The Final Nail in Intel’s Coffin?

[Xebec]

Given the first announced foundry customer deal was Microsoft, worth $15B, the other 5 customers must be worth $0 ?

I’m think many news sites are getting the MS = $15B number wrong.

If you google for “Intel 18A foundry customers”, you get a base $10B going to $15B after MS was added, around the February timeframe. (Implying MS business is worth $5B).

Intel Foundry Wins "Magnificent Seven" Customer as Backlog Rockets Past $15 Billion | The Motley Fool

Microsoft will use Intel's manufacturing services for a next-generation chip.

www.fool.com

https://www.reuters.com/technology/intel-expects-overtake-tsmc-making-fastest-chips-this-year-2024-02-21/

(And just before the MS announcement):

Intel's Foundry Business Has $10 Billion Locked In, Who are the Four Customers?

The biggest shift in Intel's (NASDAQ: INTC) strategy over the past few years under CEO Pat Gelsinger has been opening up its manufacturing operations to third parties. This isn't the first time Intel has attempted to build its own foundry business -- it tried and failed for about five years...

semiwiki.com

 

[fansink]

I’m think many news sites are getting the MS = $15B number wrong.

Thanks for the clarification

 

[Tanj]

I do not envy running any part of Intel right now. Assault on all aspects of the business

The best thing is that the marauders are all turning towards a much bigger treasure - and threat - from the green one. Intel is a distraction.

 

[Daniel Nenni]

Lots of interesting stuff in Goldman Sachs discussion:

https://seekingalpha.com/article/4696592-intel-corporation-intc-presents-goldman-sachs-global-semiconductor-conference-transcript

“We now have 6 Intel 18A external foundry customers. We've got a lifetime deal value of greater than $15 billion.“

”2/3 of the PCs shipped next year won’t be AI PCs” - Reminder - all Apple and Qualcomm-based PCs will be AI-PCs

“Graviton and other Cloud hyperscalers chips are used for internal workloads, but x86 is a lock-in for external workloads”

That was a canned discussion. "5 nodes in 4 years" blah blah blah. He clearly has no idea how the foundry business works. If IFS has $15B in lifetime deals then TSMC must have what, $15T?

 

[VCT]

I believe its a pretty big surprise to some people. TSMC sales to Intel in 2025 will be 10X higher than a couple years ago. Most leading edge Processor silicon will be from TSMC in 2025 (I call leading edge below Intel 7, Sorry Dan ). Intel's most competitive PC processors in 2025 may be 100% TSMC. The GM for CCG wants to use TSMC for some tiles long term even when Intel releases 18A, 14A.

Once Intel starts to ramp 18A significantly (2026) we will get a sense of how Intel sales to other companies compare to Intel purchases from TSMC.

This is all good for Intel product group. They have the options and flexibility they need to release multiple products. I think that by the end of 2024, we will see WHY TSMC was a positive gamechanger for Intel

"Intel ramp up 18A in 2026"
The problem is that Intel probably has to commit to TSMC's 2026 capacity 18 months in advance. Make it very difficult to decide.

 

[Daniel Nenni]

"Intel ramp up 18A in 2026"
The problem is that Intel probably has to commit to TSMC's 2026 capacity 18 months in advance. Make it very difficult to decide.

The Intel N3 agreement was signed years ago and it was pre pay. I doubt there will be a new agreement similar in scale of N3 but Intel has used TSMC for many years and will continue to do so, my opinion. When the original agreement was signed Intel design groups would get a choice between Intel manufacturing and TSMC. From what I hear that is no longer the case. Intel design groups will use Intel manufacturing for leading edge products. Intel must get internal wafer volumes up if they ever want to compete with TSMC on price.

 

[MKWVentures]

The Intel N3 agreement was signed years ago and it was pre pay. I doubt there will be a new agreement similar in scale of N3 but Intel has used TSMC for many years and will continue to do so, my opinion. When the original agreement was signed Intel design groups would get a choice between Intel manufacturing and TSMC. From what I hear that is no longer the case. Intel design groups will use Intel manufacturing for leading edge products. Intel must get internal wafer volumes up if they ever want to compete with TSMC on price.

how much was the prepay? All of the products Listed on the Client roadmap today have most of the silicon on TSMC. Intel foundry sells to Intel Product at TSMC prices which is what allows Intel product to choose Intel over TSMC (Foundry take the loss until 2028+).

 

[Daniel Nenni]

how much was the prepay? All of the products Listed on the Client roadmap today have most of the silicon on TSMC. Intel foundry sells to Intel Product at TSMC prices which is what allows Intel product to choose Intel over TSMC (Foundry take the loss until 2028+).

The prepay was not publicly disclosed. It should be in Intel's financials somewhere?

Personally I feel Intel will focus on high performance chiplets while outsourcing the support chiplets until they can get yield to a TSMC level. It looks like TSMC N2 will be another dominant node so it is a real difficult time to get back into the foundry business. You should try and get into the Samsung event later this month. I expect Samsung Foundry will up its PR game significantly since Intel Foundry is taking the wind from Samsung Foundry's sails, absolutely.

 

[VCT]

The prepay was not publicly disclosed. It should be in Intel's financials somewhere?

Personally I feel Intel will focus on high performance chiplets while outsourcing the support chiplets until they can get yield to a TSMC level. It looks like TSMC N2 will be another dominant node so it is a real difficult time to get back into the foundry business. You should try and get into the Samsung event later this month. I expect Samsung Foundry will up its PR game significantly since Intel Foundry is taking the wind from Samsung Foundry's sails, absolutely.

Regarding the upcoming N3 chips (NVIDIA, AMD, Qualcomm, MTK), TSMC has indicated that the current pricing is too low. When do you expect the current contract to expire, allowing TSMC to raise the prices for N3 chips? early 2026?

 

[Daniel Nenni]

Regarding the upcoming N3 chips (NVIDIA, AMD, Qualcomm, MTK), TSMC has indicated that the current pricing is too low. When do you expect the current contract to expire, allowing TSMC to raise the prices for N3 chips? early 2026?

Do you have a link to the pricing too low comment? I have no doubt it has been said but I'm curious about the context.

Pricing is a very confidential thing and it depends on the customer. A wafer agreement generally lasts the life of the product(s) but sometimes there are cost adjustments. If you read about pricing information take it with a grain of salt. The wafer prices I have seen published are always on the high side.

Given that TSMC owns the 3nm and 2nm nodes and demand is up I think price increases are a given. Hopefully Intel and Samsung foundry will be competitive in the angstrom era so wafer pricing gets competitive again. With chiplets fabless companies are able to spread die amongst the foundries more easily than full chips so that may help. Unless of course you want to use TSMC packaging. Exciting times in the semiconductor industry, absolutely.

 

[Mooredaddy]

Do you have a link to the pricing too low comment? I have no doubt it has been said but I'm curious about the context.

Pricing is a very confidential thing and it depends on the customer. A wafer agreement generally lasts the life of the product(s) but sometimes there are cost adjustments. If you read about pricing information take it with a grain of salt. The wafer prices I have seen published are always on the high side.

Given that TSMC owns the 3nm and 2nm nodes and demand is up I think price increases are a given. Hopefully Intel and Samsung foundry will be competitive in the angstrom era so wafer pricing gets competitive again. With chiplets fabless companies are able to spread die amongst the foundries more easily than full chips so that may help. Unless of course you want to use TSMC packaging. Exciting times in the semiconductor industry, absolutely.

I believe TSMC mentioned this in their last earnings Q&A. That 3nm was priced years ago and that’s why they aren’t seen to being extracting as much profit as analysts expected.

 

[Daniel Nenni]

I believe TSMC mentioned this in their last earnings Q&A. That 3nm was priced years ago and that’s why they aren’t seen to being extracting as much profit as analysts expected.

This makes sense. When the N3 agreements were signed TSMC had no idea Intel and Samsung would fumble. Hopefully Intel 18A and Samsung 2nm gets new N3 and N2 customers some price negotiation leverage.

 

[VCT]

Do you have a link to the pricing too low comment? I have no doubt it has been said but I'm curious about the context.

Pricing is a very confidential thing and it depends on the customer. A wafer agreement generally lasts the life of the product(s) but sometimes there are cost adjustments. If you read about pricing information take it with a grain of salt. The wafer prices I have seen published are always on the high side.

Given that TSMC owns the 3nm and 2nm nodes and demand is up I think price increases are a given. Hopefully Intel and Samsung foundry will be competitive in the angstrom era so wafer pricing gets competitive again. With chiplets fabless companies are able to spread die amongst the foundries more easily than full chips so that may help. Unless of course you want to use TSMC packaging. Exciting times in the semiconductor industry, absolutely.

from TSMC CFO
"But another reason is that we set the pricing of N3 very early, several years ahead of production. However, we experienced a lot of cost inflation pressures in the following years. So as a result, N3 will take a longer time than N5 and N7 to reach the corporate average gross margin. For N2,based on what we can see so far is that we are doing a better job in cost and selling our value, and we expect N2 to have a better margin profile than N3."

N2 margin will be good. "doing a better job in selling our value" means increasing price.

 

[VCT]

Also Intel claim

from TSMC CFO
"But another reason is that we set the pricing of N3 very early, several years ahead of production. However, we experienced a lot of cost inflation pressures in the following years. So as a result, N3 will take a longer time than N5 and N7 to reach the corporate average gross margin. For N2,based on what we can see so far is that we are doing a better job in cost and selling our value, and we expect N2 to have a better margin profile than N3."

N2 margin will be good. "doing a better job in selling our value" means increasing price.

Also Intel said hoping to get some production back from TSMC in 2026 or 2027, that mean 2026 or 2027 pricing not confirmed yet.

 

[jabo]

Speaking of TSMC and Intel manufacturing - does anyone find it interesting that even Intel, notorious for delays on its product roadmaps, has pushed/delayed everything but not Lunar Lake? Lunar Lake was planned to come out in Q4 2024 or H1 2025, and they've already announced it is coming out in Q4. Lunar Lake is all TSMC silicon, and we haven't seen Intel products actually launch on time for a while.

Really speaks to the fact that TSMC's "trusted foundry" should be considered almost completely trustable. Even Intel, which struggle to even release their products on time according to roadmap and has numerous delays - they still have not had a delay on their all TSMC products*.

*Unless you count the ARC A series on N7, which arguably was the driver teams fault.

Except that under Pat, Intel has been delivering on time. He has the execution engine running just great.

 

[IanD]

from TSMC CFO
"But another reason is that we set the pricing of N3 very early, several years ahead of production. However, we experienced a lot of cost inflation pressures in the following years. So as a result, N3 will take a longer time than N5 and N7 to reach the corporate average gross margin. For N2,based on what we can see so far is that we are doing a better job in cost and selling our value, and we expect N2 to have a better margin profile than N3."

N2 margin will be good. "doing a better job in selling our value" means increasing price.

It certainly does, going by the numbers we've seen... :-(

As for A16 with BSPD, the Rolls-Royce principle applies -- if you have to ask how much it costs, you can't afford it... ;-)