Pat Gelsinger pitched a strategic reversal to Intel’s board. They liked it so much, they made him CEO

[Daniel Nenni]

Interesting article. I'm wondering if the Intel board would have hired Pat without the foundry part of the IDM 2.0 pitch? That's the part I am still struggling with.

Pat Gelsinger pitched a strategic reversal to Intel’s board. They liked it so much, they made him CEO

He sold Intel’s board on his plan but hasn’t yet persuaded Wall Street.

www.oregonlive.com

Not only is the foundry business an unnecessary distraction, it will be a serious financial burden, not to mention the agony of foundry defeat yet again. Clearly Intel will rely heavily on TSMC for 3nm due to the delay in Intel 7nm and the lack of EUV systems and expertise. How does Intel land serious foundry customers while using TSMC as a crutch? And what happens if Intel 5nm is late and/or not competitive with TSMC and Samsung? Which is a very strong possibility. And what about price? Price has always been and always will be a determining factor in the foundry business. Can Intel really compete on price with Samsung, TSMC, and the others? It boggles the mind.

It just seems to me that Pat is setting himself and the Intel board up for serious investor disappointment on the foundry P&L.. Wouldn't it be wiser to be a little more humble and exceed your goals rather than to come up short? PT Barnum would disagree but the semiconductor industry really isn't a circus.

 

[soAsian]

Interesting article. I'm wondering if the Intel board would have hired Pat without the foundry part of the IDM 2.0 pitch? That's the part I am still struggling with.

Pat Gelsinger pitched a strategic reversal to Intel’s board. They liked it so much, they made him CEO

He sold Intel’s board on his plan but hasn’t yet persuaded Wall Street.

www.oregonlive.com

Not only is the foundry business an unnecessary distraction, it will be a serious financial burden, not to mention the agony of foundry defeat yet again. Clearly Intel will rely heavily on TSMC for 3nm due to the delay in Intel 7nm and the lack of EUV systems and expertise. How does Intel land serious foundry customers while using TSMC as a crutch? And what happens if Intel 5nm is late and/or not competitive with TSMC and Samsung? Which is a very strong possibility. And what about price? Price has always been and always will be a determining factor in the foundry business. Can Intel really compete on price with Samsung, TSMC, and the others? It boggles the mind.

It just seems to me that Pat is setting himself and the Intel board up for serious investor disappointment on the foundry P&L.. Wouldn't it be wiser to be a little more humble and exceed your goals rather than to come up short? PT Barnum would disagree but the semiconductor industry really isn't a circus.

Isn't this is the prefect time to pitch the IDM 2.0? US govrt is worry about chip shortage and rely too much on Asia for semis production. Intel come in with IDM 2.0 to get some free govrt chesses with the help from Semiconductors in America Coalition

 

[Daniel Nenni]

Isn't this is the prefect time to pitch the IDM 2.0? US govrt is worry about chip shortage and rely too much on Asia for semis production. Intel come in with IDM 2.0 to get some free govrt chesses with the help from Semiconductors in America Coalition

Yes, definitely, it is a perfect time for Intel to ask for handouts from the US Government. Unfortunately it is a fools errand. There is no way to replicate the semiconductor supply chain in the United States in a profitable manner no matter how much money is given away. Just ask China.

 

[tooLongInEDA]

Yes, definitely, it is a perfect time for Intel to ask for handouts from the US Government. Unfortunately it is a fools errand. There is no way to replicate the semiconductor supply chain in the United States in a profitable manner no matter how much money is given away. Just ask China.

Daniel, I have asked this before and got no reply. Why would the US govt not also try to form some sort of "joint venture" with TSMC ? Is this something TSMC simply refuses to entertain ? I assume you're close enough to TSMC to have some view on this.

 

[Daniel Nenni]

Daniel, I have asked this before and got no reply. Why would the US govt not also try to form some sort of "joint venture" with TSMC ? Is this something TSMC simply refuses to entertain ? I assume you're close enough to TSMC to have some view on this.

I’m not sure that would be politically correct in the current political environment. TSMC building fabs in AZ is a good start but that is only part of the chip equation. Mask shops, packaging, test, etc…

The US is the largest market for TSMC (60%?). So what will a handful of trailing edge fabs really do for the US chip supply chain?

 

[smeyer0028]

I see the problem as US government not willing to apply anti-trust
laws and break up our monopolies and cartels. If DEC and its Alpha
chip and original engineering drivernHP had been allowed to survive,
I do not think therre would be the current semiconductor problems.

 

[james juang]

Using TSMC's 3nm gives Intel a chance to compete with AMD. It is still an uphill battle against AMD's chiplet structure but a lot better fighting on both fronts -process and design.

Building 7nm (or EUV) fabs everywhere gives Intel another chance to catch up with TSMC.

One other pressing issue is what to do with those non-EUV fabs if the production of CPUs has moved to EUV notes. One way out is to convert those to IFS. If this is the plan, Intel has to move quickly. The demand for non-EUV wafers is strong as ever.

 

[Daniel Nenni]

Using TSMC's 3nm gives Intel a chance to compete with AMD. It is still an uphill battle against AMD's chiplet structure but a lot better fighting on both fronts -process and design.

Building 7nm (or EUV) fabs everywhere gives Intel another chance to catch up with TSMC.

One other pressing issue is what to do with those non-EUV fabs if the production of CPUs has moved to EUV notes. One way out is to convert those to IFS. If this is the plan, Intel has to move quickly. The demand for non-EUV wafers is strong as ever.

I don't think Intel will ever catch up to TSMC so to compete with AMD Intel will need to continue to outsource product. Intel is still taking giant process steps while TSMC does yearly half steps for better yield learning.

Even if Intel converts older fabs to the foundry model there is still a lot of technical work to do. At 28nm and above you can take your design to TSMC, SMIC, GF, and UMC without modification. The designs will need to be reworked to get them into Intel fabs. Who is going to pay for that? And what about the wafer cost difference? TSMC can make wafers for 20-30% less than Intel. Who is going to pay for that difference?

The foundry business is all about cost, usable capacity, and ecosystem. IFS is zero for three today and tomorrow is a long ways away. Just my opinion of course.

 

[james juang]

No doubt! There are challenges ahead. Let's look at the positive side of legacy-node IFS: (1) Process: Intel has the best legacy-node process in the world. With some fine-tuning, it could very well be good to go. The efforts are reasonable. (2) Idle Fabs: It is a matter of when not if the CPU production has to move to EUV Fabs. Idle Fabs are the worst nightmare for any Fab company. legacy-node IFS will keep them busy. (3) Foundry model: Intel personnel will learn how to operate in the foundry model. It has to start from somewhere. Leading-node IFS will fight both fronts- business model and process. It is tough to convince fabless to port its state-of-art designs in an IDM but easier for legacy-node designs. (4) Cost: The equipment should all be fully depreciated. UMC announced a new fab last month. The best part is that its customers will pay for it. If UMC can operate a new fab with some profits, there is no reason for Intel to operate an old fab with a loss.

 

[Portland]

Investors, shareholders, wall street ... want to see growth. Amd because where they were can grow but Intel in this pandemic can't. Is it really pat gelsinger's fault? no but he'll be held accountable.

Mike Rogoway likes to push his narrative no matter how untruthful and ridiculous. He attacks people that call him out.

 

[count]

Sounds like Intel just can't get over it's former glory. Rather than shift it's business model to adjust to the new market dynamics, it's doubled down on an already failed strategy.

 

[Daniel Nenni]

Sounds like Intel just can't get over it's former glory. Rather than shift it's business model to adjust to the new market dynamics, it's doubled down on an already failed strategy.

Maybe Pat did what he needed to do to get his dream job but in the end he will be stuck with the strategy that got Bob Swan fired, partnering with TSMC and going fab light.

 

[Jert]

I don't think Intel will ever catch up to TSMC so to compete with AMD Intel will need to continue to outsource product. Intel is still taking giant process steps while TSMC does yearly half steps for better yield learning.

Even if Intel converts older fabs to the foundry model there is still a lot of technical work to do. At 28nm and above you can take your design to TSMC, SMIC, GF, and UMC without modification. The designs will need to be reworked to get them into Intel fabs. Who is going to pay for that? And what about the wafer cost difference? TSMC can make wafers for 20-30% less than Intel. Who is going to pay for that difference?

The foundry business is all about cost, usable capacity, and ecosystem. IFS is zero for three today and tomorrow is a long ways away. Just my opinion of course.

According to this logic then we should do nothing because there is nothing to be done. I simply can't disagree more.
US is playing too nice and gentlemen in this super competitive game (actually it is a zero-sum game because winner takes it all), self-imposing rules and self-tying its onw hands. If TSMC 60% (I think maybe even higher today) is US customers, we should have pushed TSMC to open fabs in the US long ago. If it were China, they would have been already been forced to hand over the IP, or even have their fabs confiscated. Look at how aggressive the small country of South Korea government and companies are together investing in their semi tech future. How could we do nothing? Previously Intel was the pioneer for the EUV, many years ahead of TSMC or Samsung. It still has the technological capacity (and money) to catch up. We should not lose faith in the American spirit. IFS definitely won't make money in the short run, it going to take time. TSMC was not built in 2 years.

 

[Daniel Nenni]

According to this logic then we should do nothing because there is nothing to be done. I simply can't disagree more.

This is not logic nor an assumption, it is my opinion based on 30+ years experience inside the fabless semiconductor ecosystem. If you can find someone with foundry experience that disagrees let me know. I would welcome a live conversation and would make it a podcast for all to hear. Any takers?

 

[hist78]

According to this logic then we should do nothing because there is nothing to be done. I simply can't disagree more.
US is playing too nice and gentlemen in this super competitive game (actually it is a zero-sum game because winner takes it all), self-imposing rules and self-tying its onw hands. If TSMC 60% (I think maybe even higher today) is US customers, we should have pushed TSMC to open fabs in the US long ago. If it were China, they would have been already been forced to hand over the IP, or even have their fabs confiscated. Look at how aggressive the small country of South Korea government and companies are together investing in their semi tech future. How could we do nothing? Previously Intel was the pioneer for the EUV, many years ahead of TSMC or Samsung. It still has the technological capacity (and money) to catch up. We should not lose faith in the American spirit. IFS definitely won't make money in the short run, it going to take time. TSMC was not built in 2 years.

"Previously Intel was the pioneer for the EUV, many years ahead of TSMC or Samsung. "

Even that's the case, Intel screwed it up.

Will Intel have enough EUV for 7nm?

It sure does not seem like it. Big win for TSMC 3N? https://drive.google.com/file/d/1tVRejNMAoeCNKyyTODnPEFkNLomtF9DV/view

semiwiki.com

 

[Xebec]

Respectfully, Some counterarguments for consideration --

The article and leaks so far don't state any explicit goals for which manufacturing processes, how much market share, or how much $ total business Pat wants Intel to aim for with respect to re-entering the foundry market.

If Intel's primary goal is to ensure x86 dominance (and Intel market share dominance) via recapturing the lead for fab technology, I think the foundry play is fully complementary.

IMO, The near-term threats to Intel are primarily the resurgent AMD, and in Intel not being positioned well enough to gain stable marketshare in newer markets such as AI and ML. Purely from a technology perspective, AMD is looking to be 1 or 2 nodes behind TSMC bleeding edge for at least the next 3 years. (ex: Zen 4 will be a TSMC 5nm product 1 full year after Apple is already shipping 3nm products).

The long-term threats are many, including the fundamental increases in chip fab and production costs that require increasing revenue and profits for funding those node shrinks (regularly) to make sense. Re-deploying Intel foundry services in almost any capacity can serve this master: Intel can re-use older fabs even longer than are sustainable on their core markets for products that don't need the latest technology -- meaning more revenue from fully depreciated assets, or reduced costs for any "lower volume" Intel chips that can leverage older fabs. Alternative, If Intel generates foundry business on bleeding edge nodes then they'll have more money to develop those nodes (the same way Apple is funding TSMC), and be effectively denying revenue to competitors in the overall silicon space.

Intel re-entering the foundry market will/could enable them to run to the very end of scaling on Silicon (alongside TSMC) which x86 alone* might not be able to financially support.

*Especially under attack from a resurgent AMD using Apple-funded TSMC cutting edge fabs.

 

[hist78]

Respectfully, Some counterarguments for consideration --

The article and leaks so far don't state any explicit goals for which manufacturing processes, how much market share, or how much $ total business Pat wants Intel to aim for with respect to re-entering the foundry market.

If Intel's primary goal is to ensure x86 dominance (and Intel market share dominance) via recapturing the lead for fab technology, I think the foundry play is fully complementary.

IMO, The near-term threats to Intel are primarily the resurgent AMD, and in Intel not being positioned well enough to gain stable marketshare in newer markets such as AI and ML. Purely from a technology perspective, AMD is looking to be 1 or 2 nodes behind TSMC bleeding edge for at least the next 3 years. (ex: Zen 4 will be a TSMC 5nm product 1 full year after Apple is already shipping 3nm products).

The long-term threats are many, including the fundamental increases in chip fab and production costs that require increasing revenue and profits for funding those node shrinks (regularly) to make sense. Re-deploying Intel foundry services in almost any capacity can serve this master: Intel can re-use older fabs even longer than are sustainable on their core markets for products that don't need the latest technology -- meaning more revenue from fully depreciated assets, or reduced costs for any "lower volume" Intel chips that can leverage older fabs. Alternative, If Intel generates foundry business on bleeding edge nodes then they'll have more money to develop those nodes (the same way Apple is funding TSMC), and be effectively denying revenue to competitors in the overall silicon space.

Intel re-entering the foundry market will/could enable them to run to the very end of scaling on Silicon (alongside TSMC) which x86 alone* might not be able to financially support.

*Especially under attack from a resurgent AMD using Apple-funded TSMC cutting edge fabs.

I believe another immediate threat to Intel is Apple, Amazon, Google, Microsoft, Oracle, and many more found out that they don't have to wait for Intel's new CPU in order to decide their own product and service rollouts.

Actually they recognized that their own semiconductor ventures can differentiate themselves from competitors. Apple's M1, Amazon's Graviton, Google's Tensor Processing Unit, and Ampere's Arm server chips all prove just that.

IMHO, AMD's new capability is indeed a serious challenge to Intel's revenue and profit. But the in-house semiconductor projects and the fabless model adopted by Intel's customers are threatening Intel's survival.

 

[tooLongInEDA]

Respectfully, Some counterarguments for consideration --
...
*Especially under attack from a resurgent AMD using Apple-funded TSMC cutting edge fabs.

That's really "Apple customer funded" - i.e. iPhone users !

More seriously, it's a very good point you make ("The article and leaks so far don't state any explicit goals for which manufacturing processes, how much market share, or how much $ total business Pat wants Intel to aim for with respect to re-entering the foundry market."). What exactly is the business plan to justify the level of investment and public subsidies being discussed here ? The same question we should ask of the politicians in the US and EU.

Another major point is whether Intel and make a go of foundry by overcoming the organizational and cultural challenges this will bring. Can they adapt to being a #2 or #3 market player after so long as undisputed #1 ? Will enough top people want to take the risk of working in the foundry business when there's a higher status, lower risk option in server/datacenter/cpu ? Without making this an almost independent business, it's not clear this - or the conflicts of interest inherent with Intel as both foundry and competitor in some markets - are easily overcome.

 

[count]

Respectfully, Some counterarguments for consideration --

The article and leaks so far don't state any explicit goals for which manufacturing processes, how much market share, or how much $ total business Pat wants Intel to aim for with respect to re-entering the foundry market.

If Intel's primary goal is to ensure x86 dominance (and Intel market share dominance) via recapturing the lead for fab technology, I think the foundry play is fully complementary.

IMO, The near-term threats to Intel are primarily the resurgent AMD, and in Intel not being positioned well enough to gain stable marketshare in newer markets such as AI and ML. Purely from a technology perspective, AMD is looking to be 1 or 2 nodes behind TSMC bleeding edge for at least the next 3 years. (ex: Zen 4 will be a TSMC 5nm product 1 full year after Apple is already shipping 3nm products).

The long-term threats are many, including the fundamental increases in chip fab and production costs that require increasing revenue and profits for funding those node shrinks (regularly) to make sense. Re-deploying Intel foundry services in almost any capacity can serve this master: Intel can re-use older fabs even longer than are sustainable on their core markets for products that don't need the latest technology -- meaning more revenue from fully depreciated assets, or reduced costs for any "lower volume" Intel chips that can leverage older fabs. Alternative, If Intel generates foundry business on bleeding edge nodes then they'll have more money to develop those nodes (the same way Apple is funding TSMC), and be effectively denying revenue to competitors in the overall silicon space.

Intel re-entering the foundry market will/could enable them to run to the very end of scaling on Silicon (alongside TSMC) which x86 alone* might not be able to financially support.

*Especially under attack from a resurgent AMD using Apple-funded TSMC cutting edge fabs.

From this, I agree that the root of Intel's problems is that their core markets (x86 in PCs and datacenters) will not generate the volumes required for continued scaling and there are only 2 viable solutions.

1. Create a foundry business and hope to bring in the volumes required
2. Go fabless

Intel attempted option 1 before and failed miserably and I think they will fail again on this second attempt. The reasons are culture, incentives, and competition.

The foundry business requires a really cooperative, trusting culture, and Intel's culture is all about monopolization and paranoia. They want to be the king, are ultra paranoid of any threats to their monopoly, and have a lingering "not invented here" attitude, all of which is not conductive to the types of collaborative relationships required to be successful as a foundry. You see this in them taking industry accepted terms and standards and them trying to rebrand them as if they were something Intel exclusive (Intel Tiles vs chiplets, Intel Tri-Gate vs Fin Fet). You can also see how when they approached 3D Xpoint, which was a promising technology, they effectively crippled by locking it to Intel sockets. You can't pull stunts like this as a foundry and still have credibility with your customers.

Intel has every incentive not to be a good foundry partner, as they are directly or indirectly competing with many of their potential customers. The three largest foundry customers are Apple, NVidia, and AMD. With Apple now producing it's own chips and NVidia and AMD direct competitors they have 0 chance at winning any of those customers. And even if they did... Intel isn't going to want to make AMD or NVidia better.

Finally, I just don't think Intel has what it takes to compete with TSMC head on.

So that only leaves option 2. Intel needs to go fabless.

 

[Lorien]

According to this logic then we should do nothing because there is nothing to be done. I simply can't disagree more.
US is playing too nice and gentlemen in this super competitive game (actually it is a zero-sum game because winner takes it all), self-imposing rules and self-tying its onw hands. If TSMC 60% (I think maybe even higher today) is US customers, we should have pushed TSMC to open fabs in the US long ago. If it were China, they would have been already been forced to hand over the IP, or even have their fabs confiscated. Look at how aggressive the small country of South Korea government and companies are together investing in their semi tech future. How could we do nothing? Previously Intel was the pioneer for the EUV, many years ahead of TSMC or Samsung. It still has the technological capacity (and money) to catch up. We should not lose faith in the American spirit. IFS definitely won't make money in the short run, it going to take time. TSMC was not built in 2 years.

What's with the patriotic American "can-do"-ism? US "playing too nice"? When has the US ever played to nice when something serves its national interest?

You talk like the US actually does anything innovative for the sake of innovation, when it's really just all about money.The reason the US fell behind in semis is the same reason it fell behind on manufacturing in general - deregulated capitalism with no central planning. Companies move production offshore to cut cost and increase profit margins. Moving production back do domestic shores will only happen if it makes sense financially - some subsidies from the Feds is nice but ultimately just a band-aid compared to what China is able to do with their industry planning.

As for Taiwan and South Korea, they are just too far ahead by this point for any other country to seriously catch up. And that situation won't change for as long as big American tech companies continue to rely on them for their tech leadership.

Capitalism is a an ouroboro. We reap what we sow.

The foundry business requires a really cooperative, trusting culture, and Intel's culture is all about monopolization and paranoia. They want to be the king, are ultra paranoid of any threats to their monopoly, and have a lingering "not invented here" attitude, all of which is not conductive to the types of collaborative relationships required to be successful as a foundry. You see this in them taking industry accepted terms and standards and them trying to rebrand them as if they were something Intel exclusive (Intel Tiles vs chiplets, Intel Tri-Gate vs Fin Fet). You can also see how when they approached 3D Xpoint, which was a promising technology, they effectively crippled by locking it to Intel sockets. You can't pull stunts like this as a foundry and still have credibility with your customers.

This exactly. Intel embodies the textbook American corporate culture. For a long time after WWII America has enjoyed a financial & industry monopoly on international trade because, well, most other industrial countries were bombed into smoking ruins. Times have changed and we no longer have tech & industrial leadership. Intel's business model simply can't compete anymore. To truly innovate they will need to open up and allow ideas and technologies to flow, instead of guarding their secrets like some cultish cabal.