I don't see anything meaningful here. The foundry is still owned by Intel, has the same reporting chain as before, but with an extra operating board. There seems to be an idea that people will give the foundry money who wouldn't give Intel money, which seems dubious when the foundry is still Intel. Also not obvious that they'll do anything productive with more money.
Lots of emphasis on x86 and chiplets. I'm not totally sure Intel has much IP in chiplets but it definitely seems to be the winning play from AMD at present. "Continued leadership of the AI PC category, our strong position with AI in data center, and our accelerator portfolio" doesn't seem to have any correlation with reality.
Intel 20A was scrapped as part of the five nodes in four years success story. Currently shipping on TSMC instead. Intel are very confident that 18A will be great and work well and everyone will want it, shipping from a year from now, unlike the last N years of products.
I don't expect 18A to work either and Amazon will go elsewhere as that becomes increasingly clear. It's interesting times for semiconductors though.
edit: looks like there is a chiplet themed xeon shipping on intel processes, at least well enough to get review samples out, but I'm unclear on the details. E.g. is this equivalent to the AMD approach or merely using the same name to sound similar - I haven't found anything suggesting they can put GPU and CPU tiles on the same fabric for example
> I haven't found anything suggesting they can put GPU and CPU tiles on the same fabric for example
Not 100% sure what you mean by fabric, but Intel is absolutely combing CPU and GPU tiles onto a single chip with Meteor Lake. They are even using 4 different fabrication processes (two Intel, two TSMC).
Intel is being weighed down pretty dramatically by trying to do 5 nodes in 4 years AND do it at any scale. Intel's foundry business lost 2.8B last quarter, on revenue of 4.3B. That means they spent 7.1B in THREE months.
Intel currently can't use any of its new nodes to make their leading chips (AI PC, e.g.) because those nodes aren't operating at scale. So they're paying TSMC to make the chips. Which means that Intel isn't sending work to its own fabs to offset the costs of advancing nodes.
Plus, the foundry business hasn't managed to announce any flagship deals with outside parties.
By making it a sub at least they can raise funds to burn someone else's money while they try to spin things up.
The bigger headwind they face is that if Intel is tightly tied to the fabs, those fabs are going to be focused on serving intel, and not on serving the other customers it needs to service to have a leading node fab be remotely sustainable.
> Intel currently can't use any of its new nodes to make their leading chips
Xeon 6E was released on Intel 3 in June and Xeon 6P is launching next month. Data Center is currently a smaller market than client CPUs for Intel, but its still a $10B+ market segment for them.
Most of the rest of the industry is doing extremely well by separating design from foundry.
I’m not sure why one would compare Boeing when one could look at TSMC, ARM, Apple, etc instead, all of whom have been doing only one of the 2 all this while.
Heck, you could also look at AMD that was failing and split into AMD (design) and foundry (GlobalFoundries).
At the time of the split (approx 16 years ago) the company’s market cap was $8Bn.
Today AMD’s market cap is $150Bn and GlobalFoundries’s market cap is $21Bn.
Intel claims they can make the foundry break even by around 2027-2028, and be profitable beyond. My guess is once that happens, they plan to spin it off as a public company. This is a first step towards that goal.
Also, being able to report the gross margin for Intel Products and Intel Foundry separately will help Products. Currently that part of Intel (profitable) is "suffering" because it is being lumped in with the unprofitable part of Foundry.
Reporting and operating structure get affected too.
As an independent subsidiary, foundry gets to build an offering that will serve Intel and non-Intel customers best, while hopefully avoiding conflicts of interest.
The full letter from Intel's CEO is here: https://www.intc.com/news-events/press-releases/detail/1710/...
Thanks for the link, interesting reading.
I don't see anything meaningful here. The foundry is still owned by Intel, has the same reporting chain as before, but with an extra operating board. There seems to be an idea that people will give the foundry money who wouldn't give Intel money, which seems dubious when the foundry is still Intel. Also not obvious that they'll do anything productive with more money.
Lots of emphasis on x86 and chiplets. I'm not totally sure Intel has much IP in chiplets but it definitely seems to be the winning play from AMD at present. "Continued leadership of the AI PC category, our strong position with AI in data center, and our accelerator portfolio" doesn't seem to have any correlation with reality.
Intel 20A was scrapped as part of the five nodes in four years success story. Currently shipping on TSMC instead. Intel are very confident that 18A will be great and work well and everyone will want it, shipping from a year from now, unlike the last N years of products.
I don't expect 18A to work either and Amazon will go elsewhere as that becomes increasingly clear. It's interesting times for semiconductors though.
edit: looks like there is a chiplet themed xeon shipping on intel processes, at least well enough to get review samples out, but I'm unclear on the details. E.g. is this equivalent to the AMD approach or merely using the same name to sound similar - I haven't found anything suggesting they can put GPU and CPU tiles on the same fabric for example
> I haven't found anything suggesting they can put GPU and CPU tiles on the same fabric for example
Not 100% sure what you mean by fabric, but Intel is absolutely combing CPU and GPU tiles onto a single chip with Meteor Lake. They are even using 4 different fabrication processes (two Intel, two TSMC).
Outsourcing core engineering faculties went great for companies like... checks notes ... Boeing.
In all seriousness, is there anything more to a move like this than bookkeeping?
Yes and no?
Intel is being weighed down pretty dramatically by trying to do 5 nodes in 4 years AND do it at any scale. Intel's foundry business lost 2.8B last quarter, on revenue of 4.3B. That means they spent 7.1B in THREE months.
Intel currently can't use any of its new nodes to make their leading chips (AI PC, e.g.) because those nodes aren't operating at scale. So they're paying TSMC to make the chips. Which means that Intel isn't sending work to its own fabs to offset the costs of advancing nodes.
Plus, the foundry business hasn't managed to announce any flagship deals with outside parties.
By making it a sub at least they can raise funds to burn someone else's money while they try to spin things up.
The bigger headwind they face is that if Intel is tightly tied to the fabs, those fabs are going to be focused on serving intel, and not on serving the other customers it needs to service to have a leading node fab be remotely sustainable.
> Intel currently can't use any of its new nodes to make their leading chips
Xeon 6E was released on Intel 3 in June and Xeon 6P is launching next month. Data Center is currently a smaller market than client CPUs for Intel, but its still a $10B+ market segment for them.
To be fair to me, when I wrote this comment intel hadn’t announced any flagship deals.
https://www.pcmag.com/news/intel-developing-ai-chips-for-ama...
Most of the rest of the industry is doing extremely well by separating design from foundry.
I’m not sure why one would compare Boeing when one could look at TSMC, ARM, Apple, etc instead, all of whom have been doing only one of the 2 all this while.
Heck, you could also look at AMD that was failing and split into AMD (design) and foundry (GlobalFoundries).
At the time of the split (approx 16 years ago) the company’s market cap was $8Bn.
Today AMD’s market cap is $150Bn and GlobalFoundries’s market cap is $21Bn.
It went great for AMD and this is not a good comparison.
Intel claims they can make the foundry break even by around 2027-2028, and be profitable beyond. My guess is once that happens, they plan to spin it off as a public company. This is a first step towards that goal.
Also, being able to report the gross margin for Intel Products and Intel Foundry separately will help Products. Currently that part of Intel (profitable) is "suffering" because it is being lumped in with the unprofitable part of Foundry.
Reporting and operating structure get affected too. As an independent subsidiary, foundry gets to build an offering that will serve Intel and non-Intel customers best, while hopefully avoiding conflicts of interest.
[flagged]
[flagged]