC/Huawei can do 7nm and eventually 5nm with some TW/SKR training in years prior.

Semi is probably going to be less like oil and more like agriculture in coming years, it's just something strategic many govs have to subsidize.

The 7 nm uses (older) ASLM machines.

“Rhetoric”

There’s no rhetoric. When you phrase things this way it just makes you sound like an anti-USA pro-China troll.

If I was a pro-China troll I’d be supporting these sanctions. Removing the sanctions is the only thing that could strangle their nascent advanced lithography industry in the crib. US government, however, is not smart enough to see it.

That's assuming what you say is true (well it's unfalsifiable in foresight) and you are not employing some reverse psychology here. Incidentally I find that pro-China accounts or forums use similar arguments to profusely support removing sanctions.

It’s also quite literally true. There’s a reason there’s only one ASML. That reason is free trade - it makes it untenable to invest hundreds of billions into developing alternatives - you can just buy state of the art instead and move on to higher value things. Now we’ve forced China to (eventually) build “their own ASML”. Which they will eventually succeed at, given the strategic importance of this tech. Oops.

It didn't cost hundreds if billions to build EUV.

Overall, EUV lithography machines needed more than $6.3B in R&D investments over 17 years to come to market

Japan could do it too (GigaPhoton had viable light source, but no customer).

Even if they were only 10% as efficient in R&D, it's peanuts to Chinese government.

EUV is being talked as some super mega project, but this is basically peanuts. Roughly 80 F-35. Even if rhey are far less efficient i wouldn't count on it for long.

Sure, it's legacy SWE for now, vut even that us better than all countries save 3-4.

Current sanctions are bust, it's time to comtemplate blank ban on chinese made chip.

https://www.reuters.com/technology/chinese-chip-equipment-ma...

It kind of did though. You’re talking about incremental cost, building on top of what’s already there. And what’s there was already quite formidable

Blanket ban is also not going to work on account of the West only representing 1/8th of the world’s population.

That horse left the barn. That's like EU going back to Russian gas, because it's cheaper. Not going to happen.

Except that Canon released a 5nm node thats market ready about 4 months ago:

https://news.ycombinator.com/item?id=37894925#37896319

So I'd wager that this investment is to bring technologies such as this up to scale.

This technology is useful for small production volumues only though. It's too slow to produce chips at 5nm at high volumes. If it's also that much cheaper such that you can just throw money at it to produce in parallel, then it may work, but it I haven't seen such comparisons yet.

I don't follow what you are trying to say.

But do-it, do-it-cheap, do-it-at-scale, do-it-right has been the standard playbook out of Asia for 50 years now. I'd suggest that Canon is going down the road of being a player in the 5nm market.

You could summarize it as "Wah Wah, competing technology not as good as trillion dollar R&D from competitor with 20 year market advantage".

Even if it has a lower yield, having a competing technology to EUV is absolutely amazing in several ways.

He is trying to say that until they succesfully move to the "do-it-at-scale, do-it-right" parts, we're at the do-it part atm, and that's no real competition to those doing it at scale and right already.

So the playbook atm is more promise than concrete reality.

Nor is there some law that says "they did it for other technologies, so they'll manage it for this one too" (besides, this is survivorship bias, forgetting about the technologies they didn't manage to do it).

Sorry but do you have a source for this?

Everything I have read/seen not only points to the Chinese being decades behind but their efforts being black holes of money time and effort that yield NO progress.

Huawei design division was on par with apple. Ymtc was first in the market with 200+ layer NAND.

Their domestic share of semi equipment machines with publuc tenders (=legacy) has gone through the roof and is over 60% (was 10? I think five years ago). These are legacy, so unsanctioned tools.

There is a lot of waste, but there are results.

Huawei design division was on par with apple.

https://www.techinsights.com/blog/techinsights-finds-smic-7n...

7nm and 3nm are not even in the same ballpark.

And the legacy tech is useful, but this isnt "progress" its production. The distinction is large, production did not sustain declining producers.

There was a rumour at one point of a 1 T RMB fund for Chinese semiconductors but I don't think that came to be. However it appears that a number of large Chinese firms are getting unlimited state support for research in highend semiconductor production not relying on a US controlled supply chain.

I recommend this piece for anyone who is interested in the current state of the Chinese semi industry:

New Era for the Chinese Semiconductor Industry: Beijing Responds to Export Controls by Paul Triolo

https://americanaffairsjournal.org/2024/02/a-new-era-for-the...

Why are the sanctions self-defeating? I don't remember anyone saying China could never do the same thing domestically. They are a powerhouse of industrial espionage after all. Doesn't mean the west has to make it easier for them.

I have no problem with that. Look at how cheap photovoltaics became, and how it became win win overall

It's not going to change much. Most ICs don't need or benefit from 2nm features. It's only value in the West is in boosting yield of high profit margin VLSI devices where it makes sense to air freight them to SZ for assembly. Those represent a tiny slice of the semiconductor market. Cheap jellybean semiconductors will stay in SEA close to their final assembly point since their dice are already tiny and you can't squeeze more parts out of a wafer without compromising on specs.

I understand globalization for most run of the mill items but I can’t believe we’ve let the high tech manufacturing go offshore.

and it's not going super well for them: https://en.wikipedia.org/wiki/China_Integrated_Circuit_Indus...

"looking to mass produce state-of-the art 2 nanometer logic chips in 2027 from an initial starting point of zero"

They have bold goals

Historically, Canon and Nikon were some of the leading suppliers of photolithography equipment. However they seem to have missed the boat on EUV tech. Recently Canon is trying to make a comeback [1].

[1] https://arstechnica.com/reviews/2024/01/canon-plans-to-disru...

ASML took 20 years… even if they can do it twice as fast (not impossible but I have my reservations) it still means a fab is operational in about 2035-2040… better late than never I guess?

Why just twice as fast? Once you know that something is doable, it works and the rough direction things can get done really really fast. The Bannister effect is a real thing that can be observed in so many areas.

Chinese leaped ahead once the US banned exports to certain Chinese companies, Turks have become top players in the military drone business after the West didn't sell them the drones.

I mean, obviously there's lots of work to be done but I think people who believe that others can’t do things they can do are very misguided. Laws of physics is the same everywhere and scientific research is quite open, engineers are everywhere and it only takes someone to invest resources to do something.

When you look at what it takes to successfully manufacture chips at these feature sizes you start to believe in the anthropic principle, or maybe I should call it semiconductoric. Veterans in the field can hardly believe its at all possible.

Why just twice as fast? Once you know that something is doable, it works and the rough direction things can get done really really fast.

I agree on the first part. The second part is full of details. The mirrors have to be just right. The lasers have to be just right. The timings have to be just right. There's a million other things that have to be just right and the path to get there may not be very parallelizable.

Sure its still hard but unlike those who did it first, those coming after don't have to guess as much and spend resources on dead ends.

This is not Taliban getting into semiconductors, this is Japan who is already into semiconductors but lost its edge in this particular field but its leading in some other fields. Therefore they already have a considerable know how on mirrors and laser that they will have to improve on. IMHO the catch up will be much quicker and much cheaper than what the Europeans and Americans spent on to explore the path leading to what we have today.

OTOH you have Intel which never stopped R&D and they still have lots of trouble catching up; arguably they've been trying for almost a decade now, five years if you're being generous. (And they buy tools from ASML! Which other posters explain will also be the case here.)

IMHO Intel's case is more nuanced, they were milking their position until Apple did something extraordinary with TSMC. They are also a giant corporation who has to re-organise and restructure, therefore can't move as fast as a newcomer. Intel's case is a business case, more than an engineering case. Just as Nokia.

The point still stands: they have the infrastructure, the experience, the logistics and the supply chain all secured and they're still at least a couple years behind if not more. It's extremely optimistic to say that it's going to be easier to start from scratch.

Could be, even engineering projects with well defined requirements and a mature tech sometimes fail.

they were milking their position until Apple did something extraordinary with TSMC

I mean, that's true from a business perspective.

But they were trying very hard to make the next process node work. They went from a process upgrade every other design generation (the famous "tick-tock") to just tock-ing at 10nm because they couldn't get the next process generation working.

That was very much an engineering problem.

Intel was receiving continuous beatings from AMD (Ryzen/Epyc) a few years before Apple came up with M1. 2016/2017 vs 2020. I don't think Apple and Intel are in much competition: Intel doesn't make laptops and lacked Apple's moat when they did, and Apple doesn't sell (amd64) server CPUs.

Sure its still hard but unlike those who did it first, those coming after don't have to guess as much and spend resources on dead ends.

In many domains they do. Once the original creators are retiring or not involved, there are tons of tacit knowledge at every step of the way, from the high level abstract design to specific quirks of some manufacturing equipment, that's very hard to build back.

Catching up is usually easier than pushing the frontier.Once you know that it's possible, it becomes a matter of iteration.

So it will take less than sixty years. Yes, I can believe that.

There's a lot of FUD around the subject too. Take naysayers with the same degree of skepticism you take the optimistic folk.

Judging from the actual industry state, and how many have been left behind in that domain, I'd rather believe the naysayers.

Until now it's just an announcement and some money thrown at the problem, so less than vapor.

Turks have become top players in the military drone business after the West didn't sell them the drones.

Everything is cheaper in Turkey.

They can produce a lower quality drone for 1/10th the price. Bayraktar = $5M, Reaper = $32M.

Of course there's a market!

I doubt the US could produce a drone the quality of Bayraktar for $5M domestically. It's just too expensive here.

The Turks aren't succeeding because they're building better drones. They're succeeding because the market for drones is disposable, and so the quality doesn't need to be as high - in many cases it's better to have a lower price.

China didn't succeed because they can make better T-shirts than the US. They succeeded because they made cheaper T-shirts.

Don't confuse the two.

China didn't succeed because they can make better T-shirts than the US. They succeeded because they made cheaper T-shirts.

Can China make the same quality T-shirts as the US, but at a cheaper price? Then it could be considered better. For most things involving ordinary people, what matters is not what the highest possible quality thing is given no price constraints; what matters is what is economically feasible.

The largest company in the world - Apple - would beg to differ.

People care about quality in many profitable very large industries: health care, technology, military, automotives, aircraft, etc.

Do you really want to fly on the cheapest airplane money can buy?

No.

Turkey has a range of drones, the cheapo Bayraktar one is just the most well known.

Some of the Turkish drones are better off not to be well known though, as there are accusation of launching AI controlled attacks and the machines were in charge of picking the humans they killed in Libya - which is a no-no but probably developed by all the leading states.

Here is a link if you want to take a look: https://www.voanews.com/a/africa_possible-first-use-ai-armed...

So no, Turkey isn't in the business of making cheap low quality drones, the range is quite extensive with some models at the cutting edge. All thanks to being in need of drones but not being able to purchase even if they have the resources.

Bayraktar TB-2 versus Reaper:

- 55kg payload versus 1746kg payload

- 24k feet versus 50k feet altitude

- 150km versus 1850km range

It's not a fair comparison. They're very different classes of devices.

I do think the US could produce a Bayraktar-quality drone for under $5M. The major upside of the Bayraktar is that non-NATO countries can buy it.

Heck, I think many readers of HN could do the same (which obviously wouldn't help with sanctions). Stick a control system on a basic Cessna 152, and you've pretty much got the same capability. That's well south of $1M even.

You don't hear much about Bayraktar TB-2 (specifically) anymore since it's also not very useful anymore. It worked well before Russia was prepared. Now, we're generations ahead. Sky is full of drones, anti-drone warfare, and largely ones the ones used are a fraction of the cost of the Bayraktar TB-2 so it doesn't matter if they're shot down. The other end of the scale is ones less easy to shoot down.

edit: Corrected to write about the specific famed drone useful at the beginning of hostilities in Ukraine, versus Turkish drones in general.

> Chinese leaped ahead once the US banned exports to certain Chinese companies, <<

hate to nitpick, but this was actually well underway before the US sanction kicked in. SMIC's 7/5nm was already completed by former head of R&D at TSMC in 2020, which was more or less a replica of TSMC's 7nm. It certainly didn't help that the US dept of Commerce allowed over $100+B worth of export licenses to China, 1/3 of them to "blacklisted" entities, in 2022. The Dutch gov't export restriction likewise didn't really start until few months ago, which also allowed latest DUVs capable of 5/7nm.

ASML are also involved:

Dutch equipment manufacturer ASML Holding announced they would also open an office in Chitose with 40-50 technical staff to assist Rapidus. While full operation is not expected until 2027, a pilot line is expected to start in 2025.

https://en.wikipedia.org/wiki/Rapidus#cite_note-24

However they seem to have missed the boat on EUV tech.

Licensing:

To address the challenge of EUV lithography, researchers at Lawrence Livermore National Laboratory, Lawrence Berkeley National Laboratory, and Sandia National Laboratories were funded in the 1990s to perform basic research into the technical obstacles. The results of this successful effort were disseminated via a public/private partnership Cooperative R&D Agreement (CRADA) with the invention and rights wholly owned by the US government, but licensed and distributed under approval by DOE and Congress.[8] The CRADA consisted of a consortium of private companies and the Labs, manifested as an entity called the Extreme Ultraviolet Limited Liability Company (EUV LLC).[9]

Intel, Canon, and Nikon (leaders in the field at the time), as well as the Dutch company ASML and Silicon Valley Group (SVG) all sought licensing. Congress denied the Japanese companies the necessary permission as they were perceived as strong technical competitors at the time, and should not benefit from taxpayer-funded research at the expense of American companies.[10] In 2001 SVG was acquired by ASML, leaving ASML as the sole benefactor of the critical technology.[11]

[…]

This made the once small company ASML the world leader in the production of scanners and monopolist in this cutting edge technology and resulted in a record turnover of 18.6 billion € in 2021, dwarfing their competitors Canon and Nikon who were denied IP access. […]

* https://en.wikipedia.org/wiki/Extreme_ultraviolet_lithograph...

so they decided foreign technical competitors should not benefit from US taxpayer-funded research (makes sense) but then allowed ASML to buy SVG and thereby monopolizing (quite literally) this most critical of technologies? idiots.

Did Intel get the license?

Intel's focus was chip design and manufacturing, not tooling; it took a long time (and money) for ASML to develop its cutting-edge EUV tech. Besides SVG, ASML also bought other key US EUV component developers like Cymer (lasers) and HMI.

I thought the US favored companies over individuals, but evidently this isn't limited to US based corps.

The first part doesn’t make sense at all. A huge percentage of american taxpayer funded research is released openly to the world. It only makes sense in using it to create the latter monopoly, though, I guess showing where incentives really lie.

First I've heard of this, that's absolutely awful.

Why?

Do you think monopolies are generally bad or generally good?

Why are you asking me?

The state has a monopoly on violence, is that generally bad?

Yep!

Specifically, Antitrust (as is mentioned).

Canon and Nikon were leaders in lithography at the time, and ASML/Philips were a laggard 3rd.

ASML productivized (in a two decade process with many partners) foundational research licensed from US national labs, research which the Japanese (Nikon & Canon) were barred from licensing due to their then-dominant position. Though it seems unlikely to me that Canon would've gone for it, considering that they never got into ArF lithography.

I would love to know more about the inside history and origin story of ASML. Does anyone know of any good podcasts or blogs/interviews covering some of this?

'Why The World Relies On ASML For Machines That Print Chips' an 18 min CNBC piece is quite jolly https://www.youtube.com/watch?v=iSVHp6CAyQ8

Update - I just watched it through and it's actually pretty interesting. A bit on the history, then how the tech works, then on the geopolitics of not supplying China with EUV and the like. Interviews the CEO and the Chip Wars guy.

The tech's quite cool in the latest one with 13nm UV. It's absorbed by all known substances so has to operate in a vacuum and the light is produced by them squirting liquid tin at 4000 psi and then zapping the droplets with a 30 KW laser. The machine takes three fully loaded 747s to ship.

I think this stuff may be a bit of a bottle neck with Sam Altman's proposal to invest $7tr in chip making - they are the only people making the state of the art machines and are pretty much flat out already.

It is covered in Chip War, by Chris Miller. The book also details the loss of leadership by Nikon and Canon.

There's a recent article in MIT Technology Review about it (from the author of Chip Wars)

Check out Asianometry on Youtube

Both Canon and Nikon were sleeping at the wheel when GoPro and DJI both dominated the action/sports camera and drone business.

And yet it was Sony that hurt their camera market shares the most, action cams are a nieche product.

None of those have anything to do with chip making so.

And yet it was Sony that hurt their camera market shares the most, action cams are a nieche product.

Mirrorless/DSLRs and compacts have been further niche-ified by mobile phones. Many who previously would step up from compacts to a Mirrorless/DSLR have stopped and just use their (now vastly improved) phone cameras.

And mirrorless and DSLRs are maybe more of a niche than action cameras (which tons of average Joes buy, not just extreme sports fans).

Canon and Nikon make more than half of their revenue from industrial or healthcare products. They sure wished for more profit on the customer business as well, but that's not their core competency anymore.

Rapidus buys ASML machines, because it's the only available option and developing national machine isn't their focus

of all the countries in the world capable of pulling something like that off. Japan is right up there.

Unfortunately very unlikely. What's more probable is they'll end up with someone taking the blame / escaping the madness through suicide because the miracle was not delivered on time. I'm rooting for them, but without even a working sub-10nm fab there ... well, it's the uphill battle of uphill battles. Are they willing to ask for outside guidance and accept it? Getting training from IBM seems like getting training from an old gas station clerk who used to work in the big fancy factory. (And TSMC is already building fabs, but that just means they will compete with them for labor, and those 100 veterans will just go and work there.)

Luring legacy operations, sure, why not, but ... it doesn't make much sense. It's race to the bottom. If they want to do geopolitically useful thing they should put down their feet and spend on defense of Taiwan.

"If chip supplies from Taiwan halt, ..." yes, economies will collapse. Chips will be the least of our problems. Unless they mean chips for smart bombs.

https://en.wikipedia.org/wiki/List_of_semiconductor_fabricat...

Getting training from IBM seems like getting training from an old gas station clerk who used to work in the big fancy factory.

IBM still does a lot of research in chip manufacturing, they revealed a 2nm chip 3 years ago. Arguably they don't do large scale production, and the 2nm isn't mass production technology, but they aren't exactly at the retired gas station clerk level. Honestly they might be a good choice, because who else? Why would TSMC, GlobalFoundries and Intel help a potential competitor?

Honestly, for Intel, it makes a heck of a lot of sense to get in on this game. It's just a question of getting the right deal.

I really doubt they can compete with TSMC, Samsung, and GlobalFoundaries alone.

Hahahahaha...

Oh god, they're serious.

To catch up to TSMC would take decades at this point, dumping endless money into it.

I'm not saying it can't be done, but it absolutely cannot be done in 3 years.

In 2027 "2 nanometer" will be like 2 generations behind the state of the art... At least if you believe current Intel and TSMC roadmaps.

The human factor is one of the reasons Taiwan is as good in chip manufacturing as it is.

Working in a chip factory is plain terrible for humans: Having to do anything in a clean room is exhausting. Even young people in Taiwan are sceptical about the working conditions:

https://www.thinkchina.sg/taiwan-lacks-young-passionate-work...

Japanese work culture is famous for glorifying the ability to suffer for your job, and Shintoism even puts a spiritual component to „cleanness“.

It moved to Taiwan (and SK) from Japan because Japanese graduated from that kind of work culture.

The work culture in Taiwan (and SK) is miles worse than Japan today - which makes sense as both are around 10-20 years behind Japan

Median household income in Japan is around $48k [0], around $33k in SK [1], and Taiwan's I can't find but is around $28-34k based on median wage ($16k) [2]

SK+TW have entered a similar doldrums to JP around 15-20 years ago, and just like how JP began outsourcing manufacturing to China, ASEAN (mostly Thailand), and India in the 2000s, SK+TW have started doing the same (Vietnam for SK, Taiwan hasn't found a major ASEAN partner yet)

[0] - https://www.stat.go.jp/english/data/kakei/156n.html

[1] - https://m-en.yna.co.kr/view/AEN20231123001600320

[2] - https://focustaiwan.tw/business/202312010011

Taiwan is well ahead of Japan and SK in terms of GDP PPP, which takes into account the cost of living in the country. In fact, Taiwan is closing in on the US as well.[0]

If we base it on GDP PPP, then Japan and SK are actually 10-20 years behind Taiwan.

The work moved from Japan to Taiwan simply because TSMC beat the competition out.

[0]https://en.wikipedia.org/wiki/List_of_countries_by_GDP_(PPP)...

PPP and anything Per Capita isn't a convincing metric.

They are productivity metrics (aka how much shit a country can produce)

GDP is just a measurement of how much shit a country produces a year.

GDP per Capita is a measurement of how much shit is produced per entire population - from a 5 month old toddler to a 92 year old geriatric.

GDP PPP is a lossy transformation of GDP into supposed "international dollars" that have varying calculations and are based on the (imo flawed) assumption that goods can be substituted.

GDP PPP Per Capita is dividing that flawed transformation across the entire population.

because TSMC beat the competition out

Because American and Japanese companies in the 2000s viewed the foundry model as commodified and a race to the bottom, like battery tech (eg. BYD), and as such decided to divest and let much poorer/less well compensated Koreans and Taiwanese work on it in the 2000s.

This is why GlobalFoundries exists. AMD was falling drastically behind Intel in the 2000s because their foundries were a massive operational expense preventing them from investing in R&D.

As such, AMD divested their fabs into a PE funded company called GlobalFoundaries in 2009

Japanese companies did the same thing, creating joint ventures or gaining controlling stakes in Chinese, Taiwanese, and Korean conglomerates while transferring commodified IP.

This is how CATL was formed by Japanese TDK, BYD by Samsung's cellphone battery division (yes they're Korean but same story), and others

>GDP is just a measurement of how much shit a country produces a year.

No, it's a measure of much money goes through the economy, regardless if any "shit" is produced there. Look at Ireland, Netherlands and Luxembourg or any other tax heaven.

Services are production as well.

Or are you saying that any dollars made selling software you wrote is useless?

>Services are production as well.

In what fantasy universe?

>Or are you saying that any dollars made selling software you wrote is useless?

I never said anything about software. You're building your own strawmen at this point.

In what fantasy universe?

By definition - https://data.oecd.org/gdp/gross-domestic-product-gdp.htm#:~:....

I never said anything about software

Software, Banking, Foreign shell companies holding all global profits, etc all fall under SERVICES.

You're answering to claims I haven't made and not the questions I asked.

I'm pointing out that the definition you are using is fundamentally flawed, which completely undermines your point.

You're pointing out that tax heavens produce products which is flawed. They don't produce anything, they're just middlemen and their balooned GDP reflects that.

Generally, goods and services are considered to be products. And their analogy to software wasn't in bad faith, software is a service (hence the term SaaS used for subscription based software, even though non subscription based software is also a service imo) and not a good.

>Generally, goods and services are considered to be products.

False.

Products are products, services are services. Product si a Macbook or video game I just bough which I can return, but I can't return the cleaning service the handyman just offered on site.

median disposable income is another measure, on which the US leads the world https://en.wikipedia.org/wiki/Disposable_household_and_per_c...

And was why low margins high skill work left for ASEAN, EU, Mexico, China, Japan, South Korea, Israel, India, etc since the 1980s.

You can see references to this even in early Simpsons (eg. Germans buying the nuclear plant or Japanese being viewed as god tier overworked manufacturers).

Using raw median income is pointless when comparing work life quality. GDP PPP is simply a better measure though imperfect.

AMD sold their fabs for a variety of reasons including loss of design leadership to Intel Core designs, huge debt load from overpaying (at the time) for ATI, loss of node leadership, financial strain that almost led to full bankruptcy, trend towards fab cost needing multiple customers to profit, and a focus on chip design.

Further more, your point about chip nodes racing towards the bottom does not support why Intel chose to keep its fabs and won until they got stuck on 14nm.

It’s weird how you describe advanced chip manufacturing as a commodity when there is only one company in the world capable of producing 3nm chips at scale and make a profit. Furthermore, TSMC is worth more than AMD, Intel, and ARM combined. Up until recently, it was the most valuable company in the semiconductor industry before the Nvidia run.

The logical conclusion is that advanced chip design takes a tremendous amount of skill and is not a commodity. Arguably, chip design is more of a commodity in 2024 since many vendors are capable of producing fast CPUs, GPUs, NPUs, SoCs, network chips, etc.

GDP per capita is not a very useful metric for determining quality of work life (the point being discussed)

in terms of GDP PPP, which takes into account the cost of living in the country

You should be comparing PPP adjusted household income and not GDP if you want to make this argument.

having spent extensive amount of time in both countries for the last 10 years, this is just categorically not true.

Miles worse? Nah. At worst, on par maybe.

Having to do anything in a clean room is exhausting.

I bet it would do wonders for my allergies!

You have to isolate yourself from the CR so not to contaminate it, so I doubt that would help.

No actually it does help quite a bit. Allergies are primarily from allergens in the air and you still breathe the same air in the clean room (just through a mask). So you are going to be breathing the air minus allergens which for people with severe allergies might be worth the extra hassle of being in a cleanroom

You'd be trading allergies for new and exciting forms of clean room/PPE-related dermatitis.

I find these cultural arguments weak, and kind of a cliché when talking about Asia.

There are straightforward reasons for Japan to be attractive:

- Yen is cheap, at a 20-year low

- Developed infrastructure

- Attracts immigrant workers in search of better living standards

- Open to immigration for workers in manual labour jobs

- The wages at Taiwan (and China) are reaching parity with other countries

I've never been to Japan, but from what I've heard there is little scope for promotion in Japan's economy for immigrants, so I don't think the openness to immigration for manual labour jobs is a plus point for Japan when most immigrants may only find work in low paying gaijin-friendly industries (translator, teacher, etc) or MNCs with offices in Japan.

But then again that impression is based solely on information found on the Internet and may be incorrect.

from what I've heard there is little scope for promotion in Japan's economy for immigrants

Japan receives a steady influx of immigrants working on manual labour since the 80s - notoriously, from countries with large Japanese descendant populations like Brazil, Peru.

In the past, it was common for these workers to live frugal lives and make some savings before emigrating back. With the current Yen, today's workers go after better standards of living compared to their home countries.

Sorry if this sounds stupid, but can't they use robots for that kind of environment? Or if we aren't advanced enough for that yet

They use a lot of robots, but there are still a lot of things not automation. Robot maintenance is often a big one, though since I don't work in that environment I can't tell you what they can't do.

Automation is expensive (a multi-million outlay).

Most fabs are largely automated already, but you still need support staff to repair and QA processes, as well as scientists+engineers to R&D new processes and optimizations.

Shintoism even puts a spiritual component to „cleanness“

Isn't this typical of many (most?) religions? "Cleanliness is next to godliness" is a common idea in Christendom.

"Cleanliness is next to godliness"

Isn't that just a cliché? Christians don't see that as anywhere near the core of the religion.

Japanese work culture is famous for glorifying the ability to suffer for your job

That may be less and less relevant for younger generations, based on what I can observe.

This perhaps will do much more than the CHIPS act to remove the geopolitical risk associated with TSMC/Taiwan. A rational US administration would encourage this kind of development rather than stubbornly try to revive the sector domestically.

Both are pretty difficult models to pull off because none of the above has the edge Taiwan has, which is cheap skilled labor.

Until you can match labor cost with automation, it's DOA, regardless of how good your fabs are, and the automation isn't there yet.

Genuine question: is labor cost really that important for cutting-edge fabs? I would have assumed the tooling would be far more expensive.

Yes, but you amortize the cost of the tooling out over the lifetime of the tooling, and you're paying on a loan for that. Labor is an ongoing, and increasing cost, not just wages, but also insurance, payroll taxes, etc.

Also remember the tooling is a fixed cost, it'll cost you (many) millions in Taiwan the same as it will in Japan or the US, so the cost of skilled labor is what ultimately affects your margins.

I am doubtful as well. We are told these factories cost 10s of billions to build though the number of people on the floor doesn't seem very different than any other factory. I find that hard to believe that labour is a primary driver.

If you read a couple of articles featuring what TSMC said, then it's clear that work culture/ethic is a major reason why they have so much trouble getting their Arizona plant running. This work ethic is at least in some part related to cost: they expect long, possibly unpaid, overtime.

Are Taiwanese labour costs significantly lower than Japan these days?

Taiwan educated generations of people into this field, Japan probably has higher costs to hire in the same sector or simply doesn't have the people.

Could provide subsidies like we do with farming.

Taiwan's economy is deeply dependent on TSMC. By preventing TSMC from selling to the mainland while encouraging subsidies for TSMC's competitors in other countries the US is really squeezing Taiwan with unclear long-term consequences.

My understanding is that those subsidies would be available to TSMC for the fabs it is/would be setting up in the US.

So, basically TSMC as a corporate entity's profits/revenues can keep chugging along. Taiwan itself continues to have a risky/uncertain future.

It is potentially a lot of jobs and economic activity leaving Taiwan for the US, Europe or Japan.

Most of the US subsidies are going to Intel (including for them setting up their foundry services which directly competes with TSMC). However there are some for TSMC for setting up in the US, that is true.

Taiwan can be economically squeezed by western derisking and moving sensitive tech out of Taiwan or it can continue to be militarily squeezed by China if it continues to be the single point of failure in the western technology chain. The former appears to be the better choice than the latter.

Can't both be done? Everyone understood semiconductors will be the oil of the next wars when the covid supply chain disruption happened. The next few years have just been everyone trying to figure out what to do so they don't have to modify designs in times of war.

if japan can revive the sector domestically, surely usa can to

No a rational admin would accept that we can’t afford to ever be at risk of losing our flow of chips and subsidize as needed, just as with agriculture, to ensure we can provide everything we need. Which is why the last two presidents have pushed hard for this and it’s a bipartisan talking point

The sector doesn't need to be 'revived' domestically, it just needs to be reinvigorated. Intel only lost the edge from 2018 onwards and it's still among the most sophisticated producers in the world. It's not a dead industry in need of continuous subsidy to be viable.

Anyone else reading Chip War?

I have it on my list - is it good?

I’m only 100 pages in but I’m hooked! I never took the time to understand just how complex the logistics are behind the silicon that powers basically everything we rely on. Totally worth picking up.

Thanks. This comment just bumped up its importance and is now on my “must-read-soon.”

It's very entertainingly written. And in an accessible style that doesn't require any insider industry knowledge. Strongly recommend.

Would recommend. Its not as detailed as I would like but paints a great overall narrative of how we went from the discovery of the transistor to the current state of the industry.

Chip War is good.

Acquired Podcast has done a great TSMC deep dive.

https://www.acquired.fm/episodes/tsmc

Asianometry has done a lot of semicon videos with lots of ASML EUV videos.

https://www.youtube.com/@Asianometry

listened to the audiobook, highly recommend.

While the books is on the way, you can listen to an engaging talk by the author: https://www.youtube.com/watch?v=2tSs7aESx8s

In my reading-list for this year. :-)

Wikipedia for those looking https://en.wikipedia.org/wiki/Chip_War:_The_Fight_for_the_Wo...

IIRC that's more than CHIPs Act, but it's 2024, Sam wants (lol) 7T for OpenAI hardware, I can't tell what's a lot of money anymore.

7T is price anchoring at it's best. When he gets 1T it will seem normal.

Or price anchoring at its worst. When he gets 100B it will look like a failure.

I'm sure 7T is someone having a laugh, not a real thing that's actually being attempted.

I wonder what it would cost to outright buy Australia, in the sense of convincing enough voters to make you their absolute monarch…

Well, current bleeding edge semicos require like hundreds of billions if you start from 0

Meanwhile what about those materials that could allow you to make 1000x faster cpus, but we arent capable of manufacturing them at scale and also their lifespan is way shorter than silicon.

Im not saying 7T is reasonable, but if thats what he want to do, then it is very interesting and expensive as hell

why not 8

https://x.com/sama/status/1758347811786281355?s=46

I had to Google 7T. I assume you mean $7 Trillion, not some new chip or fab technology.

Besides the obvious of this being a hedge against Taiwan being invaded. This is also a response to US sanctions, Japan is trying once again to have an independent semiconductor supply chain, just as an assurance. The government has been buying up companies to shore up their base. [1]

[1]https://www.cnbc.com/2023/06/26/japan-backed-fund-to-buy-sem...

Are you suggesting Japan is concerned they could be sanctioned by the US? While technically true, it's just as much of a risk for UK, Canada, etc. If they're safeguarding their access to chips that's about the least necessary risk to address.

I’m suggesting that Japan would prefer a higher degree of control and independence on semiconductor supply chain in general, to hedge against external factors like sanctions. I did not mean to imply that US would sanction them but US sanctions on equipment to China led to losses for Japanese firms.

Having a deeper control of end to end semiconductor supply chain will help hedge against that, equipment sanction revenue loss can be made up by chips fab revenue as such. Makes more sense when you look at it from the government perspective, it is the government driving this despite how it appears.

I suspect it would be more that if a company misbehaves, Japan may lose access (since the US will sanction the company).

It's general resilience for critical materials given geopolitical risk, not a fear of the US per se.

Going to be a hell of a chip glut in a few years, and these billions may well have been flushed down the toilet.

I get that there's a "national security" component but in 2029 when your subsidized and uncompetitive chip fabs are selling a dollar for fifty cents, your choice will be to pour more taxpayer money into the bottomless pit or admit defeat.

You seem to be assuming American production cannot be competitive, and that public goods operated by the government at a loss is inherently a bad thing.

Yes, I believe both things are pretty clearly true.

And not just American. India, every major EU country + UK, and a number of others are prepared to dump billions into chip fabs. And this does not even include China, Japan, Korea, and other Asian countries. The glut is going to be incredible.

Pouring taxpayer money into a bottomless pit at least enables graft

How is it that Japan lost its foothold in the consumer electronics market? You used to see a lot more Sony or Toshiba logos on things, but I feel like I never see japanese products anymore

Commoditization driven by other, cheaper Asian manufacturers (Korea and China mostly)

And an intense Japan-US trade war in the 80s

Does AI actually need miniature chips? Or just a lot of them, of mediocre technology? They could do space heaters with AI inference. Power consumption is a concern but come on, we re wasting tons of power just to maintain bitcoin

https://en.wikipedia.org/wiki/Dark_silicon

In AI training you typically use the full hardware all the time, unlike general purpose computing, so this is probably a bigger problem.

They need small chips but most important fast and big cache

For those wondering whether this is a state-backed venture, it is not [0]. In fact it is more interesting, as most of these large Japanese ventures (including Softbank via Arm) have significant dealings with semiconductors:

Rapidus was established in August 2022 with the support of eight major Japanese companies: Denso, Kioxia, MUFG Bank, NEC, NTT, SoftBank, Sony, and Toyota.

[0] https://en.wikipedia.org/wiki/Rapidus

any etf tracking this bundle?

edit: never mind. just saw that in 2002 the bigger player is now IBM. I don't want my money near that anymore.

I bet on IBM also when they were doing well in the quantum computing space. Similarly annoyed at their slow rate of progress relative to others.

What is it about big bureaucratic orgs that they get so... tangled? It's like an institutional form of Alzheimer's.

Soft.... You cant compete with $7T

I think you got down voted because not many people know the context of your joke: https://www.cnbc.com/2024/02/09/openai-ceo-sam-altman-report...

Classic HN

67B? This is a serious amount of money for the task, and they are far from starting from scratch without experience.

Just hope they won't favor chips with shady hardware programing manual or non-free specs (like ARM since it is own by japanese softbank).

Well thats a cost of building about 4 top tier FABs.

Genuine question: Japan is in a very bad location geographically speaking: weather events like earthquakes and tsunamis are a reality.

How is this still a good investment, considering the worsening of such events in the coming decades?

Doesn't it make more sense for companies to look for a safer bet?

How are you so sure that earthquakes and tsunamis will get worse in the coming decades? Their recent frequency is in line with historical records, and if anything their impact is far less damaging now due to improved construction techniques and infrastructure.

China’s self-inflicted reversal is unbelievable. All they had to do was make an honest progress, invest in education and research, and the rest was history. Instead they got too big headed and turned insular (again). Brain drain continues and Japan/SK/Vietnam stand to gain.

China’s self-inflicted reversal

People are overreading into the geopolitics.

This transition already started 15 years ago when Intel opened it's first Semiconductor Packaging factory in Hanoi in 2009 (yes, 2009 is 15 years ago now)

By 2008-09 it already became a bit difficult to operate within Coastal China for cost reasons.

Foxconn decided to expand inwards into Interior China (eg. Zhengzhou) while Japanese+Korean companies expanded/returned to Thailand (Japan), Vietnam (South Korea), and India (Japan).

The consumer electronics industry has always been a race to the bottom.

25 years ago, ASEAN and Mexico were the primary electronics manufacturing hub (Made in Thailand, Made in Malaysia, Made in Phillipines), then it became Made in China in the 2000s-2010s, now it's Made in Vietnam/India, and in a couple years who knows where.

Great. We need Japan to be competitive again. More choice is better for everyone.

it seems that Japanese culture and the business culture specifically is particularly suited to the precision and repeatability required to produce chips. I'm surprised that they ever got behind, but I suppose there are a lot of factors at play.

Great! I would love to see more competition in this space. Both culturally and educationally, Japan seems like it's in a strong position to make such a move.

It’s astonishing to look at the future and realize that all of this capacity is being created because of the forecast expansion of technology. Like, how crazy integrated into everything are we going to get?

Sam’s massive fab for AI chips alone is huge news. Japan is pouring it on.

I know people comment about the growth and bust cycle in chips, but it does create profound societal transformation in the process and I’m really starting to wonder just how profound the change that’s coming is going to be — if history guides us it looks pretty big.

The challenge all these AI vendors are trying to solve and why national priorities are turning towards supplementing this industry is to try to capture market share to solve critical shortage of computing power specifically (in microchips or GPUs) that are necessary to advance AI technologies.

And like I mentioned Nvidia has been the hot AI chip with the H100 (and it’s lead to a rather high market valuation — a trillion dollars). But their success is getting noticed and we are seeing Microsoft, Meta, OpenAI, Amazon, and Google all throw in to create their own AI processors. And that’s why you see this massive spending by nation states, because they don’t want to be left behind in this conversion and what is presumed to be ground floor opportunity for the “next big wave” of technology.

Different companies are doing different things but almost all of them are going all in on in-house design and many on FAB itself:

Microsoft is in-housing the Maia 100 for cloud.

Google is doing the TPUs and a bunch of other stuff not public yet.

Amazon won’t be left out with the hilariously named Trainium for machine learning and a bunch of chips under development for AWS.

META is doing the MTIA and Artemis in house.

SoftBank has realized that real estate isn’t it and now has their Izanagi project.

Huawei has their 910B.

Nvidia is all-in on GPUs like the H100 (and developing the H200) and making huge piles of money helping other create custom AI chips.

And of course I already mentioned it by Sam Altman is seeking trillions for some unnamed chipset.

What other major efforts do people know about? Seems like a massive shift and if this is as big as it looks, like I said, this is going to change society incredibly.