Intel Weighs Options Including Foundry Split to Stem Losses

[Xebec]

Did people really think that Intel would already be booming in business in 2024?

From Pat's 4 year plan, this was always going to be the hardest financial year. The transition from old products and old fab to new tech. Intel 7/10nm is old, Intel 4 and 3 are ramping hard in 2024, and 20A/18A which were planned to deliver class leading products are still a year away from any kind of real volume. Other macro factors are sort of obvious too - no major new Windows releases, everyone replaced hardware in 2020-2021 and PC cycles are longer, etc. Intel misfired hard on GPUs but we can thank Raja and some of the team for that, not Pat..

If the plug gets pulled this year on Intel's turn around plan then we'll never know if it would have worked.

(It seems like it took the removal of dividends to make investors and pundits realize that, yes, Intel really was/is in bad shape and the turn around is going to be an expensive gamble.)

 

[osnium]

It’s really very simple where Intel is.

As noted by another CCG is a 200Million unit whale as large as Apple’s and bigger than Qualcomm or MediaTek. They need leading edge and beginning next year will have it again. That can/is billions++ quarter of business to some Foundry.

Intel was/is known as a silicon manufacturer and x86, some have forgotten its legacy in memories but was defining there to.

So the decisions is be a product company that is shrinking ? Let’s look at that closely and see the prognosis. We already have seen without a monopoly they have been less than successful maybe I’d call them a failure to leverage anything but the monopoly. Try as they might they simply aren’t competitive in anything outside of x86. Something very wrong in the DNA of that side. Was it decades of only little AMD and process leadership? Clearly AMD figured out how to innovate and beat Chipzilla decades ago with inferior process. The product team has shown no capability to innovate on lagging nor leadership technology, just follow the trend and leverage tech long or power. I am sure Jim Keller had audacious goals and as such was removed. As a product company they missed every big pivot and I expect them to continue to fail here without and outsider and no reason any visionary outsider wants this! Only solution must be on a node ahead of AMD, too expensive outside, let them feed the IFS and be the Apple to IFS.

Manufacturing, I’d argue that they have the core talent and DNA to get back to leaderships and stay there or lead. Remember all recent innovation in logic has come from them including the most audacious pivots in history, mostly, LOL

Their manufacturing was is not bad, not the most cost efficient but in the past did do okay as a high cost leaderships manufacturing in some of the most high cost geos. First thing a 20-30% cut in manufacturing is a must to put them at headcount parity and than the structure could follow. The layoffs are to small across the board, cut and force the organization to adapt is better than the stupid Bain and McKinsey consultants. I’ve seen their results, useless.

Pat has the right top strategy, but sadly he is the wrong CEO. Keep the strategy but lose most BoD and senior leadership. The company has the talent and business model, sadly leadership is busted at the top.

In your opinion, slowing down the fab expansions is not abandoning the strategy, correct?

After these layoffs, do you think there will be key talent left in Oregon’s TD fab to pursue technology development in a successful manner. You can argue that this crew has been the root cause of problems in the past 10 years.

 

[ai268]

Whether to split or not is a huge strategic premise / decisions.

To review and reflect on this.

Leading edge process development requires a decade ahead exploratory work, pathfinding and the final feature definition, then development. During this you need to nurture/cowork EDA, Tool and fabless ( internal products). Given the timeline and cost you need huge scale to get necessary ramp learning. Also must have huge outlay for a decade.

There is no shortcut to delivering mature and reliable robust leading edge process, you need a high volume product that can leverages leading edge and has volume to partially or fully fund the technology cadence. Clearly Intel x86 can no longer due it exclusively, just like IBM, DEC, Motorola and AMD/GF failed. Only if 10nm and Intel4/3 came on time and all the Intel products had stayed in house for the computed, graphics and base die it could have been different. Since they all went out you have a half decade of darkness before they come back in to keep IFS head above water.

It can and should work. Intel is unique in having volume hundreds of millions of units like that they can be that customer. One could argue Nvidia is almost like IBM or DEC like 30+ years ago to do this, or the systems company like MSFT, Apple, Amazon, Google could entertain, but they all have a go to supplier that continues to have leadership and scale and at least on the surface delivers that there is no reason to consider doing anything till IFS delivers parity to superior process.

This is a catch22, to unleash and separate CCG and DCAI would free them to pick TSMC. CCG with its volume can and did pay to secure capacity and we will see with Lunar Lake that. Clearly to make IFS work it needs a volume client and the only one that could/can be forced to choose it would be CCG.

To separate them, it’s a no brainer the right thing for CCG is to secure volume A16 after doing products on N3B and N3P. If they went that route where would that leave IFS? This time there is no sugar daddy and doubt the Us government nor Oil state would bankroll the billions in development and tens of billions in fab capex to make IFS successfully.

The only question is how to fund this joint IDM2.0 or really if you split and continue it’s nothing but a disingenuous play as the fabless side still isn’t free and just a shell game, someone still has to pay for the IFS side.

It was always going to be a ten year plan and the onset now with the slowing CCG and DCAI the full situation is just exposed more bare.

Sir, Since your mentioned DEC, as a Digital employee almost 35 years ago, can I "copy/paste" your article, with reference, to the DEC FB group for further discussion about DEC then vs Intel now?

 

[Brady]

Did people really think that Intel would already be booming in business in 2024?

From Pat's 4 year plan, this was always going to be the hardest financial year. The transition from old products and old fab to new tech. Intel 7/10nm is old, Intel 4 and 3 are ramping hard in 2024, and 20A/18A which were planned to deliver class leading products are still a year away from any kind of real volume. Other macro factors are sort of obvious too - no major new Windows releases, everyone replaced hardware in 2020-2021 and PC cycles are longer, etc. Intel misfired hard on GPUs but we can thank Raja and some of the team for that, not Pat..

If the plug gets pulled this year on Intel's turn around plan then we'll never know if it would have worked.

(It seems like it took the removal of dividends to make investors and pundits realize that, yes, Intel really was/is in bad shape and the turn around is going to be an expensive gamble.)

I think this hits is squarely on the head. 2024/25 was always going to be a bloodbath. The horizon for getting process leadership back was 2024/25 but that’s just the very beginning of the bottom. Then they have to pull their own leading product designs back onto internal leading processes (Panther Lake starts that), and win some external customers (this is going more slowly than anyone would have hoped).

Now, absolutely Pat and leadership were overly optimistic and out of touch on customer desire for their products, likely significantly because of the seismic shift caused by GenAI. But, again: 2024/25 was never going to be roses for Intel, neither Product nor Foundry.

So mostly you either still believe this strategy will work, or you still think it will never work and the company must be split to save either side.

All I know is I see a lot of banker and investment types circling Intel like vultures (again, as they did in 2021) and that is very bad for an innovation company critical to America. Very bad.

 

[ai268]

to deal with TSMC monopoly another option is unleash it's Chinese rivals. client could easily see a 50% drop in cost.

intc is dead already, other wise they might have different opinions.

Is this a joke? You forgot about the "mask" problem during 2019 Covid. Spokesman of China foreign ministry: "No Soup (mask) for you"

 

[osnium]

Is this a joke? You forgot about the "mask" problem during 2019 Covid. Spokesman of China foreign ministry: "No Soup (mask) for you"

If the US is incapable of sustaining manufacturing due to societal problems and economic problems, the best way might be to relax all silicon technology restrictions on China by allowing ASML to sell EUV to China and allowingIntel to sell/transfer silicon technology to SMIC. Hopefully that would create some level of goodwill between the US and China.

 

[ai268]

If the US is incapable of sustaining manufacturing due to societal problems and economic problems, the best way might be to relax all silicon technology restrictions on China by allowing ASML to sell EUV to China and allowingIntel to sell/transfer silicon technology to SMIC. Hopefully that would create some level of goodwill between the US and China.

Drink poison to quench your thirst.​

 

[blueone]

If the US is incapable of sustaining manufacturing due to societal problems and economic problems

On a per capita basis, the US is a more prolific manufacturer than China:

Top 10 Manufacturing Countries in the World in 2024

Considering some supply chain options? Here are the top 10 manufacturing countries in the world in 2024.

www.safeguardglobal.com

China has a population of 1.2 billion, while the US has a population of about 330 million. Yet, the US produces 15.9% of the world's manufacturing output, and China 31.6%. And China's system over-produces many products for political reasons, like solar panels, cars, and steel, at huge losses to keep their population employed, which is not practical in the US political system (except for certain agricultural products, but let's not go there).

The US "problems" are that it has a market economy and is mostly a democracy (the US is actually a constitutional federal republic, but let's keep it simple), the former driving economic efficiency and demands for highest returns on investment, while the latter decreases efficiency in a lot of ways compared to autocracies. This combination makes it more complicated for the US government to provide support or favorable regulations for specific industries, so the US isn't very good at it. If you have another country that is more motivated to support and subsidize a local industry, the US has historically not been successful in holding off subsidized competition, or products produced in countries with very low labor costs and weak or non-existent environmental regulations. A well-functioning market economy will tend to want to take advantage of lower cost sources, and even with US subsidies like the CHIPS Act, tariffs or import restrictions are often required to defend US-local manufacturing.

 

[BruceA]

In your opinion, slowing down the fab expansions is not abandoning the strategy, correct?

After these layoffs, do you think there will be key talent left in Oregon’s TD fab to pursue technology development in a successful manner. You can argue that this crew has been the root cause of problems in the past 10 years.

Foundry should build capacity to order. Pat was ridiculous in his expansion talk. There simply was no reason to believe that much demand was coming on the timeline he talked. In many ways the BoD has some responsibility in controlling the CEO. Here the BoD is lacking, sad.

Given then efficiency need a layoff and retirement of people would simply be clearing of the dead wood and even if 15% of TD left I’d say they could still deliver. Actually the more that leave maybe the better, only keep the committed, everyone else can leave.

 

[BruceA]

If the US is incapable of sustaining manufacturing due to societal problems and economic problems, the best way might be to relax all silicon technology restrictions on China by allowing ASML to sell EUV to China and allowingIntel to sell/transfer silicon technology to SMIC. Hopefully that would create some level of goodwill between the US and China.

Maybe create a real competitive situation. Nothing motivates people better than fair competition and nothing like unfair to start wars.

 

[blueone]

Maybe create a real competitive situation. Nothing motivates people better than fair competition and nothing like unfair to start wars.

What in your view would a real competitive situation look like in semi fabrication?

 

[blueone]

Foundry should build capacity to order. Pat was ridiculous in his expansion talk. There simply was no reason to believe that much demand was coming on the timeline he talked. In many ways the BoD has some responsibility in controlling the CEO. Here the BoD is lacking, sad.

I think your expectation for the BoD, pretty much any BoD in my experience, is overly optimistic. Intel does have a more balanced board than the BoD structure than a lot of companies, because the Chairperson of the Board is a different person than the CEO. A lot of companies have a structure such that the CEO and Chair are the same person. For example, in AMD Lisa Su is Chairperson and CEO. So is Satya Nadella at Microsoft, just to name another familiar company. At Intel, the CEO technically reports to the Chair, who is Frank Yeary.

That said, if you're thinking that BoD members are going to be adversaries to the CEO's agenda within a short timeline, that doesn't happen normally in my experience. For one thing, in the US BoD members are often not subject matter experts in product or market matters of the company. And this BoD chose PG apparently because he had a plan to reinvigorate manufacturing. If PG really screws things up they may vote to replace him, but that's different than getting involved in business management issues.

 

[jim_22222]

What in your view would a real competitive situation look like in semi fabrication?

first thing is within a US semi company, if there is a Mark liu showing great potential to be a senior management, he will get promoted and empowered. current ly this maybe only happening in huang or su s company.
if you understand how critical talent is to this industry you will know this is just the most basic requirement.

TSMCs loss of liang moon song caused them massive damage, US side lost multiple liang level of people in past few decades, what you think would happen.

second thing is how to deal with the capable Chinese engineers. you either employ them or they will incorporate a company to beat you. that is what happened in tw. in Israel engineers can get a job in one of the r&d facilities set up by a multinational. that is why there is no Israel TSMC.
in tws case they might want to treat it as a assembly house only and didn't put any r and d there. look at what happened later.

 

[tooLongInEDA]

Foundry should build capacity to order. Pat was ridiculous in his expansion talk. There simply was no reason to believe that much demand was coming on the timeline he talked. In many ways the BoD has some responsibility in controlling the CEO. Here the BoD is lacking, sad.

Given then efficiency need a layoff and retirement of people would simply be clearing of the dead wood and even if 15% of TD left I’d say they could still deliver. Actually the more that leave maybe the better, only keep the committed, everyone else can leave.

It's very difficult to build capacity to order, isn't it ? Not least because of the time lag between planning capacity and bringing it on line. You can ultimately only build to forecast. The reason why semis has historically been such a cyclical boom and bust industry.

The fact that semis currently seems to be less cyclical than in the past is - in my view - a comforting illusion (the classic "this time it's different" delusion). Some of the end markets for semis have fairly stable and predictable demand. But a significant proportion are not (I suggest AI is one of those).

Coming back to Intel's planning, I guess the modern get out here is to say that if it all goes wrong that you're only building fab shells. Which makes me curious - does anyone other than Intel go in for this fab shells model ?

 

[benb]

I agree with hist78 that keeping foundry and design together means neither will succeed, and best chance for one to survive, means separating.

I also agree with BruceA Intel x86 assets form the foundation of a foundry. It's exactly what is missing at Samsung.

The one thought I can add to this forum is, it is correct to state TSMC is a total monopoly. They have the scale, no one else does. Lack of scale motivates the solution, which will probably involve transactions with Samsung, UMC, GF and Intel, to result in a cobbled-together entity with enough scale to compete.

I sort of wish the status-quo, keeping trying-until-2027 idea was true, but that doesn't work at public businesses. Maybe Intel should have gone private.

Agree the worst case for Intel is to invite the Wall Street types in, and yet, an objective sum-of-the-parts-now analysis may be just what is needed. The Wall Street types may unlock future AI value, with a high multiple, that may help achieve the new not-TSMC entity.

So, here's my opinion of what Intel has that has a future:
-TD organization has proven its worth. 5 nodes 4 years, amazing.
-The EUV fabs (Portland, Chandler, Lexlip) have a future
-Royal core cancellation gives a clue about the future of design at Intel: No future. Time to leave.

 

[C-4]

Most major mature industries have a duopoly structure (airplanes, operating systems, even political parties). There ought to be room for a second player in the near-leading-edge lithography space too. That’s probably Samsung and/or Intel. (Or maybe SMIC in about a decade…)

More to the point, if Intel fabs separate from Intel design, then the following are likely to happen:
1. Intel fabs will be the only supplier with enough capacity for Intel design. AMD had a hard time securing wafers at the leading edge at TSMC and they were much smaller at the time.
2. Intel design might be locked to using Intel fabs by contract. AMD had this problem until GF gave up on catching up to the leading edge.
3. Intel design will have 0 ability to help/force the fab team to catch up.

That sounds like a lose-lose proposition.

I think at this point Intel is committed to the fab model, and in any case it is too early to judge their success. I think it all depends on their next big node 18A. If that is truly good enough then they will attract customers on both product and fab sides.

 

[BruceA]

The other alternative to Pat’s audacious IDM2.0…. Was the bean counter approach and likely what wall street thinks too it they had a go with Bob Swan as CEO.

Lunarlake platform has compute tile as well as the GPU and SOC all at TSMC follow that with PantherLake and its respective chips on N2/A16 and N4p and N5. Intel fabs are no more and close up shop as they have no other customers and little interest in their 1276, 1274, 1222 derivatives.

I guess still a possible narrative but feel the prroduct groups have no love for the fab side and given independence would pick TSMC over IFS just to spite their treatment from them for the past two decades pretty much a feeling.

It’s this or kiss Intel fans and any US technology leadership goodbye.

 

[BruceA]

Most major mature industries have a duopoly structure (airplanes, operating systems, even political parties). There ought to be room for a second player in the near-leading-edge lithography space too. That’s probably Samsung and/or Intel. (Or maybe SMIC in about a decade…)

More to the point, if Intel fabs separate from Intel design, then the following are likely to happen:
1. Intel fabs will be the only supplier with enough capacity for Intel design. AMD had a hard time securing wafers at the leading edge at TSMC and they were much smaller at the time.
2. Intel design might be locked to using Intel fabs by contract. AMD had this problem until GF gave up on catching up to the leading edge.
3. Intel design will have 0 ability to help/force the fab team to catch up.

That sounds like a lose-lose proposition.

I think at this point Intel is committed to the fab model, and in any case it is too early to judge their success. I think it all depends on their next big node 18A. If that is truly good enough then they will attract customers on both product and fab sides.

TSMC given 18 month forecast can easily and happily ramp up to supply Intel, nothing would make them happier to be Intels, AMD, Nvidia, Qualcomm, Apple, Microsoft, Amazon and the worlds only supplier! Dang another 10% price increase or more and they would than invest and target far more than low 50% GM.

 

[hist78]

Most major mature industries have a duopoly structure (airplanes, operating systems, even political parties). There ought to be room for a second player in the near-leading-edge lithography space too. That’s probably Samsung and/or Intel. (Or maybe SMIC in about a decade…)

More to the point, if Intel fabs separate from Intel design, then the following are likely to happen:
1. Intel fabs will be the only supplier with enough capacity for Intel design. AMD had a hard time securing wafers at the leading edge at TSMC and they were much smaller at the time.
2. Intel design might be locked to using Intel fabs by contract. AMD had this problem until GF gave up on catching up to the leading edge.
3. Intel design will have 0 ability to help/force the fab team to catch up.

That sounds like a lose-lose proposition.

I think at this point Intel is committed to the fab model, and in any case it is too early to judge their success. I think it all depends on their next big node 18A. If that is truly good enough then they will attract customers on both product and fab sides.

Let's go back to the basics. Are we debating IDM vs. the Fabless/Foundry model? Are we questioning whether Intel's IDM model is still viable and capable in the 21st century? Are we considering whether fabless or fab-lite semiconductor companies regret not returning to the IDM model?

Aren't we witnessing so many fabless companies eating into Intel's lunch, or even enjoying the dinner that Intel can't afford to have?

 

[BruceA]

Let's go back to the basics. Are we debating IDM vs. the Fabless/Foundry model? Are we questioning whether Intel's IDM model is still viable and capable in the 21st century? Are we considering whether fabless or fab-lite semiconductor companies regret not returning to the IDM model?

Aren't we witnessing so many fabless companies eating into Intel's lunch, or even enjoying the dinner that Intel can't afford to have?

Chicken or Egg debate? If 10nm was / had been no more than 3-6 months late instead of years late and Canon Lake and Ice Lake came out on schedule and 10nm didn’t suck all the Intel4/3 resources slipping that program I could venture that Intel would have continued their cadence of leadership tech long and some parity in real product performance. Of course if they did that would their designs have done Lunar Lake the way it is now?

But for sure Intel fabs would be full of Intel products instead of paying TSMC billions upon billions instead of their own fabs.

All this likely would have slowed or aborted the urgency for an IFS model.

But there is none of that and only speculation. What we can be certain is if Intel delivers 18A on schedule and ramps Panther Lake in 2025 and the product is competitive and if their PDK1.0 is real and healthy there should be enough momentum for fabless to reconsider Intel as real and a real alternative to drive price concession at TSMC and than Intel and TSMC could be competitive choice on 14A versus A16. It would be good for the world to have a choice. TSMC has scale and likely would accept a sub 50% GM and the world fabless would benefit. The question is does Intel have the cashflow and runway to drive this as well assume they can execute given the cash pinch they must be taking in RD. TSMC has none of those cash constraints currently and the N3 and N2 lineup looks like another cash cow.

Clearly Pat and his bean counter failed to fully comprehend the cost on this path and miscalculated the need as well as the risk of really delivering the advanced nodes compared to TSMC cost no concern for the same node.