With Intel’s latest layoffs, will the Ohio plant ever be built?

[ChrisGar]

Ireland is 4nm and 3nm? They should be ramping up. I just hope that Intel can add some foundry business to that so they stay full. Intel 3/4nm are good solid nodes. Where is the NOT TSMC business here? My concern is that the Intel Foundry sales team are mostly Intel sales people and not foundry sales people. If I was Lip-Bu I would re energize the sale team/replace them. The process nodes are ready, now the rubber hits the road.

I'm not sure about the entire staff or who attends the meetings -- but the head of foundry, Kevin O'Buckley, is an engineer.

 

[MKWVentures]

I must have missed this announcement. At the Intel event only one foundry customer was named and that was MediaTek on Intel 16. I spoke to Intel directly about this as well.

I am wondering about the cozy relationship with MediaTek. They are a very Taiwanese company and the CEO is the former TSMC CEO. Maybe there is some bad blood there? But again, if you want good TSMC pricing sing the praises of Intel Foundry.

I do not believe Intel has 9 committed 18A tapeouts that will result in real products. I believe that there are 9 plans to check out Intel on some products some of which will disappear. Others will be added.

I think Revenue (which is dropping) and Tapeouts (which is not growing fast) is the indicator. it may be trailing but until we get some non-intel input, it is the only data that we should trust IMHO.

We should have projected revenue for 2026 and 2027 and Projected Tapeouts for 2025, 2026 from Intel. Then compare actuals. Intel has this data and has info on how it is changed over the past 2 years.

 

[MKWVentures]

@nghanayem

So your thought is that it is too early to see revenue or tapeouts for 18A or Intel 3.


When would you expect to see >10 tapeouts total on 18A?
When would you expect to see 1.5B in annual wafer revenue (making intel a top 10 foundry).

When did TSMC release its PDK 1.0 for N2 compared to Intel for 18A?

 

[nghanayem]

I do not believe Intel has 9 committed 18A tapeouts that will result in real products. I believe that there are 9 plans to check out Intel on some products some of which will disappear. Others will be added.

Intel has by my count over the past 2 years announced 7 signed wafer agreements for 18A. In 2023 two unamed defense related ones and one unamed customer who in their contract will at some point pay out Intel a prepay for their capacity. In 2024 Microsoft AI chip, Amazon AI fabric chip, and in like the Q3 earnings Intel announced two more unamed 18A wafer agreements. Is it really so impossible to believe that I missed the last two 18A design wins or that there were two that weren't mentioned when Intel was busy talking about things investors cared more about like Falcon Shores, Intel products over projecting revenue again, or the CEO leaving?

I think Revenue (which is dropping)

I already explained why revenue isn't a useful metric to understand momentum and why Intel foundry external revenue now isn't a like for like with what it was back when they were selling substrates to OSATs during the pandemic and while they were also including IMS tool sales in IFS revenue.

and Tapeouts (which is not growing fast)

18A literally couldn't have had a single non Intel products tapeout until PDK 1.0 shipped in the back half of last year. That would be as foolish as saying "mmmmm TSMC N2 customer interest looks pretty weak to me if they only taped out 0 customer designs in 2023.".

So your thought is that it is too early to see revenue or tapeouts for 18A or Intel 3.

Yes. That is obvious. It isn't opinion it is an indisputable fact.

When would you expect to see >10 tapeouts total on 18A?

I don't know depends on customer schedule and when a 10th design win is landed (if it hasn't already). Based on their being 9 paying customers by EOY24, at the absolute earliest 10 18A tapeouts by EOY2027 or EOY2028.

When would you expect to see 1.5B in annual wafer revenue (making intel a top 10 foundry).

Before 2030 but after 2026 since Intel said their plans for a 2027 break even assumed negligible external revenue.

When did TSMC release its PDK 1.0 for N2 compared to Intel for 18A?

Mid 2024 for TSMC and late 2024 for 18A. I have noted reducing the gap between HVM readiness and foundry readiness needing to be a major focus point going forward on a couple of prior occasions. But since the top ten design houses started working on N2 designs since at least 2020, Intel foundry only started in mid 2021, and first customers were only signed in 2022, the gap between first Apple N2 product launch and first 18A external product launch will probably be closer to a year (rather than just the 6mo gap between PDK1.0 release). Lines up well with first 18A tapeout supposedly being due in 2H25 while first TSMC tapeout (almost assuredly Apple) was like 2H24.

 

[MKWVentures]

@nghanayem
I will stick by my comment "I do not believe Intel has 9 committed 18A future tapeouts today that will result in real products."
We will see what happens. I believe there is a distinction from announcements to commitments and real products.

I thought Intel shipped PDK1.0 in July 2024 (That is what Intel said). and TSMC was Mid 2024, correct?

And the first 18A external tapeout was announced as 1H2025 or 2H2025?

negligible revenue in 2027 and 1.5B by 2030 is very reasonable (you are way more pessimistic than my model LOL) .... but that is not what was said in April of 2024.

I think Tan's hints will become clear statements over the next 6 months as he shares the actual info on external wafer starts and revenue based on the planning numbers.

We shall see......

 

[nghanayem]

@nghanayem
I will stick by my comment "I do not believe Intel has 9 committed 18A future tapeouts today that will result in real products."

If you want to willingly ignore public fact and assume that Intel is commiting SEC fraud, you are always welcome to file a complaint with the SEC. That is what they are there for. In the meantime I will assume that Intel executives don't like the idea of jail time and are not claiming they have signed wafer agreements with people when they have not.

We will see what happens. I believe there is a distinction from announcements to commitments and real products.

A wafer agreement is ironclad unless the foundry fails to deliver or the fabless firms buys out their contract. If a chip is being designed with 18A in mind there is no reason to cancel the product unless the product is completely incapable of matching PDK criteria and the chip has unsellablly bad variations or power performance.

I thought Intel shipped PDK1.0 in July 2024 (That is what Intel said). and TSMC was Mid 2024, correct?

That sounds roughly right. I think it is like Feb or March for TSMC and I think the 18A PDK went out to early customers in like July with Intel announcing public availability in like August or something. Near enough to 6mo anyhow.

And the first 18A external tapeout was announced as 1H2025 or 2H2025?

As of last year they said 2H25, and they have been sticking with that claim so there isn't reason to assume anything has changed. But you are also welcome to assume Intel is once again lying if you want. Free country after all.

negligible revenue in 2027 and 1.5B by 2030 is very reasonable (you are way more pessimistic than my model LOL) .... but that is not what was said in April of 2024.

First stop making random claims up.

Second Pat, and DZ have said at least 10 times from late 2022 to today that Foundry break even in 2027 is assuming negligible external revenue, and that there may be upside (presumably mostly from advanced packaging since the time from design to HVM is much shorter).

Third I think it will be well before 2030 since Intel's goal is supposedly $15B per year to beat Samsung internal plus external on only external revenue. Going from a 300M baseline to 15B in 6 years means a 1.9x CAGR is needed. Seems plausible I guess with such a low baseline plus first wafer revenues coming this year from Tower and Mediatek, and a steady growth of advanced packaging revenue. Then in 2027 or 2028 having first 18A and Intel 3 revenues. With that said I would assume the revenue growth might be even more back loaded than that. Either way unless Intel expects revenue to go up 10x from 1.5B in 2029 to 15B in 2030, 1.5B probably needs to happen around 2026 or 2027. While I would call 1.5B non negligible I guess if we are talking less than 10% of total foundry revenue I guess I could see how DZ and Pat would call 1.5B negligible.

 

[MKWVentures]

If you want to willingly ignore public fact and assume that Intel is commiting SEC fraud, you are always welcome to file a complaint with the SEC. That is what they are there for. In the meantime I will assume that Intel executives don't like the idea of jail time and are not claiming they have signed wafer agreements with people when they have not.

A wafer agreement is ironclad unless the foundry fails to deliver or the fabless firms buys out their contract. If a chip is being designed with 18A in mind there is no reason to cancel the product unless the product is completely incapable of matching PDK criteria and the chip has unsellablly bad variations or power performance.

That sounds roughly right. I think it is like Feb or March for TSMC and I think the 18A PDK went out to early customers in like July with Intel announcing public availability in like August or something. Near enough to 6mo anyhow.

As of last year they said 2H25, and they have been sticking with that claim so there isn't reason to assume anything has changed. But you are also welcome to assume Intel is once again lying if you want. Free country after all.

First stop making random claims up.

Second Pat, and DZ have said at least 10 times from late 2022 to today that Foundry break even in 2027 is assuming negligible external revenue, and that there may be upside (presumably mostly from advanced packaging since the time from design to HVM is much shorter).

Third I think it will be well before 2030 since Intel's goal is supposedly $15B per year to beat Samsung internal plus external on only external revenue. Going from a 300M baseline to 15B in 6 years means a 1.9x CAGR is needed. Seems plausible I guess with such a low baseline plus first wafer revenues coming this year from Tower and Mediatek, and a steady growth of advanced packaging revenue. Then in 2027 or 2028 having first 18A and Intel 3 revenues. With that said I would assume the revenue growth might be even more back loaded than that. Either way unless Intel expects revenue to go up 10x from 1.5B in 2029 to 15B in 2030, 1.5B probably needs to happen around 2026 or 2027. While I would call 1.5B non negligible I guess if we are talking less than 10% of total foundry revenue I guess I could see how DZ and Pat would call 1.5B negligible.

We shall see.
Couple items: Intel said first external tapeout in 1H25, this was announced Aug 6th 2024 on the Intel website when they said they released PDK1.0 in July LOL
then it was changed to 2H 2025.

at the Intel foundry discussion in april 2024. Intel told multiple analysts that external revenue was expected to be 15B external in 2030. 5B in 2027-8. This is documented by many analysts at the meetings and published all over. This was confirmed by me with executives at the company.

Cancelling or changing a plan is not fraud. and even contracts are changed all the time with mutual agreement. those "9" plans will be adjusted. Intel never lies.... but sometimes you need to read between the lines to see when its a plan vs a commit. half of those plans are expected to change or disappear.

I appreciate your commitment to Intel foundry! You have incredible loyalty.

 

[Daniel Nenni]

@nghanayem
I will stick by my comment "I do not believe Intel has 9 committed 18A future tapeouts today that will result in real products."
We will see what happens. I believe there is a distinction from announcements to commitments and real products.

I have yet to see proof that Intel has said this publicly. I remember them saying publicly they were engaged with 7 of the top 10 semiconductor companies. Engaged does not mean wafer agreements have been signed. The answer I have gotten from people who would know is nope, they do not. This does not include the government stuff. I have no visibility there.

Lip-Bu communicates differently than Pat Gelsinger and Lip-Bu has not mentioned it. Lip-Bu was at CDNLive today singing the praises of Cadence. Same with Jensen Huang. Jensen absolutely LOVES Cadence. If not for Cadence, Jensen would still be a busboy at Denny's, or so he said.

 

[count]

In terms of Intel projections… surely in all investor presentations and announcements you’d have seen a long list of disclaimers providing legal cover for making announcements that are not outright fraud but very optimistic projections or numbers that are manipulated.

A agreement could be something like “If Intel meets all these targets at this price, the customer will buy $X in chips”, but the targets and the price could be pretty unrealistic.

 

[vslee]

I will assume that Intel executives don't like the idea of jail time and are not claiming they have signed wafer agreements with people when they have not.

A wafer agreement is ironclad unless the foundry fails to deliver or the fabless firms buys out their contract. If a chip is being designed with 18A in mind there is no reason to cancel the product unless the product is completely incapable of matching PDK criteria and the chip has unsellablly bad variations or power performance.

Did Intel actually say that all 9 companies have ironclad wafer agreements? Or are you just inferring that?

 

[Daniel Nenni]

Did Intel actually say that all 9 companies have ironclad wafer agreements? Or are you just inferring that?

That is the question. It was harder to tell with Pat. I believe Lip-Bu will be clearer on this type of thing. He will under promise and over perform. So if Lip-Bu says 6 design engagements he will deliver 9. That is my experience, observation, and opinion after working with Cadence the last 15 years.

 

[nghanayem]

Did Intel actually say that all 9 companies have ironclad wafer agreements? Or are you just inferring that?

I think I recall in the past Intel mentioning they had 35 engagements on 18A, with 7/10, and later "nearly every" leading fabless companies investigating 18A. By my count they have explicitly mentioned 7 in Intel's own words "signed 18A customers", "18A design wins", "signed a memorandum of understanding to use 18A for an upcoming product", "commited to using 18A and even signed for a prepay per their contract". I don't know what a signed customer, design win, a MOU to use 18A for a specific product, or a wafer prepay can mean other than a wafer agreement. That phrasing is also distinctly different than wording used like "strong customer engagement on 18A", "a flurry of activity after PDK1.0 release", "strong interest from fabless firms in 18A", "kicking the tires on 18A", "running test chips for 35 different potential foundry customers", and "35 interested potential customers". Logically it also doesn't really make sense. If Intel was talking about engagements rather than design wins they would be quoting that 35 number (or now that 18A is real and ready to go for foundry you would assume that engagement number might be even better than it was in like 2022/23) rather than 9. For 9 to be the number of people kicking the tires on 18A, the only alternative would be that more that 75% of those 35 bailing and Intel being unable to convince a single other company to kick the tires on 18A. For that to be true would mean there is less interest in 18A than in Intel custom foundey 10nm. Which would make no sense given 18A yield is solid, is in HVM now, PPA is competitive, it isn't a design nightmare, it has industry standard design tools/flows, and it has more than 0 industry standard IP ported and validated. If Intel could only land Amazon and Microsoft that is literally as bad or worse than ICF 10nm foundry (Phillips, LG, Altera, and another 1-2 FPGA guys). I'm sorry, but 18A being a worse failure than 10nm; that just doesn't pass a sniff test to me.

 

[count]

I think I recall in the past Intel mentioning they had 35 engagements on 18A, with 7/10, and later "nearly every" leading fabless companies investigating 18A. By my count they have explicitly mentioned 7 in Intel's own words "signed 18A customers", "18A design wins", "signed a memorandum of understanding to use 18A for an upcoming product", "commited to using 18A and even signed for a prepay per their contract". I don't know what a signed customer, design win, a MOU to use 18A for a specific product, or a wafer prepay can mean other than a wafer agreement. That phrasing is also distinctly different than wording used like "strong customer engagement on 18A", "a flurry of activity after PDK1.0 release", "strong interest from fabless firms in 18A", "kicking the tires on 18A", "running test chips for 35 different potential foundry customers", and "35 interested potential customers". Logically it also doesn't really make sense. If Intel was talking about engagements rather than design wins they would be quoting that 35 number (or now that 18A is real and ready to go for foundry you would assume that engagement number might be even better than it was in like 2022/23) rather than 9. For 9 to be the number of people kicking the tires on 18A, the only alternative would be that more that 75% of those 35 bailing and Intel being unable to convince a single other company to kick the tires on 18A. For that to be true would mean there is less interest in 18A than in Intel custom foundey 10nm. Which would make no sense given 18A yield is solid, is in HVM now, PPA is competitive, it isn't a design nightmare, it has industry standard design tools/flows, and it has more than 0 industry standard IP ported and validated. If Intel could only land Amazon and Microsoft that is literally as bad or worse than ICF 10nm foundry (Phillips, LG, Altera, and another 1-2 FPGAOPER guys). I'm sorry but 18A being a worse failure than 10nm; that just doesn't pass a sniff test to me.

I have no doubt that every semiconductor company is looking at 18A, because there is a lot of information that can be gleaned at very little cost, and maybe a lot of upside if things work out. But at the end of the day Intel needs a very cost competitive process and that's where I think they will struggle.

We've already seen that Intel 4/3 is not cost competitive, even if the performance is good, so Intel has shifted production back to Intel 7.

As many people have said, Intel is trying to become the second supplier to TSMC... well the entire point of having a second supplier is to keep the primary supplier in check on costs. And if the second supplier is way more expensive than the primary supplier, there is no reason to have a second supplier at all. Might as well consolidate your volumes and get better pricing and economies of scale.

 

[Daniel Nenni]

I think I recall in the past Intel mentioning they had 35 engagements on 18A, with 7/10, and later "nearly every" leading fabless companies investigating 18A. By my count they have explicitly mentioned 7 in Intel's own words "signed 18A customers", "18A design wins", "signed a memorandum of understanding to use 18A for an upcoming product", "commited to using 18A and even signed for a prepay per their contract". I don't know what a signed customer, design win, a MOU to use 18A for a specific product, or a wafer prepay can mean other than a wafer agreement. That phrasing is also distinctly different than wording used like "strong customer engagement on 18A", "a flurry of activity after PDK1.0 release", "strong interest from fabless firms in 18A", "kicking the tires on 18A", "running test chips for 35 different potential foundry customers", and "35 interested potential customers". Logically it also doesn't really make sense. If Intel was talking about engagements rather than design wins they would be quoting that 35 number (or now that 18A is real and ready to go for foundry you would assume that engagement number might be even better than it was in like 2022/23) rather than 9. For 9 to be the number of people kicking the tires on 18A, the only alternative would be that more that 75% of those 35 bailing and Intel being unable to convince a single other company to kick the tires on 18A. For that to be true would mean there is less interest in 18A than in Intel custom foundey 10nm. Which would make no sense given 18A yield is solid, is in HVM now, PPA is competitive, it isn't a design nightmare, it has industry standard design tools/flows, and it has more than 0 industry standard IP ported and validated. If Intel could only land Amazon and Microsoft that is literally as bad or worse than ICF 10nm foundry (Phillips, LG, Altera, and another 1-2 FPGAOPER guys). I'm sorry but 18A being a worse failure than 10nm; that just doesn't pass a sniff test to me.

A recently ex Intel person told me that Pat was fired due to this type of language. They could not pin him down on foundry futures. TSMC makes it very clear. Intel foundry needs to communicate like TSMC if they want the NOT TSMC business. We should not need a decoder ring to figure out how many tape-outs Intel Foundry has on each node.

 

[vslee]

I don't know what a signed customer, design win, a MOU to use 18A for a specific product, or a wafer prepay can mean other than a wafer agreement.

Take a look at the definition for MOU then.

From legaltemplates.net:

A Memorandum of Understanding (MOU) is a formal agreement between two or more parties that outlines their intentions and the terms of their collaboration. It typically includes shared goals, tasks, and expectations, but it is generally not legally binding. MOUs are often used to establish a mutual understanding before formal contracts are created, and they can vary in complexity depending on the nature of the agreement.

From wiki:

A memorandum of understanding (MoU) is a type of agreement between two (bilateral) or more (multilateral) parties. It expresses a convergence of will between the parties, indicating an intended common line of action. It is often used either in cases where parties do not imply a legal commitment or in situations where the parties cannot create a legally enforceable agreement. It is a more formal alternative to a gentlemen's agreement.

So if you're seeing MOU, then it specifically means that a legally binding contract has not been signed.

 

[alicemia12]

Intel’s recent layoffs certainly raise questions about the future of the Ohio plant, but it’s hard to say for sure. Similar uncertainty surrounded tech launches, like the iPhone15, but they still moved forward let's hope the same happens here!