Intel on the Brink of Death

[Xebec]

A little over dramatic as usual. Here is my 2 cents:

What is Wrong with Intel? - Semiwiki

One of the most popular topics on the SemiWiki forum is Intel, which I understand. Many of us grew up with Intel, some of us have worked there, and I can say that the vast majority of us want Intel to succeed. The latest Intel PR debacle is the abrupt departure of CEO Pat Gelsinger.…

semiwiki.com

Excellent writeup as always. I really hope someone takes you up on your offer -->

Bottom line: Some very difficult decisions have to be made by some very qualified people. Take a look at the current Intel Board of Directors and convince me that they are the right ones to do it. You have an open invitation to be a guest on our podcast or post a written response to this blog.

 

[MKWVentures]

I tend to agree with this; however, I also agree with @siliconbrush999 that Intel has enjoyed a big advantage having vertical integration with their design and fab:

@siliconbruh999 @OneEng

It was an advantage in 1992-2010. It is not an advantage anymore. Almost everyone else figured that out in the last 25 years. Intel CEO figured it out in 2020. Then they "unfigured" it out. Intel seems to still be committed to IBM2.0. .... Stories of glory from the past don't work anymore.

 

[Xebec]

@siliconbruh999 @OneEng

It was an advantage in 1992-2010. It is not an advantage anymore. Almost everyone else figured that out in the last 25 years. Intel CEO figured it out in 2020. Then they "unfigured" it out. Intel seems to still be committed to IBM2.0. .... Stories of glory from the past don't work anymore.

Can we break this out into a couple of areas --

I think it's a potential technical advantage* to have your own fabs, but maybe not a cost advantage because fabs have gotten so expensive. At the highest level is that your thinking? or do you see the IDM as a technical disadvantage too?

*Especially as Intel Foundry appears to be ahead of TSMC on packaging tech, and BSPD/transistor perf with 18A.

I'd also argue Swan didn't quite figure it out - he kept spending on Intel 10nm/7, and Intel even released a roadmap for 5nm, 3nm at the time indicating they were still investing in fab tech. I would imagine if Intel was trying to wholly transition to TSMC under Swan they'd have to be more open in disclosures about that.

 

[MKWVentures]

@Xebec
cost is an important part of technology. intel doesnt have the structure or scale to be cost effective. outsourcing was a good decision

We will see if intel is ahead of tsmc in Jan 2026... they certainly are not now. Intel often is ahead on the powerpoint slides.

18A is doing well from what i hear. but the cost/finances are a problem.

the data has told and will tell the story. lets see what happens next.

 

[hist78]

Former Intel CEO Pat Gelsinger calls for prayer and fasting for employees.

Former Intel CEO Pat Gelsinger calls for prayer and fasting for employees

Former Intel CEO Pat Gelsinger took to social media to invite people to pray and fast for the chipmaker's employees, saying the company is going through a "difficult period."

www.foxbusiness.com

 

[delong.height]

It will kill both the design and product and they will loose the scale

What Intel need is partner, now nobody really want to be theirs. So you got to lure with with benefits.

If Intel split into two independent companies, foundry and product design, I believe at least one of them will survive. If they don't do it quickly now, Intel product design and manufacturing divisions may go down together and to be sold in pieces.

Sorry, it may hurt many people's feelings, mine too.

They cannot split into two, one must sacrifice for the other. And given foundry is national security. Government should let Intel sell product side, for them to get future wafer purchase agreements from national champions like Apple, AMD, NVIDIA, Broadcom, and Qualcomm. The book value of IFS is even more than the Intel stock along. With the cash injected and wafer purchase agreement, that's what can keep iNTEL survived.

I fucking hate the board, they should all being fired, hopefully there will be an activist because this shit-hole they've been digging deeper and deeper.

Formally we have IDM 2.0, and if foundry come back then bon voyage! Now they want to take one and give other. Fucking stupid.

 

[delong.height]

Unfortunately, a article I published on reddit seemed to be verified, but I really don't want it to be.

https://www.reddit.com/r/wallstreetbets/comments/1eiu09u

I hope that there's an activist coming up soon. Or we all getting screwed by this fucking board

 

[delong.height]

Do activists know who they should be targeting this time? I hope they do
.

https://www.cnbc.com/2024/08/23/intel-intc-activist-defense-sources.html

 

[OneEng]

Your analysis is very thorough and comprehensive, and really hits the core challenges Intel is facing right now. Transforming into a company that provides services to foundries is indeed much more complex, takes longer, and requires a greater financial investment than producing designs directly on TSMC nodes. Intel's vertical integration model itself does have advantages similar to Apple's, but as you said, the exponential increase in costs at new nodes combined with the decline in market share does create a "perfect storm."

AMD's strategy on wafer costs is also worth watching. This approach effectively optimizes resource utilization by designing in high volume on older nodes while focusing on high-margin parts on leading-edge nodes. But as you pointed out, PC consumers are price-sensitive, and the limitations of the entire industry on pricing will undoubtedly limit strategic development.

As for Intel's future direction, I agree that it is unrealistic to continue operating at the current rate of losses. They do need to expand their customer base through foundry services to achieve revenue growth. However, as you mentioned, this transformation not only faces technical challenges, but also takes time to build a reliable foundry business.

Regarding the view of splitting up the company, I think that in the long run, this may indeed be a strategic option worth considering. Profitability in the design division may be easier to achieve, while the independent operation of the fab can focus on improving efficiency and attracting more customers. But this process is bound to be accompanied by major changes and challenges.

Your analysis is thorough and your observation of industry dynamics is sharp. I hope Intel can find ways to adapt to market changes while leveraging its own strengths. This is a challenging era, but it may also breed new opportunities!

Thank you for your esteem and high opinion of my thoughts.

While I stand by these beliefs as stated, as an American, and a military veteran, I would very much like to see Intel succeed for the sake of my countries security (preferably without tax payer money ); however, their current model is clearly not sustainable and must change.

 

[OneEng]

@siliconbruh999 @OneEng

It was an advantage in 1992-2010. It is not an advantage anymore. Almost everyone else figured that out in the last 25 years. Intel CEO figured it out in 2020. Then they "unfigured" it out. Intel seems to still be committed to IBM2.0. .... Stories of glory from the past don't work anymore.

LOL. Got a good chuckle from your reply "Unfigured" .

As in the posts above, I was also referring to a technical advantage in having a vertically integrated fab ..... always assuming that the fab is producing leading edge technology that supports the designs.

Your point (as is mine) is that this no longer makes economic sense for Intel. The empirical data needed to come to this conclusion has been obvious for all to see for some time now.

 

[sit]

The product side is in a slow death spiral and the only piece of Intel that can survive long term is Foundry. Apart from improved versions with each gen, there has been little real product invention since the 1970's. Intel could lead vs. AMD due to better process technology, that is reversed today, but both Intel and AMD are dealing with ARM's ascendency and it is just time before ARM consumes the client business. The future is not x86.

The big challenge is having the funds to bridge the foundry transformation. Intel needs to poach enough business to roughly double its current wafer demand to be of sufficient scale. Could take a decade+. At the current burn rate that is at least $50B. Is that best achieved by keeping the product side and harvesting margin or by divesting of the product businesses? (or keep some, sell some) Last resort, a giant bailout likely happens to cover any residual need under a national security pretense.

 

[siliconbruh999]

The biggest problem as people have said is filling Fabs they need something to fill them

They need to bring back everything that was outsourced to TSMC in couple of years now that they will have competitive process

Something along the line of Custom chips for Hyperscalers like Amazon should be their way to fill fabs get competitive design and sell them for cheaper than competition

AMD alone can't pull x86 like Intel has pulled it for years.AMD itself is helped by Intel's Software Improvements over the years
Tthe ISA doesn't matter if it is working fine and x86 still has life in it it comes down to implementation.

 

[delong.height]

The product side is in a slow death spiral and the only piece of Intel that can survive long term is Foundry. Apart from improved versions with each gen, there has been little real product invention since the 1970's. Intel could lead vs. AMD due to better process technology, that is reversed today, but both Intel and AMD are dealing with ARM's ascendency and it is just time before ARM consumes the client business. The future is not x86.

The big challenge is having the funds to bridge the foundry transformation. Intel needs to poach enough business to roughly double its current wafer demand to be of sufficient scale. Could take a decade+. At the current burn rate that is at least $50B. Is that best achieved by keeping the product side and harvesting margin or by divesting of the product businesses? (or keep some, sell some) Last resort, a giant bailout likely happens to cover any residual need under a national security pretense.

I think here's the deal for board who prefer product side over foundry. Do you want a IDM that has a high possibility of failure, and not like by anyone in the industry.

Or do you prefer a business that is dominated by two companies and only, TSMC and Intel. The profit margin of IFS may be much larger than product side at the end of decade given the growth momentum in packaging and other advanced manufacturing techniques, and more durable.

 

[blueone]

The product side is in a slow death spiral and the only piece of Intel that can survive long term is Foundry. Apart from improved versions with each gen, there has been little real product invention since the 1970's. Intel could lead vs. AMD due to better process technology, that is reversed today, but both Intel and AMD are dealing with ARM's ascendency and it is just time before ARM consumes the client business. The future is not x86.

The big challenge is having the funds to bridge the foundry transformation. Intel needs to poach enough business to roughly double its current wafer demand to be of sufficient scale. Could take a decade+. At the current burn rate that is at least $50B. Is that best achieved by keeping the product side and harvesting margin or by divesting of the product businesses? (or keep some, sell some) Last resort, a giant bailout likely happens to cover any residual need under a national security pretense.

The biggest problem as people have said is filling Fabs they need something to fill them

They need to bring back everything that was outsourced to TSMC in couple of years now that they will have competitive process

Something along the line of Custom chips for Hyperscalers like Amazon should be their way to fill fabs get competitive design and sell them for cheaper than competition

AMD alone can't pull x86 like Intel has pulled it for years.AMD itself is helped by Intel's Software Improvements over the years
Tthe ISA doesn't matter if it is working fine and x86 still has life in it it comes down to implementation.

I think here's the deal for board who prefer product side over foundry. Do you want a IDM that has a high possibility of failure, and not like by anyone in the industry.

Or do you prefer a business that is dominated by two companies and only, TSMC and Intel. The profit margin of IFS may be much larger than product side at the end of decade given the growth momentum in packaging and other advanced manufacturing techniques, and more durable.

I agree with several ideas in these posts. I think the long-term play is IFS. We can debate the merits of Intel's products, but the reality is that more and more of the entire computing environment is moving into the cloud, and the cloud companies are all designing their own chips (at least at the digital level; they partner with the likes of Broadcom and Marvell to do the analog stuff) and using a foundry for manufacturing. Yeah, they still buy CPUs and GPUs and memory (DRAM and flash), and some networking parts, but in the future most of their spending on merchant chips will be for memory and storage. When the cloud companies design their own chips the cost savings are very large. And Cadence and Synopsis and the wide availability of IP from them and numerous other companies are making the design effort ever easier too. The merchant chip market is going to shrink substantially over time as a percentage of the overall logic chip volume, like the next ten years, and even if Intel has winning designs they'll be facing a smaller market.

I'm especially aligned with @delong.height, in that when you're one of three or one of two, you're in a more favorable position regardless of what it takes to win, regardless of cost. Even private equity companies are going to be smart enough to see this. When your customers can build a better version of a product for themselves than you can build, it's time to reconsider your strategy.

IFS needs a $100B Go-Fund-Me and the right leadership. IMO, the world is dumb not to make it easier for this to happen.

 

[Xebec]

@Xebec
cost is an important part of technology. intel doesnt have the structure or scale to be cost effective. outsourcing was a good decision

We will see if intel is ahead of tsmc in Jan 2026... they certainly are not now. Intel often is ahead on the powerpoint slides.

18A is doing well from what i hear. but the cost/finances are a problem.

the data has told and will tell the story. lets see what happens next.

Totally agree.

My point was more that losing access to the fabs directly will reduce their technical competitiveness in other markets too. If the cost factor is say 10-15% higher than TSMC, it may be more than worth it to get some extra clock speed, priority, or other benefits (mindshare from a better product, etc.) out of the IDM model where it's possible to do more fine tuning*.

Likewise if the cost factor is 2X, that's pretty hard to justify without a lot of gymnastics.

*Totally recognize this has a pile of pros/cons too; but in-house process changes should mean faster and closer to "what you want" than outsourcing.

 

[Xebec]

We can debate the merits of Intel's products, but the reality is that more and more of the entire computing environment is moving into the cloud,

The merchant chip market is going to shrink substantially over time as a percentage of the overall logic chip volume, like the next ten years, and even if Intel has winning designs they'll be facing a smaller market.

Just a thought here. The same thing was said in the 1990s/2000s before Mobility took off. Large companies were consolidating apps to data centers, HTML5 was supposed to place the primary need for compute on the back end, and people were supposed to replace thick apps with web apps that would be device agnostic (further commoditizing the device market). A lot of thin clients were talked about.

In reality we had (substantial) growth on both ends because increased data center and connectivity performance enabled more need for access. The Data center growth was certainly more fantastic than handheld/PC, but there are always drivers for more compute on the receiver end too. GPUs took off, and now AI has a chance to require increased investment on the user device side too.

I'd nuance this that the Intel addressable portion of the overall semiconductor pie is getting smaller, but that doesn't mean that they can't grow in terms of $$, since the overall pie is growing too.

(that said, I totally agree with you that data center providers making their own custom chips represents a grave threat to Intel's product business. Intel *should* have prevented this by making their products too good to consider replacing.)

 

[blueone]

Just a thought here. The same thing was said in the 1990s/2000s before Mobility took off. Large companies were consolidating apps to data centers, HTML5 was supposed to place the primary need for compute on the back end, and people were supposed to replace thick apps with web apps that would be device agnostic (further commoditizing the device market). A lot of thin clients were talked about.

I remember that thin client nonsense, and it was nonsense, but as Apple has proven, you don't need merchant CPUs, GPUs, or NPUs to be a leader in personal computing. Could Dell, HP, and Lenovo do their own CPUs? A few years ago I would have snarled at that thought, now I'm leaning the other way. Qualcomm may be the last one on the train just before the plane is about to push back from the gate.

In reality we had (substantial) growth on both ends because increased data center and connectivity performance enabled more need for access. The Data center growth was certainly more fantastic than handheld/PC, but there are always drivers for more compute on the receiver end too. GPUs took off, and now AI has a chance to require increased investment on the user device side too.

Agreed, but like I said, now I'm thinking client CPUs could be at risk for Intel, AMD, and Qualcomm too. Merchant chips just have huge sales, marketing, product management, and the-sum-of-the-requirements costs that in-house designs don't. The advantages of in-house designs just look too good for rich companies to pass up.

I'd nuance this that the Intel addressable portion of the overall semiconductor pie is getting smaller, but that doesn't mean that they can't grow in terms of $$, since the overall pie is growing too.

(that said, I totally agree with you that data center providers making their own custom chips represents a grave threat to Intel's product business. Intel *should* have prevented this by making their products too good to consider replacing.)

I've met a few of the cloud company design teams, especially Amazon's Annapurna Labs team. They're hungry and motivated in a way I haven't seen from any merchant chip company.

 

[OneEng]

Some very good points here.

100bn is a pretty tough pill for Intel to swallow at this time to create a leading competitor to TSMC with IFS. While that may well be the best strategy, I don't know if it is a palatable one.

With respect to the DC takeover by big DC companies vs purchasing Intel/AMD x86 .... possibly true. There are significantly fewer apps to port and optimize in this space than in client (PC and mobile). In client, business still make up the lions share of x86 I believe. Asking every company around the world to get all their apps working on something other than x86 will be a long journey so I think this market is secure for Intel/AMD for at least another 10+ years.

Still, losing the lucrative DC market would be a huge blow to both Intel and AMD. With respect to this line of thought I do have a question though. In what way is x86 inferior to ARM in DC? Seems like there isn't an ARM out there that is competitive in this market yet.

Furthermore, there are issues with the "lots of little efficient cores" strategy. There are lots of DC software out there that charge an annual license per core. This is why very powerful cores are chosen for most applications. x86 first decode stage is just changing the variable length CISC to equal length "RISC like" instructions, so I don't see a fundamental advantage for ARM in this space. Also, features like SMT appear to be lacking in ARM to date.

Lots to think about I guess.

 

[Xebec]

I've met a few of the cloud company design teams, especially Amazon's Annapurna Labs team. They're hungry and motivated in a way I haven't seen from any merchant chip company.

Unfortunately I can't shake the feeling that it may be in 10 years we look back at this time as the final death throes of Intel... and draw the conclusion that there was nothing that would have saved Intel after Otellini gave up the Apple business.

Even if Swan and Gelsinger had pivoted the company to take advantage of AI, and/or their 10nm debacle hadn't happened (Hi BK), those things may only have prolonged the inevitable. (Pat should have done more for AI, but realistically given development timelines -- it would have needed aggressive action from Swan too).

If Intel had won the Apple business (+ was hungry for more business afterwards), they might have ended up (2030-2035?) being the company that could justify further node investments beyond what TSMC could justify.

 

[blueone]

With respect to the DC takeover by big DC companies vs purchasing Intel/AMD x86 .... possibly true. There are significantly fewer apps to port and optimize in this space than in client (PC and mobile).

For the cloud data centers, the cloud companies own the entire stack, including the OS distributions. It makes software stack porting and tuning a process under their control.

In client, business still make up the lions share of x86 I believe. Asking every company around the world to get all their apps working on something other than x86 will be a long journey so I think this market is secure for Intel/AMD for at least another 10+ years.

Yup, by unit volume there's no comparison. Client CPUs are a multi-hundred-million unit market, server CPUs are tens of millions, certainly less than 50 million per year, even including proprietary CPUs.

Still, losing the lucrative DC market would be a huge blow to both Intel and AMD.

Very much so.

With respect to this line of thought I do have a question though. In what way is x86 inferior to ARM in DC? Seems like there isn't an ARM out there that is competitive in this market yet.

x86 CPUs aren't inferior to Arm CPUs in any way except possibly power consumption, but being "competitive" has a different meaning when the CPU is custom-designed for the customer. The benchmarks that AMD and Intel like to tout are mostly for marketing purposes. When the customer has a target application in mind where they design and develop the software, stuff like memory systems and I/O systems mean more to most applications than instructions per clock, and what the clock speeds are. Intel and AMD talk about that stuff because they're trying to differentiate cores, and they have to make compromises on caches and memory systems, because they're trying to develop one-design-fits-many solutions. If you're in charge of running EC2, which consumes a huge number of Annapurna CPUs per year, you're only interested in the overall application price-performance, response times, power consumption, reliability, and implementation of the most important features to EC2.

Furthermore, there are issues with the "lots of little efficient cores" strategy. There are lots of DC software out there that charge an annual license per core. This is why very powerful cores are chosen for most applications. x86 first decode stage is just changing the variable length CISC to equal length "RISC like" instructions, so I don't see a fundamental advantage for ARM in this space. Also, features like SMT appear to be lacking in ARM to date.

Arm Neoverse cores can be impressive enough. This is how Ampere won over Oracle, though Ampere now has an Arm architecture license for the instruction set, and is now designing their own custom cores, which they claim are even more impressive.

It is also a common misconception that Arm cores do not use variable length instructions. Some Arm cores do. And Arm has complicated-up their designs to compete with x86 with stuff like vector operations. If you want more RISC-like designs, I think you need to stick to Cortex IP, but I'm not an Arm expert.

Arm cores do not support hardware multi-threading, and I doubt they will in the near future.