Intel's Conflict of Interest? Can it overcome TSMC

[Arthur Hanson]

No matter how Intel changes its corporate structure, there is still an inherent conflict of interest unless the split into two separate companies or divisions with a wall between them. I have rarely seen a structure like this succeed, but it has happened, but rarely. TSM has mastered their model and nurtured it since inception with the ability to service differing customers without conflict. This is a skill set TSM has mastered for decades not years. Already by counting internal business as part of their foundry business they have broken down a wall that needs to be maintained if they are going to succeed. They have built in to their model this flaw from the beginning. It is not too late to correct it. I hope they take appropriate action and soon. Any thoughts, comments or additions sought and appreciated. As a full time investor, I see this model having flaws from the start, but it's early in the game and not too late to correct.

 

[Tech101]

When Intel dominated both chip designs (CPUs etc) and manufacturing (foundry) with "Intel Inside", the IDM 1.0 was a big success quarter century ago. Howevr, situation has changed. Pat Gelsinger is smart enough, and now Intel is in the right direction in preparation to split into two separate companies with a wall to avoid the conflict of interest.
Intel is formidable and could regain the domination position in 4 years with the support of CHIPS Act.

 

[hist78]

When Intel dominated both chip designs (CPUs etc) and manufacturing (foundry) with "Intel Inside", the IDM 1.0 was a big success quarter century ago. Howevr, situation has changed. Pat Gelsinger is smart enough, and now Intel is in the right direction in preparation to split into two separate companies with a wall to avoid the conflict of interest.
Intel is formidable and could regain the domination position in 4 years with the support of CHIPS Act.

But TSMC and Samsung are also having the support of Chips Act and local government's support.

 

[Daniel Nenni]

It is simple really. The pure-play foundry model is preferred by all fabless companies. If GF or UMC had leading edge processes IDM foundries would not be needed. But TSMC is the only pure-play foundry at the leading edge so here we are, IDM foundries are in the thick of things and customers will line up. If Intel or Samsung can compete with TSMC at he process level it would be an exciting three horse race. Today it is just TSMC and who wants to watch a one horse race?

Money is not the issue, it is technology. If Intel can get back on track, which from what I see they are, then we have a two horse race. Samsung is going to have to make some pretty big structural changes inside to get back in the race. Personally I hope they do but I'm not sure how that is going to happen.

My concern about Intel is that expectations are set pretty high. My suggestion is always speak softly and carry a big stick. That builds trust. Unfortunately Wall Street is on Intel's back so managing expectations is very hard.

 

[Mooredaddy]

When Intel dominated both chip designs (CPUs etc) and manufacturing (foundry) with "Intel Inside", the IDM 1.0 was a big success quarter century ago. Howevr, situation has changed. Pat Gelsinger is smart enough, and now Intel is in the right direction in preparation to split into two separate companies with a wall to avoid the conflict of interest.
Intel is formidable and could regain the domination position in 4 years with the support of CHIPS Act.

The money from the chips act is way over emphasized. It’s $52 billion spread over many years and a fraction is going to manufacturers never mind Intel specifically. It helps but nothing that will change any individual companies trajectory.

 

[hist78]

The money from the chips act is way over emphasized. It’s $52 billion spread over many years and a fraction is going to manufacturers never mind Intel specifically. It helps but nothing that will change any individual companies trajectory.

The CHIPS Act and Science Act 2022 provides US$52.7 billion in federal funding to revitalize the U.S. semiconductor industry. It's including US$39 billion in semiconductor manufacturing, US$13.2 billion in R&D and workforce development, and US$500 million to strengthen global supply chains.

Let's assume 8 major companies will build new fabs with federal money from the $39 billion manufacturing grant pool: :

Wolfspeed: $2 billion
SK Hynix: $3 billion
Globalfoundries: $4 billion
Texas Instruments: $4 billion
Samsung: $5 billion
Micron: $5 billion
TSMC: $6 billion
Intel: $7 billion
All other smaller semiconductor companies: $3 billion

We can debate the amount I guessed on each company is too much or too little. But one thing for sure is that the impact on each recipient is much smaller than the impression of the $52 billion total budget.

 

[Tanj]

The other consideration of one company vs. two is the business model. Capital intensive manufacturing, people and IP intensive design. They should be listening to entirely different sets of customers. They have different business cycles. The design side would benefit from diverse choices in manufaccturing. The manufacturing side needs multiple, diverse design customers to achieve scale and smooth out business cycles.

The only reason they are joined at the hip is because of history, including that exceptional period from 1985 to 2010 where their fab powered their designs ahead of all others, to an apex of a 3 year lead around 2005 for the most valuable PC and server markets. But the end of the Dennard era was eroding their lead even before their foundry stumbled, and the tick-tock model amplified instead of smoothed the business cycle. They are never going to have that exceptional advantage again.

The best case is clearly to split into separate companies as soon as they have the foundry model credible with other customers. The recent IFM Update showed them deepening the separation, which will be necessary to get to launch speed, even if some of the claimed cost savings prove illusory it is the structural changes which will be important.

 

[Daniel Nenni]

The money from the chips act is way over emphasized. It’s $52 billion spread over many years and a fraction is going to manufacturers never mind Intel specifically. It helps but nothing that will change any individual companies trajectory.

The more I talk to the CHIPs Act people the more encouraged I am. It's not just the money, it's a badge of honor to get funding from the US government and more funding will be coming, my opinion. There are some well placed semiconductor professionals involved with CHIPs Act so it isn't just politicians doling this money out to support their re-election campaigns.

Best case: Intel, TSMC, Micron, and Samsung get money to recognize and support their manufacturing efforts in the United States, absolutely.

 

[hskuo]

 

[Fred Chen]

No matter how Intel changes its corporate structure, there is still an inherent conflict of interest unless the split into two separate companies or divisions with a wall between them. I have rarely seen a structure like this succeed, but it has happened, but rarely. TSM has mastered their model and nurtured it since inception with the ability to service differing customers without conflict. This is a skill set TSM has mastered for decades not years. Already by counting internal business as part of their foundry business they have broken down a wall that needs to be maintained if they are going to succeed. They have built in to their model this flaw from the beginning. It is not too late to correct it. I hope they take appropriate action and soon. Any thoughts, comments or additions sought and appreciated. As a full time investor, I see this model having flaws from the start, but it's early in the game and not too late to correct.

Can Intel do all its chiplets by itself? Seems not right now, this highlights Intel's foundry competing with TSMC.

 

[Hunterdolphin]

When Intel dominated both chip designs (CPUs etc) and manufacturing (foundry) with "Intel Inside", the IDM 1.0 was a big success quarter century ago. Howevr, situation has changed. Pat Gelsinger is smart enough, and now Intel is in the right direction in preparation to split into two separate companies with a wall to avoid the conflict of interest.
Intel is formidable and could regain the domination position in 4 years with the support of CHIPS Act.

Yer, I agree. I personally always thought that Intel would work better as a conglomerate within this current industry and geopolitical environment.

 

[Fred Chen]

True separation means IFS changes its name to not include Intel, also get some of its own fabs from Intel.

 

[Nokokara]

It is simple really. The pure-play foundry model is preferred by all fabless companies. If GF or UMC had leading edge processes IDM foundries would not be needed. But TSMC is the only pure-play foundry at the leading edge so here we are, IDM foundries are in the thick of things and customers will line up. If Intel or Samsung can compete with TSMC at he process level it would be an exciting three horse race. Today it is just TSMC and who wants to watch a one horse race?

Money is not the issue, it is technology. If Intel can get back on track, which from what I see they are, then we have a two horse race. Samsung is going to have to make some pretty big structural changes inside to get back in the race. Personally I hope they do but I'm not sure how that is going to happen.

My concern about Intel is that expectations are set pretty high. My suggestion is always speak softly and carry a big stick. That builds trust. Unfortunately Wall Street is on Intel's back so managing expectations is very hard.

Yes, indeed it is quite right to take all that into account to have a better/more correct vision of the semiconductor foundries evolution, including IDM’s ones as
(i) it is important for the foundry leaders to be providing most advanced technologies with good performance and yield…
(ii) but fabless companies will always whenever possible prefer to use pure-play foundries rather than IDM which could then internally duplicate some component design usage/solution…

Given all these elements, the IDM foundries are not quickly nor strongly progressing and will/should not become the biggest foundry above TSMC who is now already above Intel and Samsung global semiconductor business in term of revenue in 2022 (TSMC 75.88 B$ , Samsung 65.59 B$, Intel 63.05 B$) as,
(i) Samsung the IDM who created in 2005 their “foundry” with the objective/dream to become soon the first/biggest foundry which they finally not yet succeed after so many years, both because they often had low yield on last advanced technologies and many fabless preferring other pure-play foundries…
(ii) It will then be very similar for Intel who had same expectation/dream in their more recent IDM 2.0 business model including the evolution within foundry (which is still very small in term of revenue, only 895 M$ in 2022…)…

 

[Daniel Nenni]

Yes, indeed it is quite right to take all that into account to have a better/more correct vision of the semiconductor foundries evolution, including IDM’s ones as
(i) it is important for the foundry leaders to be providing most advanced technologies with good performance and yield…
(ii) but fabless companies will always whenever possible prefer to use pure-play foundries rather than IDM which could then internally duplicate some component design usage/solution…

Given all these elements, the IDM foundries are not quickly nor strongly progressing and will/should not become the biggest foundry above TSMC who is now already above Intel and Samsung global semiconductor business in term of revenue in 2022 (TSMC 75.88 B$ , Samsung 65.59 B$, Intel 63.05 B$) as,
(i) Samsung the IDM who created in 2005 their “foundry” with the objective/dream to become soon the first/biggest foundry which they finally not yet succeed after so many years, both because they often had low yield on last advanced technologies and many fabless preferring other pure-play foundries…
(ii) It will then be very similar for Intel who had same expectation/dream in their more recent IDM 2.0 business model including the evolution within foundry (which is still very small in term of revenue, only 895 M$ in 2022…)…

Samsung Foundry revenue in 2022 is closer to $20B I believe, not $65.59B. The Samsung semiconductor revenue which includes memory is $65.59B. TSMC 2022 is $73.67B and Intel reported $63.05B. So now under Intel’s new foundry reporting it is a two horse race.

 

[delong.height]

Yes, indeed it is quite right to take all that into account to have a better/more correct vision of the semiconductor foundries evolution, including IDM’s ones as
(i) it is important for the foundry leaders to be providing most advanced technologies with good performance and yield…
(ii) but fabless companies will always whenever possible prefer to use pure-play foundries rather than IDM which could then internally duplicate some component design usage/solution…

Given all these elements, the IDM foundries are not quickly nor strongly progressing and will/should not become the biggest foundry above TSMC who is now already above Intel and Samsung global semiconductor business in term of revenue in 2022 (TSMC 75.88 B$ , Samsung 65.59 B$, Intel 63.05 B$) as,
(i) Samsung the IDM who created in 2005 their “foundry” with the objective/dream to become soon the first/biggest foundry which they finally not yet succeed after so many years, both because they often had low yield on last advanced technologies and many fabless preferring other pure-play foundries…
(ii) It will then be very similar for Intel who had same expectation/dream in their more recent IDM 2.0 business model including the evolution within foundry (which is still very small in term of revenue, only 895 M$ in 2022…)…

I always thought it is best to keep Intel as this type of conglomerate, with design and manufacturing capability. I don't think IFS being spin off from Intel will help the former or the latter in any way or form. Sure, it can gain some attention from other fabless companies. But the nature of IFS, in which it's branded as the only western foundry since inception, is going to be more costly to operate compared to TSMC and Samsung. To match the price tag from TSMC, they must either lower their price tag or shrinking down their capacity. As a result, it's going to have less profit due to a limitation on both revenue and profit margin.

It may be okay to compete with TSMC temporarily after IFS being spin off from Intel. But that does only harm to both organizations (business units and foundry). As the business units have no constraints or desires from the top that's asking them to put some of their wafer to be manufactured in house. Instead, the business nature of enabling greater profit margins will drive these business units to allocate majority or all of their supply to be manufactured at TSMC because they are cheaper to produce due to economic of scales and the fact it's based in Asia.

Now consider what that does to the future, long after IFS being spin off, IFS will be permanently weakened. And if the capex required and R&D required for more leading-edge process node or packaging keeps increasing, which it was given the past ~50 years. I see no future in IFS or Intel anymore.


(ADD: While I believe spin off is still possible, it's only likely that Intel can do a Mobileye-like spin off)

 

[blueone]

(ii) but fabless companies will always whenever possible prefer to use pure-play foundries rather than IDM which could then internally duplicate some component design usage/solution…

While I think dedicated fab companies are a better idea than IDMs, my reasoning is that it is nearly impossible to find a single CEO capable of effectively leading both design and fabrication organizations. I think Intel has proven this for decades, and so did IBM.

As for the hypothesis that Intel would steal design IP from a fab customer, I'm having trouble believing it. The most likely scenario would probably be diverting some of the fab customer's early samples to Intel's competitive analysis lab. There would be several, perhaps numerous, engineers and managers who would need to be involved in the conspiracy. If you really wanted to assess design features you're going to need to put the chip into a functional test system and perhaps need special microcode, firmware, or software to do it. Any one of these Intel employees could decide to be a whistleblower. The blow to IFS's reputation would be fatal, should the diversion be revealed. The entire scenario could easily result in a huge disaster for Intel and its stockholders. The risk-award ratio looks very high indeed.

I just don't believe it.

 

[nghanayem]

While I think dedicated fab companies are a better idea than IDMs, my reasoning is that it is nearly impossible to find a single CEO capable of effectively leading both design and fabrication organizations. I think Intel has proven this for decades, and so did IBM.

As for the hypothesis that Intel would steal design IP from a fab customer, I'm having trouble believing it. The most likely scenario would probably be diverting some of the fab customer's early samples to Intel's competitive analysis lab. There would be several, perhaps numerous, engineers and managers who would need to be involved in the conspiracy. If you really wanted to assess design features you're going to need to put the chip into a functional test system and perhaps need special microcode, firmware, or software to do it. Any one of these Intel employees could decide to be a whistleblower. The blow to IFS's reputation would be fatal, should the diversion be revealed. The entire scenario could easily result in a huge disaster for Intel and its stockholders. The risk-award ratio looks very high indeed.

I just don't believe it.

To add on, Apple (that company that is renowned for never forgiving or forgetting) went back to Samsung for second sourcing a mere two years after the alleged IP theft had occurred. If Samsung can still get major business from their competitors after that, and intel doesn’t have this kind of black mark on their record, I don’t really see people not using IFS due to IP concerns (assuming intel has shown they have put in place sufficient firewalls for their customers).

(ii) It will then be very similar for Intel who had same expectation/dream in their more recent IDM 2.0 business model including the evolution within foundry (which is still very small in term of revenue, only 895 M$ in 2022…)…

And where pray tell did intel say they “planned to be the largest foundry”? The messaging they always gave was number two. Dave also said something to the tune of:

“…funnily enough our internal business kind of makes us number 2. But the goal is to grow external to make us number 2…”.

Once again paraphrasing off of memory since I didn’t feel like digging through the transcript.

 

[blueone]

To add on, Apple (that company that is renowned for never forgiving or forgetting) went back to Samsung for second sourcing a mere two years after the alleged IP theft had occurred. If Samsung can still get major business from their competitors after that, and intel doesn’t have this kind of black mark on their record, I don’t really see people not using IFS due to IP concerns (assuming intel has shown they have put in place sufficient firewalls for their customers).

Correct me if I'm wrong, but as far as I can tell, none of the Apple infringement claims against Samsung were related to chip design or fabrication. All I remember and have found in a few internet searches related to physical product design, user interface designs, and various physical phone and tablet attributes. Like, for example, "swipe to unlock" functionality on touch screens.

The thesis on this forum relates to chip design, which seems like much more of a stretch, for Intel to derive value from examining or testing IFS customer sample chips. I remember seeing the use of X-Ray or electron microscopy imaging to differentiate between, say, custom logic and a structured ASIC, which can be illuminating, but I haven't heard it's possible to interpret circuit-level logic as it relates to functionality. Is it? In my engineering past I sometimes questioned the validity of circuitry patents if there was no known test for infringement. The answers from the patent attorneys were essentially, "A patent is a legal document, and you are not qualified to judge one.", and that a large patent portfolio is more important than almost any individual patent.

 

[Daniel Nenni]

Correct me if I'm wrong, but as far as I can tell, none of the Apple infringement claims against Samsung were related to chip design or fabrication. All I remember and have found in a few internet searches related to physical product design, user interface designs, and various physical phone and tablet attributes. Like, for example, "swipe to unlock" functionality on touch screens.

The thesis on this forum relates to chip design, which seems like much more of a stretch, for Intel to derive value from examining or testing IFS customer sample chips. I remember seeing the use of X-Ray or electron microscopy imaging to differentiate between, say, custom logic and a structured ASIC, which can be illuminating, but I haven't heard it's possible to interpret circuit-level logic as it relates to functionality. Is it? In my engineering past I sometimes questioned the validity of circuitry patents if there was no known test for infringement. The answers from the patent attorneys were essentially, "A patent is a legal document, and you are not qualified to judge one.", and that a large patent portfolio is more important than almost any individual patent.

I believe there was some IP involved but the point is that Samsung, Apples partner/source for the iPhone, directly competed with Apple at the product level. Had Apple known Samsung would do that do you think that partnership would have gone forward? Absolutely not. In fact, Apple does its best to NOT include Samsung products in their iProduct family BOM to this day. Apple's first partner choice for the SoC was Intel but we all know how that turned out.

When IDMs first started to manufacture outside designs back in the 1980s customers were very nervous about IP and available capacity. Xilinx was one of the first fabless companies to use an IDM for manufacturing. They quickly moved to a pure play foundry (UMC) to avoid risk of losing capacity to IDM internal products. They also got better pricing. That is the history of IDM foundries.

Will AMD use Intel IFS? I don't think so and it is not necessarily an IP security issue but it is certainly a competitive issue. Will Nvidia use IFS for GPUs that are competitive with Intel products? I would be surprised if they do, especially since Intel uses TSMC for their GPUs.

There are plenty of other fabless companies that do not compete with Intel and that is where the IFS market is, my opinion, and it is a very big and growing market.

The one thing that I would encourage IFS to do is to aggressively chase the ASIC business. Samsung does ASICs, TSMC has GUC and a large collection of ASIC partners. Broadcom, MediaTek and Marvell are in the ASIC business amongst many others. The ASIC business is back and growing and a big part of the foundry ecosystem, absolutely.

 

[blueone]

I believe there was some IP involved but the point is that Samsung, Apples partner/source for the iPhone, directly competed with Apple at the product level. Had Apple known Samsung would do that do you think that partnership would have gone forward? Absolutely not. In fact, Apple does its best to NOT include Samsung products in their iProduct family BOM to this day. Apple's first partner choice for the SoC was Intel but we all know how that turned out.

I searched a bit more, and all of the patents Apple tried to defend appear to be so-called "design patents" and those which describe user interface features.

When IDMs first started to manufacture outside designs back in the 1980s customers were very nervous about IP and available capacity. Xilinx was one of the first fabless companies to use an IDM for manufacturing. They quickly moved to a pure play foundry (UMC) to avoid risk of losing capacity to IDM internal products. They also got better pricing. That is the history of IDM foundries.

I believe the available capacity concern could be valid for IFS customers, using any process involving EUV equipment. I suspect for the foreseeable future IFS pricing will be very competitive, but that seems pretty obvious.

Will AMD use Intel IFS? I don't think so and it is not necessarily an IP security issue but it is certainly a competitive issue. Will Nvidia use IFS for GPUs that are competitive with Intel products? I would be surprised if they do, especially since Intel uses TSMC for their GPUs.

FWIW, I'm not convinced this will necessarily be the case, since Intel would be the risk-reduction second source. If TSMC's geopolitical risk went away, I don't think Intel would get any business from Nvidia or AMD.

There are plenty of other fabless companies that do not compete with Intel and that is where the IFS market is, my opinion, and it is a very big and growing market.

Agreed.

The one thing that I would encourage IFS to do is to aggressively chase the ASIC business. Samsung does ASICs, TSMC has GUC and a large collection of ASIC partners. Broadcom, MediaTek and Marvell are in the ASIC business amongst many others. The ASIC business is back and growing and a big part of the foundry ecosystem, absolutely.

Agreed.