AMD Exec Says Intel IFS is Destined to Fail

[Daniel Nenni]

This reminds me of the time a Senior Fellow at Intel said TSMC could not succeed, the fabless model was broken, blah blah blah. Not long after that TSMC could have said the same, that the IDM model was broken, do to the missteps of Samsung and Intel, but they did not. The only thing statements like this do is motivate the competition and annoy people like me. Hopefully Darren retracts this statement or is fired. I personally would not want a leader at my company publicly stepping on his man parts and unduly motivating the competition.

Intel will in fact succeed for reasons that have been discussed here. In the battle for the NOT TSMC business against Samsung Intel will succeed and if the latest PDKs from TSMC. Samsung, and Intel (N2, 3nm, and 18A) are any indication it will be a heated battle for the next generation of foundry process technologies, absolutely.

And lets be honest, the real reason AMD went fabless is because they were horribly behind Intel and had no other choice. Then they were locked to GF and still behind Intel. Now that AMD is with TSMC they are catching up but not to the point they should be bad mouthing the competition like this:

Asked at the Canalys EMEA Forum 2023 if Intel can succeed, Darren Grasby, exec VP for strategic partnerships and president of AMD EMEA, replied emphatically: "Of course not."

He hinted that the decision to embrace contract manufacturing could be a turn that Intel might come to regret.

"Intel has gone down these paths," he said, "and if you think about the journey of AMD we had our own fabs many years ago and we chose to go fabless, and it was the turning point of the company that allowed us to invest those R&D dollars into the roadmap, and they're the roadmaps that are bringing that product and leading edge technology to market today."

Intel building fabs? Turnaround? Think again, says AMD exec

Exiting fab biz was 'turning point' for House of Zen, claims Darren Grasby

www.theregister.com

 

[hist78]

This reminds me of the time a Senior Fellow at Intel said TSMC could not succeed, the fabless model was broken, blah blah blah. Not long after that TSMC could have said the same, that the IDM model was broken, do to the missteps of Samsung and Intel, but they did not. The only thing statements like this do is motivate the competition and annoy people like me. Hopefully Darren retracts this statement or is fired. I personally would not want a leader at my company publicly stepping on his man parts and unduly motivating the competition.

Intel will in fact succeed for reasons that have been discussed here. In the battle for the NOT TSMC business against Samsung Intel will succeed and if the latest PDKs from TSMC. Samsung, and Intel (N2, 3nm, and 18A) are any indication it will be a heated battle for the next generation of foundry process technologies, absolutely.

And lets be honest, the real reason AMD went fabless is because they were horribly behind Intel and had no other choice. Then they were locked to GF and still behind Intel. Now that AMD is with TSMC they are catching up but not to the point they should be bad mouthing the competition like this:

Asked at the Canalys EMEA Forum 2023 if Intel can succeed, Darren Grasby, exec VP for strategic partnerships and president of AMD EMEA, replied emphatically: "Of course not."

He hinted that the decision to embrace contract manufacturing could be a turn that Intel might come to regret.

"Intel has gone down these paths," he said, "and if you think about the journey of AMD we had our own fabs many years ago and we chose to go fabless, and it was the turning point of the company that allowed us to invest those R&D dollars into the roadmap, and they're the roadmaps that are bringing that product and leading edge technology to market today."

Intel building fabs? Turnaround? Think again, says AMD exec

Exiting fab biz was 'turning point' for House of Zen, claims Darren Grasby

www.theregister.com

"NOT TSMC" is not enough to save the IDM Intel.

It will take years' effort and heavy investment to build capacity and revenue for the IFS business. But Intel overall revenue is shrinking faster than the revenue IFS can generate.

I sense while we are talking about this "NOT TSMC" strategy, Intel itself is moving forward with outsourcing to TSMC to de-risk on over relying on IFS.

 

[Xebec]

"NOT TSMC" is not enough to save the IDM Intel.

It will take years' effort and heavy investment to build capacity and revenue for the IFS business. But Intel overall revenue is shrinking faster than the revenue IFS can generate.

I sense while we are talking about this "NOT TSMC" strategy, Intel itself is moving forward with outsourcing to TSMC to de-risk on over relying on IFS.

Intel seems to be recognizing this in part by building limited capacity for certain nodes. Intel 4 (+packaging) is a big leap but they’re not going to build enough capacity to service desktop in addition to Mobile. Intel 3 likewise looks to be not intended for all of Intel’s markets.

I think their strategy is - keep capacity on leading nodes at a balance between cost/reward while ensuring node advancement gets priority. Then once Intel has the lead (20A/18A?) raise capacity in accordance with confidence.

 

[Daniel Nenni]

Things are changing in the ecosystem now more than ever before. I think the TSMC market share will jump quite a bit as a result of the N3 tidal wave of customers and that will give rise to the NOT TSMC efforts just in time for the new Intel process technologies. Intel knew 4/3 would not hit the foundry business in time so they are betting on 18a and below. We all know there must be an alternative to TSMC at the leading edge so business will go to Intel. If 18A is a success and the Intel process roadmap goes deep, which is what Intel used to be known for, I see opportunity for foundry success, absolutely.

Will they be the next TSMC? No. Can they be a strong #2? Easily. And it is not just about the revenue. Intel will gain customer/ecosystem experience which is what made TSMC great.

 

[MKWVentures]

Things are changing in the ecosystem now more than ever before. I think the TSMC market share will jump quite a bit as a result of the N3 tidal wave of customers and that will give rise to the NOT TSMC efforts just in time for the new Intel process technologies. Intel knew 4/3 would not hit the foundry business in time so they are betting on 18a and below. We all know there must be an alternative to TSMC at the leading edge so business will go to Intel. If 18A is a success and the Intel process roadmap goes deep, which is what Intel used to be known for, I see opportunity for foundry success, absolutely.

Will they be the next TSMC? No. Can they be a strong #2? Easily. And it is not just about the revenue. Intel will gain customer/ecosystem experience which is what made TSMC great.

View attachment 1485

The NOT TSMC is nice, but the NOT TSMC people have to be competitive. One can choose UMC or Samsung or GF or Intel. People choose not to

is this Samsung number included Samsung Internal? in other words Samsung is roughly equivalent to GF for external shipments. Side note: Intel numbers include packaging and mask making. I believe actual wafer sales (classic foundry) is less than $500M per year.

 

[Daniel Nenni]

The NOT TSMC is nice, but the NOT TSMC people have to be competitive. One can choose UMC or Samsung or GF or Intel. People choose not to

is this Samsung number included Samsung Internal? in other words Samsung is roughly equivalent to GF for external shipments. Side note: Intel numbers include packaging and mask making. I believe actual wafer sales (classic foundry) is less than $500M per year.

Yes Samsung includes internal products, Intel will as well but since Intel is outsourcing to TSMC this number has more relevance. Samsung, to my knowledge, does not use TSMC.

I am talking about leading edge foundry (Whale) business and today it is TSMC, Samsung, and Intel. As I have mentioned, the new PDKS are out (N2, 3nm, and 18A) and the IP people are saying that they are very close to being silicon ready so we have a new race for the GAA technology generation. Intel18a has a performance advantage, TSMC N2 has density and power, Samsung 3nm has cost. Given that TSMC N3 has 90% market share it will be interesting to see where the new design starts land in 2025.

 

[MKWVentures]

Yes Samsung includes internal products, Intel will as well but since Intel is outsourcing to TSMC this number has more relevance. Samsung, to my knowledge, does not use TSMC.

I am talking about leading edge foundry (Whale) business and today it is TSMC, Samsung, and Intel. As I have mentioned, the new PDKS are out (N2, 3nm, and 18A) and the IP people are saying that they are very close to being silicon ready so we have a new race for the GAA technology generation. Intel18a has a performance advantage, TSMC N2 has density and power, Samsung 3nm has cost. Given that TSMC N3 has 90% market share it will be interesting to see where the new design starts land in 2025.

great info, thanks

 

[hist78]

Things are changing in the ecosystem now more than ever before. I think the TSMC market share will jump quite a bit as a result of the N3 tidal wave of customers and that will give rise to the NOT TSMC efforts just in time for the new Intel process technologies. Intel knew 4/3 would not hit the foundry business in time so they are betting on 18a and below. We all know there must be an alternative to TSMC at the leading edge so business will go to Intel. If 18A is a success and the Intel process roadmap goes deep, which is what Intel used to be known for, I see opportunity for foundry success, absolutely.

Will they be the next TSMC? No. Can they be a strong #2? Easily. And it is not just about the revenue. Intel will gain customer/ecosystem experience which is what made TSMC great.

View attachment 1485

This CNBC graph titled "World's Largest Chips Foundries (Q2 2023 revenue)" is questionable and misleading:

1. CNBC said it obtained the foundry revenue data from company earnings reports. But Samsung only published Device Solutions(DS) Q2 revenue of $14.73 KRW Trillion and of that $8.97 KRW Trillion was from the memory products. Samsung DS division has Samsung memory, LSI (SoCs, CPU, AP, etc.), foundry, and display businesses. As far as I know, Samsung doesn't disclose the foundry service revenue in its earning reports. Where did CNBC get the Samsung Foundry revenue number from?

2. In this CNBC graph, Intel is the seventh largest foundry in the world. This is wrong.

Based on the Q2 2023 revenue, there are two public traded foundry companies that are bigger than Intel.

Powerchip Semiconductor: US$353.39 million
Vanguard International Semiconductor: US$316.32 million
(NT$1 = US$0.0321, June 30, 2023)

Source:

https://images.samsung.com/is/content/samsung/assets/global/ir/docs/2023_2Q_conference_eng.pdf

https://www.powerchip.com/upload/media/financials/quarterly_result/2023-Q2-CFS_EN.pdf

https://media-vis.todayir.com/20230829165114991275707_en.pdf

 

[Daniel Nenni]

This CNBC graph titled "World's Largest Chips Foundries (Q2 2023 revenue)" is questionable and misleading:

1. CNBC said it obtained the foundry revenue data from company earnings reports. But Samsung only published Device Solutions(DS) Q2 revenue of $14.73 KRW Trillion and of that $8.97 KRW Trillion was from the memory products. Samsung DS division has Samsung memory, LSI (SoCs, CPU, AP, etc.), foundry, and display businesses. As far as I know, Samsung doesn't disclose the foundry service revenue in its earning reports. Where did CNBC get the Samsung Foundry revenue number from?

2. In this CNBC graph, Intel is the seventh largest foundry in the world. This is wrong.

Based on the Q2 2023 revenue, there are two public traded foundry companies that are bigger than Intel.

Powerchip Semiconductor: US$353.39 million
Vanguard International Semiconductor: US$316.32 million
(NT$1 = US$0.0321, June 30, 2023)

Source:

https://images.samsung.com/is/content/samsung/assets/global/ir/docs/2023_2Q_conference_eng.pdf

https://www.powerchip.com/upload/media/financials/quarterly_result/2023-Q2-CFS_EN.pdf

https://media-vis.todayir.com/20230829165114991275707_en.pdf

My guess is that Samsung outside foundry business is comparable to UMC and SMIC so there would be a 3 way tie for second. The interesting this is the gap between TSMC and Others. And it will get bigger when N3 ramps up in 2025.

 

[count]

The Intel market isn't the not TSMC market. It's more narrowly defined than that. It's leading edge + high cost/performance + not TSMC. How big of a market is this, especially considering 3 biggest customers in that market (AMD/Nvdia/Apple) are dedicated to TSMC? Is the leading edge + high cost/performance + not TSMC - AMD/Nvidia/Apple market big enough to justify the tens of billions of capex required to keep up with TSMC at leading edge?

I have my doubts.

 

[hist78]

The Intel market isn't the not TSMC market. It's more narrowly defined than that. It's leading edge + high cost/performance + not TSMC. How big of a market is this, especially considering 3 biggest customers in that market (AMD/Nvdia/Apple) are dedicated to TSMC? Is the leading edge + high cost/performance + not TSMC - AMD/Nvidia/Apple market big enough to justify the tens of billions of capex required to keep up with TSMC at leading edge?

I have my doubts.

To Intel, "NOT TSMC" also means not Made-in-Taiwan. With the TSMC fab expansions in Japan, US, and Germany, this is going to be less significant. And unfortunately the current tragedy in Israel has made Intel's supply chain less reliable than it said.

 

[delong.height]

"NOT TSMC" is not enough to save the IDM Intel.

It will take years' effort and heavy investment to build capacity and revenue for the IFS business. But Intel overall revenue is shrinking faster than the revenue IFS can generate.

I sense while we are talking about this "NOT TSMC" strategy, Intel itself is moving forward with outsourcing to TSMC to de-risk on over relying on IFS.

NOT TSMC is enough to save Intel for sure. Let me explain, among all of the fabless companies, there are 2 renowned companies that's trying to diversify their supply chain. Nvidia and Qualcomm, we have been talking about these two in the past on this forum and why they had been trying to do so. So, it always bothered me to think of any other reason that they won't choose IFS for their wafer/packaging capacity.

Intel itself can contribute 20B of revenue to IFS, and that is 2024, a time where Intel has to depend on TSMC for some of its capacity because it hasn't got ready for its EUV process node. If Intel can get any of the two to sign, it will definitely be a great moment for celebration.

Second, I don't think we should discount of any possibility for Intel and Samsung to go in partnership. If Intel can master five nodes in four years, Intel can very well go in a partnership with Samsung and have a slightly leading position in this joint partnership. The two will be the best complements of each other. Samsung has a vast manufacturing capacity on leading edge, and a very large number of EUV that's under-utilized, it can make cheap wafer with thin margin, but it cannot compete with TSMC on quality. While Intel can provide technological advancement like RibbonFET and PowerVia that's ahead of two. We have seen a number of joint partnerships over the last couple of years, like ST micro and GF, Intel and Tower (on New Mexico plant). If Intel and Samsung ever go in joint partnership, the two will be very likely to attract majority of fabless players out there. And it may even include Apple and AMD. Intel will also need to manufacture some of its cheap part in Asia, so Samsung can be one of the great partners if they execute well. For Intel, the biggest problem is it costs a lot to build new fabs and bring those capacity online. So, it makes sense to me that Intel can play the role IBM was playing, and Samsung can fund Intel's R&D and be the biggest franchisee of Intel process node and packaging technology. In the past and present AMD and Nvidia are the greatest competitors to Intel, so there would be worried that IFS may potentially steal their IP, etc. But if Samsung now become the middleman, the two can very well balance their supply chain, and shift some of their wafer production to be at Samsung.

To me, the biggest hurdle for Samsung and Intel is TSMC because of its reputation and quality, which is why they can get >90% of leading-edge market share. I don't think they should view each other as competitors because they control a very small amount of market. If they can join the R&D and capacity, it will definitely bring balance to the chip world.


The stronger TSMC is, more likely this will happen.

 

[blueone]

NOT TSMC is enough to save Intel for sure. Let me explain, among all of the fabless companies, there are 2 renowned companies that's trying to diversify their supply chain. Nvidia and Qualcomm, we have been talking about these two in the past on this forum and why they had been trying to do so. So, it always bothered me to think of any other reason that they won't choose IFS for their wafer/packaging capacity.

Intel itself can contribute 20B of revenue to IFS, and that is 2024, a time where Intel has to depend on TSMC for some of its capacity because it hasn't got ready for its EUV process node. If Intel can get any of the two to sign, it will definitely be a great moment for celebration.

Second, I don't think we should discount of any possibility for Intel and Samsung to go in partnership. If Intel can master five nodes in four years, Intel can very well go in a partnership with Samsung and have a slightly leading position in this joint partnership. The two will be the best complements of each other. Samsung has a vast manufacturing capacity on leading edge, and a very large number of EUV that's under-utilized, it can make cheap wafer with thin margin, but it cannot compete with TSMC on quality. While Intel can provide technological advancement like RibbonFET and PowerVia that's ahead of two. We have seen a number of joint partnerships over the last couple of years, like ST micro and GF, Intel and Tower (on New Mexico plant). If Intel and Samsung ever go in joint partnership, the two will be very likely to attract majority of fabless players out there. And it may even include Apple and AMD. Intel will also need to manufacture some of its cheap part in Asia, so Samsung can be one of the great partners if they execute well. For Intel, the biggest problem is it costs a lot to build new fabs and bring those capacity online. So, it makes sense to me that Intel can play the role IBM was playing, and Samsung can fund Intel's R&D and be the biggest franchisee of Intel process node and packaging technology. In the past and present AMD and Nvidia are the greatest competitors to Intel, so there would be worried that IFS may potentially steal their IP, etc. But if Samsung now become the middleman, the two can very well balance their supply chain, and shift some of their wafer production to be at Samsung.

To me, the biggest hurdle for Samsung and Intel is TSMC because of its reputation and quality, which is why they can get >90% of leading-edge market share. I don't think they should view each other as competitors because they control a very small amount of market. If they can join the R&D and capacity, it will definitely bring balance to the chip world.


The stronger TSMC is, more likely this will happen.

An interesting joint strategy concept for Intel and Samsung, it sounds like good business sense, but the two companies do not appear to have a positive relationship. It seems a long stretch to think they could be partners.

 

[delong.height]

Samsung 3nm has cost

I supposed here you meant that Samsung is willing to sell at a thin margin unlike the other two.

 

[delong.height]

The Intel market isn't the not TSMC market. It's more narrowly defined than that. It's leading edge + high cost/performance + not TSMC. How big of a market is this, especially considering 3 biggest customers in that market (AMD/Nvdia/Apple) are dedicated to TSMC? Is the leading edge + high cost/performance + not TSMC - AMD/Nvidia/Apple market big enough to justify the tens of billions of capex required to keep up with TSMC at leading edge?

I have my doubts.

I see Intel trying to convert themselves into performance per watt. But it will take time. And for the question "How big of a market is this", you can think of all segments where Intel is participating right now. PC, Server, Edge, HPC, GPU, AI. The unit may not be comparable to Arm, the ones done at TSMC, but it's high margin business.

 

[delong.height]

An interesting joint strategy concept for Intel and Samsung, it sounds like good business sense, but the two companies do not appear to have a positive relationship. It seems a long stretch to think they could be partners.

Yeah, but I think that's the only shot they can possibly have. And this chart perfectly illustrates the other side. As time goes on, there will be even less participants in this market. In 2001, we had 18 leading edge companies, in 2020, it goes down to 2. Samsung definitely will need Intel's help if they are going to miss 3nm again, and Intel needs Samsung in order to expand its own capacity and reduce manufacturing costs and better fund its R&D objectivity.

 

[hist78]

NOT TSMC is enough to save Intel for sure. Let me explain, among all of the fabless companies, there are 2 renowned companies that's trying to diversify their supply chain. Nvidia and Qualcomm, we have been talking about these two in the past on this forum and why they had been trying to do so. So, it always bothered me to think of any other reason that they won't choose IFS for their wafer/packaging capacity.

Intel itself can contribute 20B of revenue to IFS, and that is 2024, a time where Intel has to depend on TSMC for some of its capacity because it hasn't got ready for its EUV process node. If Intel can get any of the two to sign, it will definitely be a great moment for celebration.

Second, I don't think we should discount of any possibility for Intel and Samsung to go in partnership. If Intel can master five nodes in four years, Intel can very well go in a partnership with Samsung and have a slightly leading position in this joint partnership. The two will be the best complements of each other. Samsung has a vast manufacturing capacity on leading edge, and a very large number of EUV that's under-utilized, it can make cheap wafer with thin margin, but it cannot compete with TSMC on quality. While Intel can provide technological advancement like RibbonFET and PowerVia that's ahead of two. We have seen a number of joint partnerships over the last couple of years, like ST micro and GF, Intel and Tower (on New Mexico plant). If Intel and Samsung ever go in joint partnership, the two will be very likely to attract majority of fabless players out there. And it may even include Apple and AMD. Intel will also need to manufacture some of its cheap part in Asia, so Samsung can be one of the great partners if they execute well. For Intel, the biggest problem is it costs a lot to build new fabs and bring those capacity online. So, it makes sense to me that Intel can play the role IBM was playing, and Samsung can fund Intel's R&D and be the biggest franchisee of Intel process node and packaging technology. In the past and present AMD and Nvidia are the greatest competitors to Intel, so there would be worried that IFS may potentially steal their IP, etc. But if Samsung now become the middleman, the two can very well balance their supply chain, and shift some of their wafer production to be at Samsung.

To me, the biggest hurdle for Samsung and Intel is TSMC because of its reputation and quality, which is why they can get >90% of leading-edge market share. I don't think they should view each other as competitors because they control a very small amount of market. If they can join the R&D and capacity, it will definitely bring balance to the chip world.


The stronger TSMC is, more likely this will happen.

"Intel itself can contribute 20B of revenue to IFS, and that is 2024,"

Use this 20 billion as an example, what's the gross profit margin do you expect for IFS and for the Intel Design/Product division?

 

[Daniel Nenni]

I supposed here you meant that Samsung is willing to sell at a thin margin unlike the other two.

Yes, wafer cost to the customer.

 

[count]

It takes minimum $20b of annual foundry capex (and growing) to stay on leading edge. $20b in revenue is a fraction of what is needed. I would guess ~$50b in revenue would be bare minimum to support a leading edge foundry business.

 

[delong.height]

"Intel itself can contribute 20B of revenue to IFS, and that is 2024,"

Use this 20 billion as an example, what's the gross profit margin do you expect for IFS and for the Intel Design/Product division?

for IFS, I would expect 30%, for Design/Product, I think upper 50s ~ 60s is highly likely.