With the threat of Taiwan being invaded by China, I think we are going to see a lot more of this. At least I hope we do.

This news doesn't make much sense right now. In the long run, sure, but "considering" switching to Intel, which is years off being where Nvidia would need them to be leaves them with Asia (and practically TSMC and Samsung) as their only options. Also, if Taiwan gets attacked, one shouldn't have the idea that all's good since "oh, our GPUs are made in South Korea anyway"... everybody should be well aware, that issues such as the ones implied, are felt under any circumstances. So yeah, nice move of some PR department, but ultimately, just talk. edit: sure they might fab in whatever Intel had. Which is already worse than what Nvidia used for Samsung for the 3000 series, so...

Can't wait to buy an RTX in 14nm+++++++++++++++

rtx 2060 12gb on 14nm++++++++ but seriosuly can't see the reason why not though,intel's next nodes are gonna be quite advanced.it's good to diversify. seems like nvidia wants to buy a share of every node seeing how bad the situation has been for the past two years.

They're just hedging their bets in case of future supply disruptions. Would be foolish to place all their eggs in one basket.

alanm:

They're just hedging their bets in case of future supply disruptions. Would be foolish to place all their eggs in one basket.

Does GloFo have any non-Taiwan fabs?

Here's what I don't get: Nvidia raked in almost $10 billion in net revenue for 2021. They could easily build their own facility wherever they wanted with that kind of money. It makes far more sense for them to supply their own chips than it was for them to buy ARM. They don't have to have excuses for supply chain issues. They don't have to complain about someone's node not being good enough. They don't have to deal with AMD bragging about being on a smaller node. They could pay less by not having to compete with companies like AMD and Apple.

schmidtbag:

Here's what I don't get: Nvidia raked in almost $10 billion in net revenue for 2021. They could easily build their own facility wherever they wanted with that kind of money. It makes far more sense for them to supply their own chips than it was for them to buy ARM. They don't have to have excuses for supply chain issues. They don't have to complain about someone's node not being good enough. They don't have to deal with AMD bragging about being on a smaller node. They could pay less by not having to compete with companies like AMD and Apple.

While I'd generally agree with you, the parts I made bold happened to Intel, although they do have their own fabs. You actually are responsible for a lot more things if you have your own fabs, and can't make somebody else be the one to blame when another earthquake hits Asia 😀 edit:

BLEH!:

Does GloFo have any non-Taiwan fabs?

Actually, they have non in Taiwan.... apparently, Singapore, US, Germany, unless I missed something.

fantaskarsef:

While I'd generally agree with you, the parts I made bold happened to Intel, although they do have their own fabs. You actually are responsible for a lot more things if you have your own fabs, and can't make somebody else be the one to blame when another earthquake hits Asia 😀

Funny thing is, I thought the same thing but just omitted it from my post. It's much easier to point fingers and say someone else is at fault than to take on such a huge responsibility. Nvidia saw how Intel and Samsung struggled to compete with TSMC and perhaps they [somehow] had the humility to realize the same could happen to them. Or perhaps they saw supply chain issues and didn't want to be blamed for it.

schmidtbag:

Here's what I don't get: Nvidia raked in almost $10 billion in net revenue for 2021. They could easily build their own facility wherever they wanted with that kind of money. It makes far more sense for them to supply their own chips than it was for them to buy ARM. They don't have to have excuses for supply chain issues. They don't have to complain about someone's node not being good enough. They don't have to deal with AMD bragging about being on a smaller node. They could pay less by not having to compete with companies like AMD and Apple.

Creating an high end factory for GPUs/CPUs is not that easy and is not worth it anyway. For some reason Apple doesn´t own their own fabs and they have more money that should be possible for a single company...

schmidtbag:

Here's what I don't get: Nvidia raked in almost $10 billion in net revenue for 2021. They could easily build their own facility wherever they wanted with that kind of money. It makes far more sense for them to supply their own chips than it was for them to buy ARM. They don't have to have excuses for supply chain issues. They don't have to complain about someone's node not being good enough. They don't have to deal with AMD bragging about being on a smaller node. They could pay less by not having to compete with companies like AMD and Apple.

Everyone in semi's took a look at Intel's fabs including AMD. Make no mistake they all did it to get info on Intel. This does not translate into them being customers.

H83:

Creating an high end factory for GPUs/CPUs is not that easy and is not worth it anyway. For some reason Apple doesn´t own their own fabs and they have more money that should be possible for a single company...

In today's market, it very much is worth it. I wouldn't be surprised if Nvidia loses hundreds of millions of dollars every year by contracting someone else to make their chips for them. Remember too that Nvidia could also take in other orders, much like Intel and Samsung do, further increasing their profits. But yeah, considering how much of the TSMC pie Apple takes (you could say it's an... Apple pie), it surprises me they don't make their own fab either.

schmidtbag:

Here's what I don't get: Nvidia raked in almost $10 billion in net revenue for 2021. They could easily build their own facility wherever they wanted with that kind of money.

They definitely cannot easily build their own facility. A modern fabrication plants costs ~$20B and about 8-10 years to build and those are costs projected from companies with existing supply chains/fabrication engineers/experience in fabrication/etc. Nvidia would have to build/acquire/develop all that. And in the end they wouldn't even get a much better facility than TSMC because all the tools these companies use to actually manufacture chips essentially come from ASML - which is a joint venture between all the major fabs anyway. It would be a massive gamble.. cost nearly two years of their yearly entire operating revenue and by the time the project is built it could be surpassed by any number of companies operating fabs already. Could they do it, sure? Would it be easy? Not even close to easy. It would be a massive expansion to the company, into a field where there are very few players, for almost little to no gain because the issues with global chip supply isn't the number of fabs but the supply chain itself.

reality bites. Intel has signed a contract with ASML for lithography machines worth hundreds of millions of dollars for their new fabs in Ohio and Magdeburg (and for the older fabs in Beaverton, Leixlip, and Israel) that will be breaking ground this year. which makes Intel the only other fab than TSMC with EUV lithography. there have been close relations with Nvidia and Intel (especially the engineers in mobile) for decades and both companies are just off of Lawrence Expwy. and eat at the same places for lunch (when they want something off campus) where everybody in (mid) Silicon Valley mingles. at least now with the less restrictive Covid rules. imho, this was all made possible - ironically - by Intel having TSMC as a bridge to their new fabs while Nvidia was very unhappy with Samsung iteration of the n5. this may be "inside baseball" or worse, "inside Cricket" but i may need to remind folks - no matter how fabulous AMD fabbed by TSMC is, Intel has a far greater circuit density @ node. which is why the 10n (++?) Intel is equivalent to an TSMC 7n when Intel gets their own ASML lithography they will be BONKERS @ 5,4, and (eventually) 3n.

... and ASML has already stated they're full until way into 2024 in regards to Denial's post.

Denial:

They definitely cannot easily build their own facility. A modern fabrication plants costs ~$20B and about 8-10 years to build and those are costs projected from companies with existing supply chains/fabrication engineers/experience in fabrication/etc. Nvidia would have to build/acquire/develop all that. And in the end they wouldn't even get a much better facility than TSMC because all the tools these companies use to actually manufacture chips essentially come from ASML. It would be a massive gamble.. cost nearly two years of their yearly entire operating revenue and by the time the project is built it could be surpassed by any number of companies operating fabs already.

Yes, they could. It's not going to cost $20B up-front. With their profit margins, they could keep up with construction costs as it is being built and not go into debt over it. Nvidia is a big enough name that I'm sure much of the challenges you speak of would not hold them back. They're aggressive enough to undercut the supply chain. Who said anything about getting something better than TSMC? The whole idea is they can more reliably keep up with production and lower production costs. The only reason I even bring this up is because Nvidia keeps switching manufacturers and they have more than once complained about the production process. Well, if they don't like how things are, they should do it themselves. EDIT: Let's not forget that this could potentially appeal to investors, which would help pay off the facility quicker.

schmidtbag:

In today's market, it very much is worth it. I wouldn't be surprised if Nvidia loses hundreds of millions of dollars every year by contracting someone else to make their chips for them. Remember too that Nvidia could also take in other orders, much like Intel and Samsung do, further increasing their profits. But yeah, considering how much of the TSMC pie Apple takes (you could say it's an... Apple pie), it surprises me they don't make their own fab either.

Steve Jobs looked very closely at Apple fabbing their own chips. when Intel was stuck in the mud, Apple looked closely again but decided that the AMD model was sound. Yes, it is possible for more money to be made, but at a huge upfront cost with the vagaries of weather, labor, and materials changing in the 18 months it takes to get a fab built and running (if you're like TSMC and have the Gov't running interference). but if you want your fab close to you and away from hostile political entities, that 18 months is an Environmental review taken with folks on Meth 😛 let alone permitting, purchasing, and outfitting. all of that versus an established company at the highest level operating with precision engineering and no octopus of various agencies with oversight with various agendas. i'm just being real.

schmidtbag:

Yes, they could. It's not going to cost $20B up-front. With their profit margins, they could keep up with construction costs as it is being built and not go into debt over it. Nvidia is a big enough name that I'm sure much of the challenges you speak of would not hold them back. They're aggressive enough to undercut the supply chain. Who said anything about getting something better than TSMC? The whole idea is they can more reliably keep up with production and lower production costs. The only reason I even bring this up is because Nvidia keeps switching manufacturers and they have more than once complained about the production process. Well, if they don't like how things are, they should do it themselves. EDIT: Let's not forget that this could potentially appeal to investors, which would help pay off the facility quicker.

I completely disagree. If Nvidia announced right now that they were building a fab i'd immediately sell all my shares. It's such a huge risk and massive gamble. You're paying half that $20B up front just to get in line with ASML. You need to sell capacity 5-6 years in advance because every other foundry is. Scouting locations takes years. The approval process for government regulation takes years. You'd have to poach so much talent from TSMC because there's a shortage of fabrication engineers. And in the end what exactly are you gaining? You keep saying "lowering production costs' but you're not.. you're spending $20B, and it doesn't even stop there. 2nm comes you're spending another $20-25B keeping up with everyone else. Honestly Nvidia would probably have to start a 2nm fab before the 3nm is done.. or 2/1 or whatever NM is a decade off. By then the supply chain issues are probably completely solved and TSMC is operating a much lower cost because they have 20+ years of experience/investments in supply chain and their production is spread out over tons of companies. Apple maybe has the revenue to try it and not let it completely tank them if it goes awry.. Nvidia? No way. In theory could they? Sure but it would be a horrible idea with almost zero benefit. Easily? There's dozens of articles describing chip manufacturing as the most difficult task ever. Just look at Intel for example. They dumped ~$10B over 5-6 years and basically showed nothing for it. Imagine if Nvidia dumped $10-15B over the next 5 years and ran into similar problems? And again more fabs doesn't even solve the current problems: https://www.extremetech.com/computing/333025-chip-making-companies-cant-buy-enough-lithography-machines They simply can't get machines built fast enough. Having another player doesn't help alleviate this. https://fortune.com/2022/02/22/india-chip-shortage-semiconductor-manufacturing-crisis-design-solution-rakesh-kumar/

Chip manufacturing is arguably the hardest technical exercise humankind currently undertakes—correctly etching and connecting tens to hundreds of billions of transistors 50 times smaller than a virus—and it’s getting harder. Twenty years back, there were 30 chip companies at the leading edge; now there are only two, and it’s unclear if a new entrant can break in.

On a related note, the semiconductor industry is under threat of disruptions due to neon shortages. Half the worlds neon production comes from plants in Ukraine which are now shuttered down. Litho machines cant function without it. https://www.cnbc.com/2022/03/25/russia-ukraine-war-laser-neon-shortage-threatens-semiconductor-industry.html

@Denial While I'll admit I don't know anything about the logistics of Nvidia building their own fab, you seem to be looking at the situation rather narrowly and directly comparing to TSMC, which is much larger than Nvidia needs to be. Nvidia doesn't need to match the production level of companies like TSMC or Intel. I'm sure they could be 1/4 the size of TSMC and still produce everything they need. If demand is still too high, they could outsource (which would likely cost less than it does now since they wouldn't have to undercut competitors as much). So no, they're not going to need $20B. In any case, what Nvidia is doing today is a gamble. They can't depend on TSMC because they either won't be able to produce enough or they're going to be too expensive. TSMC is also threatened by natural disasters and PRC. They can't depend on Samsung because apparently their process doesn't yield good enough results. As for Intel, they struggled to get to 10nm and Intel themselves are using TSMC to manufacture GPUs, so if I were an investor of Nvidia, I would not feel all that confident in their switch.

Just look at Intel for example. They dumped ~$10B over 5-6 years and basically showed nothing for it. Imagine if Nvidia dumped $10-15B over the next 5 years and ran into similar problems?

TSMC managed to pull it off sooner, so just because Intel struggled for so long, that doesn't mean Nvidia would too. I understand they would have a much harder time getting ahead considering the situation they'd be in, but there are a lot of variables involved. One of the variables that slowed down Intel was how they kept bumping clock speeds of their 14nm parts every time they had to delay 10nm. This just made it harder for them to release something that would actually be competitive. Intel managed to release low-end 10nm parts in 2019 so it's not like they didn't figure things out, but their approach made it difficult to proceed the way they intended. Nvidia's architecture is very different. Even if their R&D to do 10nm turned out exactly the same as Intel's, that doesn't mean they wouldn't have had to wait as long to get something released. Of course, the inverse could be true, where they would have to delay even longer. That's why no matter what Nvidia does, it's a gamble. So here's the thing about gambling: It's only stupid when you can't afford to lose what you put in. Nvidia can swap between manufacturers but should anything go catastrophically wrong because a tsunami or a missile wipes out a factory, they've lost everything for that generation, which would be difficult to recover from. Should such a situation happen, they're not going to be the only ones to immediately switch over to a competitor. Spending their spare funds (that came from their net revenue) on a new factory in a place that isn't likely to be threatened by nature or politics is, in my opinion, less risky. It's not like they're in a rush to get it built today. Of course being a public company, it would be a big issue if a new facility were to fail, but they could also sell what they attempted to minimize the damage.