Unfortunately this is the story of how Intel lost it's Foundry business.
The same exact thing happened when Otellini wasn't willing to fab Apple's ARM chips right before iPhone took off..
Samsung to Produce Tesla Chips in $16.5 Billion Multiyear Deal
- Thread starter kevin01
- Start date
Unfortunately this is the story of how Intel lost it's Foundry business.
The same exact thing happened when Otellini wasn't willing to fab Apple's ARM chips right before iPhone took off..
user nl
Active member
Yes, I can imagine CC Wei (and all of TSMC customers) would not appreciate it when Elon is messing around TSMCs Fabs in Arizona and pissing off all his engineers, like also the new (star)-CEO of TSMC-Arizona, who's job it is to make Arizona a well-run and yielding TSMC-Gigafab, as if it was located in Taiwan
This situation marks four wins and one loss:
(1) Tesla successfully diversified its supply chain, cutting costs;
(2) Samsung landed a major deal for the Texas facility and is positioned to support GAA development;
(3) TSMC, though losing the contract, relieved itself of the strain of enlarging investment in U.S.
(4) The U.S. saw a boost in semiconductor production.
The outlier? Intel Foundry
(1) Tesla successfully diversified its supply chain, cutting costs;
(2) Samsung landed a major deal for the Texas facility and is positioned to support GAA development;
(3) TSMC, though losing the contract, relieved itself of the strain of enlarging investment in U.S.
(4) The U.S. saw a boost in semiconductor production.
The outlier? Intel Foundry
This may have been a Texas thing. Elon moved Tesla HQ from Palo Alto to Austin in 2021 due to a tiff with Gavin Newsome, Governor of California. It was a COVID and tax problem.
“California used to be the land of opportunity... now it’s become more so the land of overregulation, overlitigation, over taxation.” Elon.
Plus Tesla has a Giga Factory in Texas so I suspect a backroom political deal here. Maybe Elon will make Samsung Foundry great again?
user nl
Active member
I would add, the biggest win, Elon is not messing around with his slaves in Washington, tearing down stuff like Science (funding) and Education (funding), the systems with the largest ROI!! And that he has no clue about (from his actions) on how they work and why they work the way they work.
Having an opinion on bureaucracy, on running Science and Education, isn't making you a competent and strategic leader to steer bureaucracy in those domains.
Tearing down stuff you do not understand is easy, especially if you think that you are the smartest guy on earth who seems to have an urgency to verbalize his opinion on everything that he has an opinion about.
Last edited:
It is also a point against the TSMC monopoly argument. Given how loud Elon Musk is it is a big point. Go Elon!
user nl
Active member
This may have been a Texas thing. Elon moved Tesla HQ from Palo Alto to Austin in 2021 due to a tiff with Gavin Newsome, Governor of California. It was a COVID and tax problem.
“California used to be the land of opportunity... now it’s become more so the land of overregulation, overlitigation, over taxation.” Elon.
Plus Tesla has a Giga Factory in Texas so I suspect a backroom political deal here. Maybe Elon will make Samsung Foundry great again?
Perhaps Elon's long-term aim is running for governor of Texas, with his new political party! Texas is important for DT and his successors (JDV) to keep in their control.....
user nl
Active member
Except Tesla volumes will never be Apple volumes.
And Intel didn't have a foundry business at that point in time.
True re: volumes, but "Tesla" is running a lot of custom silicon these days - which can represent both licensing fees as well as manufacturing profits:
- annual 2 million infotainment/car computers (think equivalent to 2 million gaming consoles) - Ryzen APUs
- annual 4 million FSD chips (each car has 2 chips for redundancy)
- annual ~ 20,000 large die Tesla Dojo "D1" chips: D1 chip is 645mm2 on TSMC N7 process (50B transistors)
This will increase to include AI6 chips for xAI's Grok Supercomputer, potentially replacing the 200,000 (current) to 1 million (projected) high end Nvidia GPUs Grok is using. In addition, AI6 may also replace some Nvidia functions on Tesla Dojo.
Whatever it is, it appears to be enough business to justify an advanced Samsung fab in Texas, meanwhile Intel can't justify a fab past 18A yet, and some folks think 18A will lose money for a long while too..
Tesla Dojo - Wikipedia
True re: volumes, but "Tesla" is running a lot of custom silicon these days - which can represent both licensing fees as well as manufacturing profits:
- annual 2 million infotainment/car computers (think equivalent to 2 million gaming consoles) - Ryzen APUs
- annual 4 million FSD chips (each car has 2 chips for redundancy)
- annual ~ 20,000 large die Tesla Dojo "D1" chips: D1 chip is 645mm2 on TSMC N7 process (50B transistors)
This will increase to include AI6 chips for xAI's Grok Supercomputer, potentially replacing the 200,000 (current) to 1 million (projected) high end Nvidia GPUs Grok is using. In addition, AI6 may also replace some Nvidia functions on Tesla Dojo.
Whatever it is, it appears to be enough business to justify an advanced Samsung fab in Texas, meanwhile Intel can't justify a fab past 18A yet, and some folks think 18A will lose money for a long while too..
Tesla Dojo - Wikipedia
en.m.wikipedia.org
I agree, this is the type of deal that Intel Foundry can do. It is more than just wafer sales, they really are joint development agreements (JDAs). I'm not sure how well Musk will integrate into the Samsung culture when trying to automate their fabs with robots but it will certainly be a learning experience and fun to watch. Kind of like Elon's integration into Washington DC Politics, like a bull in a china shop.
Practically Tesla's fab now. Maybe they can handle the tool installations from the very beginning!
And with robots tesla made.
Yes, this is what I meant about Tesla wanting a lot of control over how suppliers do things. They will send their engineers to supplier factories and everyone, both Tesla engineers and suppliers will be forced to work 24/7 on anything that is causing an issue. Samsung will be giving up a lot of operational control and potentially will risk a lot of IP exposure.
Tesla will also strong arm Samsung at every turn, withhold payments over minor issues, threaten lawsuits constantly, steal company secrets, ect. I've consistently seen this kind of behavior from Tesla people firsthand. Samsung will probably end up regretting this.
That's interestingTrue re: volumes, but "Tesla" is running a lot of custom silicon these days - which can represent both licensing fees as well as manufacturing profits:
- annual 2 million infotainment/car computers (think equivalent to 2 million gaming consoles) - Ryzen APUs
- annual 4 million FSD chips (each car has 2 chips for redundancy)
- annual ~ 20,000 large die Tesla Dojo "D1" chips: D1 chip is 645mm2 on TSMC N7 process (50B transistors)
This will increase to include AI6 chips for xAI's Grok Supercomputer, potentially replacing the 200,000 (current) to 1 million (projected) high end Nvidia GPUs Grok is using. In addition, AI6 may also replace some Nvidia functions on Tesla Dojo.
Whatever it is, it appears to be enough business to justify an advanced Samsung fab in Texas, meanwhile Intel can't justify a fab past 18A yet, and some folks think 18A will lose money for a long while too..
Tesla Dojo - Wikipedia
TSMC has been making Tesla's chips without problem. The first one was on N7 and the new one is on N4. Samsung made Tesla chips on 14nm and 7nm. Samsung's 5nm had serious yield problems as did 3nm. Samsung in Texas will make 4nm and 2nm chips. It must be a good value proposition for Tesla for them to switch from TSMC N4 to Samsung 4nm/2nm.
This story and What does it have to do with Strongarm?
I think it's a bit unreasonable
Apart from that, instead of strongarm, there must have been a way to create x86 chips with ultra-low power segments by focusing on low power consumption designs.
I don't think it would have been impossible if I had worked hard at that point.
In fact, the ATOM for the early days of mobile In fact, I don't think the ATOM SOC for the early days was bad.
I can understand and agree if you say this Intel has started the foundry business earlier.
Even TSMC may have become an untouchable existence.
Personally, I was thinking about TSMC's N4 and later processes.
It wasn't a problem with TSMC's technology, but I wanted to say about how much it can be supplied.
In that regard, I don't think there's much room after N4,
Moreover, if you think that the original chip for FSD has become more versatile, a certain number would be necessary.