Question Intel's future after Pat Gelsinger

[DavidC1]

they are wrapping up NEX and Automotive as well

Yea I heard.

I don't think some hail mary aka E core team is going to save them this time. I love what they are doing there but Intel sounds like hell. Some part of Intel sounds like despite this they were able to achieve what they were able to do. Now they seem to be draining the soul and spirit of the company.

2028 is way far off.

 

[511]

Yea I heard.

I don't think some hail mary aka E core team is going to save them this time. I love what they are doing there but Intel sounds like hell. Some part of Intel sounds like despite this they were able to achieve what they were able to do. Now they seem to be draining the soul and spirit of the company.

Their current hail Mary is 18A for the process at least let it come out lol

2028 is way far off.

Yeah I am waiting for earnings lol for the spin they are going to give

 

[johnsonwax]

So does Apple but Tim is hands off luckily the hardware side of Apple VPs engineers are very competent. The software side of Apple is struggling now in AI because of being caught off guard but its failure falls on Tim. Apple has money so it can catch up, Intel though is in the deep end.

Your company can survive bean counter CEOs if some parts of your company excels, which for Apple is the iPhone sales. Intel suffered enough damage that NO parts of the company are excelling.

AI is a cultural mismatch for Apple, which is why they're struggling. You can't protect user privacy and operate the way OpenAI does. They're trying to build a competing business model for AI, and I'm not sure they're failing, but it's going to move slower than the 'let's shove all user data into the cloud and steal as much IP as possible during training'. What Apple really screwed the pooch on was overpromising. Historically Apple has been a very 'don't get too far over your skis' kind of company and that's broken down pretty badly around AI. Apple (actually everybody) does best when they ignore the fact they have investors. 'Maximizing shareholder value' might prove to be the most expensive idea ever floated in humanity.

 

[DavidC1]

'Maximizing shareholder value' might prove to be the most expensive idea ever floated in humanity.

It might be, but like many things it's not the idea that's wrong, but the inherently flawed human nature.

Shareholder stake exists because it's the easiest way to get wide exposure to mass capital. If you do it private it's much harder. Many things in advertising is alerting the world that you and your product simply exists. Many products fail commercially because it lacks such exposure, not necessarily because the product sucks.

Even simpler, saying Intel or Apple is making x,y,z mistake just means like any living entity(including a corporation which is made of living entities), it has a lifespan and when it's at an end, it perishes.

 

[johnsonwax]

It might be, but like many things it's not the idea that's wrong, but the inherently flawed human nature.

Shareholder stake exists because it's the easiest way to get wide exposure to mass capital. If you do it private it's much harder. Many things in advertising is alerting the world that you and your product simply exists. Many products fail commercially because it lacks such exposure, not necessarily because the product sucks.

Even simpler, saying Intel or Apple is making x,y,z mistake just means like any living entity(including a corporation which is made of living entities), it has a lifespan and when it's at an end, it perishes.

That's not what I'm saying. Maximizing shareholder value is not an inherent property of capitalism - it's a relatively new idea, cooked up in the 70s and started to get implemented in the 80s and 90s. Prior to that the general philosophy was 'create a new customer'. It was a philosophy centered on the needs of the customer and therefore revenues and reach of the business, not an investor focused philosophy centered on profits and rent seeking. That's not to say those things didn't apply previously - but they were considered economically undesirable. A heap of policies reinforced that, and when the new philosophy took over those policies changed to favor investors. Throughout that period the stock market was not a spectacular place to invest because the benefits of the business were being conferred to the customers and society. That is, after all, what the benefits of competition in capitalism are promised to be - cheaper and better products, not higher returns for investors. Bonds were a better investment because companies were raising capital by borrowing - because expanding the business was the goal.

In the modern telling, we're still fighting communism and selling the benefits of competition, but the incentives are aligned toward monopoly and away from competition. The policies can either support of resist that flawed human nature and they've shifted form resisting to supporting.

 

[Doug S]

and not seeing ARM cores support 64-bit in 2013 meant that Intel's x86 monopoly was starting to show cracks. Why would phones need more than 4GB in 2013. ARM cores going 64-bit means it will go in servers and laptops etc.

I'm not sure what you mean by this. Yes phones didn't "need" (by that you really mean "weren't yet shipping with") more than 4 GB in 2013. But ARM, Apple, and everyone else saw the writing on the wall. They could extrapolate DRAM density graphs and talk to DRAM OEMs about their roadmap for the next few years and know that phones would be shipping with more than 4 GB soon enough. And that would happen whether they needed it or not, since in the Android world at least for a time the amount of RAM a flagship had was (along with number of cores and number of megapixels) one of the numbers they used in advertising to claim their phone as better than the competition.

Even if there was never an ARM Mac, never a Snapdragon PC, no one was running ARM servers. ARM was going to go 64 bits even if was going to be forever limited to smartphones.

 

[mikegg]

Lip-Bu Tan doesn't seem like a solution either. They are cutting Linux engineering teams now.

At Phoronix one guy says they are cutting big on the fab side. Also it isn't just middle managers being cut, lot of engineers are being cut.

If you look at Tan's history he has a Venture Capitalist background. Intel, account company that sometimes uses engineers.

LBT is cutting too much we will see the earnings what was the cut

https://katu.com/resources/pdf/32e15aae-0951-486b-859b-b2b671e6d6c3-WARN9293OregonJobListing070725.pdf

Yea.... Tan lied about only firing middle managers. Exist is saying such claims are almost always lies. Can't do precise firing without being in a company for few years at least. He just got hired!

Those same managers do most of the firing too. I would not buy Intel products at this point.

Just some context for people who criticize Tan.

Intel has $22b in cash on hand. Intel lost $19b in 2024. One more year of this loss and Intel is bankrupt. This is what Tan inherited. A company on verge of bankruptcy. Couldn't acquire any external fab customers. Losing in cutting edge fabs. Losing in AI chips. Losing in x86 chips. Losing in virtually every single market.

 

[511]

Just some context for people who criticize Tan.

Intel has $22b in cash on hand. Intel lost $19b in 2024. One more year of this loss and Intel is bankrupt. This is what Tan inherited. A company on verge of bankruptcy. Couldn't acquire any external fab customers. Losing in cutting edge fabs. Losing in AI chips. Losing in x86 chips. Losing in virtually every single market.

That is not a cash loss read properly first also they are loosing share but you can't revert in like a month or two
https://www.intc.com/news-events/pr...-reports-third-quarter-2024-financial-results

 

[Doug S]

AI is a cultural mismatch for Apple, which is why they're struggling. You can't protect user privacy and operate the way OpenAI does. They're trying to build a competing business model for AI, and I'm not sure they're failing, but it's going to move slower than the 'let's shove all user data into the cloud and steal as much IP as possible during training'. What Apple really screwed the pooch on was overpromising. Historically Apple has been a very 'don't get too far over your skis' kind of company and that's broken down pretty badly around AI. Apple (actually everybody) does best when they ignore the fact they have investors. 'Maximizing shareholder value' might prove to be the most expensive idea ever floated in humanity.

Yes I agree. I think Apple made a real PR blunder when they started talking about AI just because everyone else was. Rather than promising something they didn't have at the time but planned to have later, they should have said "we're working on it, but in a careful way to protect privacy and insure it delivers up to our quality standards". Don't promise a date, promise it will be worth a wait. Yes they'd get hit a bit in the market that was bidding up everything AI related, but they ended up losing that premium when they didn't deliver on time anyway.

If anyone said "why are you taking so long" they could point to stuff like Grok going full nazi as evidence for why they are taking a more careful approach than everyone else racing to get it out there and alpha test on the general public.

 

[mikegg]

That is not a cash loss read properly first also they are loosing share but you can't revert in like a month or two
https://www.intc.com/news-events/pr...-reports-third-quarter-2024-financial-results
View attachment 127416

Ok thanks. Given that a TSMC 2nm fab costs $28 billion (estimate). Intel has $20 billion on hand right now. How do you think Intel can make a 16A fab?

 

[511]

Ok thanks. Given that a TSMC 2nm fab costs $28 billion (estimate). Intel has $20 billion on hand right now. How do you think Intel can make a 16A fab?

you mean 14A? The best way would be to double pattern EUV instead of High Na and reuse Arizona for their own need when someone asks for it just ask them to pay in advance and bring Ohio online this way they can skip building more fabs and save quite a lot. For foundry they have already spent a fortune.
The learning curve is expensive where you need volume and money

 

[mikegg]

you mean 14A? The best way would be to double pattern EUV instead of High Na and reuse Arizona for their own need when someone asks for it just ask them to pay in advance and bring Ohio online this way they can skip building more fabs and save quite a lot. For foundry they have already spent a fortune.
The learning curve is expensive where you need volume and money

They already bought High NA machines.

 

[511]

They already bought High NA machines.

those aren't enough they need at least 8-10 of those

 

[Io Magnesso]

those aren't enough they need at least 8-10 of those

No, Intel has already ordered about 6 Hi-NA EUVs.

 

[Io Magnesso]

Not all processes are exposed with Hi-Na EUV…
For example, even in the EUV process, there is a DUV process.

 

[Doug S]

Not all processes are exposed with Hi-Na EUV…
For example, even in the EUV process, there is a DUV process.

Yep even if they're using high NA EUV with 14A (and I'm not convinced there is any reason to do so) they'd be using it on only 2 or 3 layers at most. If anything BSPDN makes the need for high NA and certainly the number of layers for high NA less than a process that's designed to allow choice to use or not use BSPDN like TSMC. So Intel may have an advantage on that front if they are BSPDN only.

 

[Joe NYC]

Yea I heard.

I don't think some hail mary aka E core team is going to save them this time. I love what they are doing there but Intel sounds like hell. Some part of Intel sounds like despite this they were able to achieve what they were able to do. Now they seem to be draining the soul and spirit of the company.

2028 is way far off.

One way to interpret Intel's roadmap is that Intel does not foresee its main CPU design team (in Israel) delivering a whole lot in next 2-3 generations, and at the end of the 2-3 year period, it will be disbanded.

During those 2-3 years, Intel will have to survive selling products, for which they already have limited expectations (given the roadmap).

So it seems like the gap in product performance between Intel and competition (Intel is now reduced to competing only with AMD, its only competition) that gap is likely to only grown wider.

Widening technology deficit will push Intel out of all of the premium segments, reduce ASPs. Intel's low cost offering (Alder Lake - Raptor Lake) will soon be shut out of the market place. Which will lead to Intel's costs rising and ASPs falling, profits disappearing.

Will Intel be able to limp to 2028 to catch that Hail Mary? What's the probability of that? I would say 1 in 3 at best.

 

[Geddagod]

One way to interpret Intel's roadmap is that Intel does not foresee its main CPU design team (in Israel) delivering a whole lot in next 2-3 generations, and at the end of the 2-3 year period, it will be disbanded.

During those 2-3 years, Intel will have to survive selling products, for which they already have limited expectations (given the roadmap).

So it seems like the gap in product performance between Intel and competition (Intel is now reduced to competing only with AMD, its only competition) that gap is likely to only grown wider.

Widening technology deficit will push Intel out of all of the premium segments, reduce ASPs. Intel's low cost offering (Alder Lake - Raptor Lake) will soon be shut out of the market place. Which will lead to Intel's costs rising and ASPs falling, profits disappearing.

Will Intel be able to limp to 2028 to catch that Hail Mary? What's the probability of that? I would say 1 in 3 at best.

Idk, it's hard to tell.
The P-core team does seem like they will be delivering a decent bit in the next 2-3 years. Griffin Cove is rumored to be a tock, as well as the P-core in NVL.
The team seems to be disbanded as they want a core better in power and area, not even necessarily perf, even though the current tock and the next tock seem to underwhelm in terms of perf as well.

In mobile Intel has seemed to gain a great bit of competitiveness, and their server CPU position is improving too. GNR vs Turin Dense is likely worse than DMR vs Venice Dense is going to be, IMO.

The real problem is that their desktop situation is deteriorating, but given gains in other markets...

Even if ASPs decrease, Intel's margins vs right now will prob improve. Wildcat Lake and even NVL-H seem like great cost competitive replacements to RPL. Intel 7 is already notoriously expensive (same cost as 18A), so Intel can simultaneously be replacing a bit of the ARL lower end sku market, as well as the RPL segment, with cost effective solutions.

I don't think unified core is a hail marry either. It doesn't seem like it will be as revolutionary as RYC (which could be argued as a hail marry), and at best I think we can expect it to be competitive with Apple, who has the industry leading cores rn. I think a reasonable expectation would be for it to reach parity with AMD's cores.

 

[Joe NYC]

Even if ASPs decrease, Intel's margins vs right now will prob improve. Wildcat Lake and even NVL-H seem like great cost competitive replacements to RPL. Intel 7 is already notoriously expensive (same cost as 18A), so Intel can simultaneously be replacing a bit of the ARL lower end sku market, as well as the RPL segment, with cost effective solutions.

It is not just the CPU tile (and I am not sure I agree with 18A cost being the same as Intel 7, there is a lot of very expensive equipment and fabs to depreciate).

The problem is that Intel is adding all these other tiles and expensive packaging.

I have not seen any analysis of this, but my opinion is that the Foveros packaging Intel is using is very similar to CoWoS, and its cost should then also be in same range (not cheap). And this is being sold not for $20,000, like datacenter GPUs but $200 CPUs.

I don't think unified core is a hail marry either. It doesn't seem like it will be as revolutionary as RYC (which could be argued as a hail marry), and at best I think we can expect it to be competitive with Apple, who has the industry leading cores rn. I think a reasonable expectation would be for it to reach parity with AMD's cores.

If you consider the size of the performance gap Intel will likely face in 2028, it will be another leapfrog-like attempt, to bridge that gap in on generation.

 

[Io Magnesso]

Idk, it's hard to tell.
The P-core team does seem like they will be delivering a decent bit in the next 2-3 years. Griffin Cove is rumored to be a tock, as well as the P-core in NVL.
The team seems to be disbanded as they want a core better in power and area, not even necessarily perf, even though the current tock and the next tock seem to underwhelm in terms of perf as well.

In mobile Intel has seemed to gain a great bit of competitiveness, and their server CPU position is improving too. GNR vs Turin Dense is likely worse than DMR vs Venice Dense is going to be, IMO.

The real problem is that their desktop situation is deteriorating, but given gains in other markets...

Even if ASPs decrease, Intel's margins vs right now will prob improve. Wildcat Lake and even NVL-H seem like great cost competitive replacements to RPL. Intel 7 is already notoriously expensive (same cost as 18A), so Intel can simultaneously be replacing a bit of the ARL lower end sku market, as well as the RPL segment, with cost effective solutions.

I don't think unified core is a hail marry either. It doesn't seem like it will be as revolutionary as RYC (which could be argued as a hail marry), and at best I think we can expect it to be competitive with Apple, who has the industry leading cores rn. I think a reasonable expectation would be for it to reach parity with AMD's cores.

Well, Intel's Laptop products aren't that bad
Originally, recent architecture changes are supposed to be for laptops
If you keep doing it, it will be better
Isn't that ironic?