Industry faces “acute” CPU shortage with hope that Intel 18A yields improve

[MKWVentures]

I think that this organizational setup makes the problem even worse.

The problem is that new nodes are becoming exponentially more expensive to develop AND require more processing time per wafer. Intel is trying to write off the same CapEx as TSMC is, but over 8 times fewer wafers.

Intel needed to be making the move to separate IFS and compete head-to-head with TSMC 10 years ago to avoid the current problem.

Of course, the math is simple while resolving the issue is clearly not.

you are 100% correct. The math is clear and has been clear since 2019. There is a reason the world went to a foundry model and then that the foundry model contracted to one successful advanced foundry. math

Excel shows this very simply

The answer is a) do not do IDM2.0 or b) get 15B in external foundry business. everything else is a rationalization about why math and profit doesnt matter. One CEO chose a, One CEO chose b. they both 100% agreed on the options. LBT has figure out how to deal with "what if neither happens?"

Side fun excel spreadsheet experiment "if you are not already making money and you are losing money while trying to grow, you cannot catch up to have a positive return.

 

[siliconbruh999]

Sometimes we forget Brian Kranzich Contribution of destroying the Company which would have made IFS lot easier and product wouldn't be in this mess in the first place cause there wouldn't have been any delays

 

[MKWVentures]

Market share numbers are in. AMD made record gains in DC CPU share..... and is selling units at higher ASP than Intel. Intel lost almost 5% share in a quarter and had a shortage of units (all DC CPUs are made at Intel Fabs).

The stock price went up 4x because IFS may have significant material revenue in 2029

 

[swka]

Market share numbers are in. AMD made record gains in DC CPU share..... and is selling units at higher ASP than Intel. Intel lost almost 5% share in a quarter and had a shortage of units (all DC CPUs are made at Intel Fabs).

The stock price went up 4x because IFS may have significant material revenue in 2029

Intel losing shares to AMD has been going on for a long time, question is has that slowed down in the last couple quarters. Agree IFS optimism is the main driver behind market surge

 

[siliconbruh999]

honestly that's a problem for DCAI they are the ones that are constantly delaying stuff and why hasn't it been fixed yet

 

[Brady]

Intel losing shares to AMD has been going on for a long time, question is has that slowed down in the last couple quarters. Agree IFS optimism is the main driver behind market surge

As MK pointed out, both share and ASP relative to Intel has accelerated in the past quarter.

Continued terrible competitiveness by Intel DCAI.

 

[MKWVentures]

As MK pointed out, both share and ASP relative to Intel has accelerated in the past quarter.

Continued terrible competitiveness by Intel DCAI.

DC CPU Share loss was higher than expected. Intel gained share in desktop but also lost share in Notebooks.
DC roadmaps can be very slow to change. As an example, Some intel people were predicting 2 years ago that AMD would get 50% of DC CPU revenue share in 2027 (looks like they were correct) . I think the shortage adds noise to this slow erosion (In either direction).

My point since last quarter has been that Intel is losing share in DC CPUs and saying it has a shortage at the same time.... and Q1 DC CPU units were down 5% YoY. it seems odd how this can happen.

On the positive side, I believe this will push Intel customers to Intel 3 and 18A which will simplify Intels roadmap.

 

[swka]

DC CPU Share loss was higher than expected. Intel gained share in desktop but also lost share in Notebooks.
DC roadmaps can be very slow to change. As an example, Some intel people were predicting 2 years ago that AMD would get 50% of DC CPU revenue share in 2027 (looks like they were correct) . I think the shortage adds noise to this slow erosion (In either direction).

My point since last quarter has been that Intel is losing share in DC CPUs and saying it has a shortage at the same time.... and Q1 DC CPU units were down 5% YoY. it seems odd how this can happen.

On the positive side, I believe this will push Intel customers to Intel 3 and 18A which will simplify Intels roadmap.

the data can only mean Intel is losing DC shares faster than the total demand increase. Not good at all

 

[OneEng]

you are 100% correct. The math is clear and has been clear since 2019. There is a reason the world went to a foundry model and then that the foundry model contracted to one successful advanced foundry. math

Excel shows this very simply

The answer is a) do not do IDM2.0 or b) get 15B in external foundry business. everything else is a rationalization about why math and profit doesnt matter. One CEO chose a, One CEO chose b. they both 100% agreed on the options. LBT has figure out how to deal with "what if neither happens?"

Side fun excel spreadsheet experiment "if you are not already making money and you are losing money while trying to grow, you cannot catch up to have a positive return.

LOL. Yea, it doesn't help a company at all to keep changing tracks. Each track change takes time and money to do, and the company is generally losing money throughout the transition.

Market share numbers are in. AMD made record gains in DC CPU share..... and is selling units at higher ASP than Intel. Intel lost almost 5% share in a quarter and had a shortage of units (all DC CPUs are made at Intel Fabs).

The stock price went up 4x because IFS may have significant material revenue in 2029

AMD's "Server First" design decisions are a good strategy IMO. DC is the highest margin market, and the fastest growing market.

For Intel to lose 5% share in a quarter when the market is growing like a weed is a very bad sign IMO.

If you look at things from a technical perspective, the picture for Intel doesn't look great for some time. AMD will release Venice before Intel releases Granite Rapids (Oak Stream) on 18A, but it will be in 2027 (6 or more months behind Venice), and will not include SMT. Intel will release Clearwater Forest with 288 cores. Venice D with 256 cores and 512 threads (and more memory bandwidth) seems like more than a match to me though. Generally speaking in server loads, AMD's SMT has achieved 1.3-1.4x performance per core. A little simple math shows that Venice D is going to be a tough competitor. It also has double (I think) the PCI bandwidth for integration with AI processors (specifically AMD's MI400 stuff).

What I am guessing is that the roadmap would indicate that AMD will retain its advantage with the next generation of DC processors.

Now, I am thinking that Intel's Nova Lake is going to be quite impressive; however, gaining market share in a market that is rapidly declining like desktop or in a commodity market like laptop is a hard living to make when you are paying for premium lithography nodes.

As MK pointed out, both share and ASP relative to Intel has accelerated in the past quarter.

Continued terrible competitiveness by Intel DCAI.

Yea, the bottom line is that Intel pays premium prices for the process, and is losing share and margins. Not a great situation for team blue.

 

[hist78]

LBT has figure out how to deal with "what if neither happens?"

What could his approach be?

 

[hist78]

Intel gained share in desktop but also lost share in Notebooks.

But Intel’s Q1 2026 Client Computing Group (CCG) revenue is essentially flat compared to Q1 2025, and it has been declining since Q2 2025.

If the PC market is as hot as Intel describes, that strength is not reflected in CCG revenue. Intel attributed the weak results to capacity constraints at Intel Foundry. But there is another possibility: the PC market demand may not be as strong as Intel claims.

 

[siliconbruh999]

If the PC market is as hot as Intel describes, that strength is not reflected in CCG revenue. Intel attributed the weak results to capacity constraints at Intel Foundry. But there is another possibility: the PC market demand may not be as strong as Intel claims.

Everything circles back to Intel 7(Raptor Lake) Majority of Volume.

 

[hist78]

Everything circles back to Intel 7(Raptor Lake) Majority of Volume.

Yes, Intel 7 may be part of the reason, but it cannot be responsible for all the problems Intel is facing. I remember Intel’s former CEO Pat Gelsinger once saying that Intel 7 was neither technologically nor cost competitive. He expected that the arrival of Intel 4 and Intel 3 would resolve those issues. So why couldn’t Intel’s customers, Intel itself, or both transition to Intel 4 and Intel 3 as planned?

Is it because the yields of Intel 4 and Intel 3 were much lower than expected, the cost of Intel 4 and Intel 3 was even higher than the already expensive Intel 7, the performance and pricing of Intel 4/3 based products were not attractive, or something else?

Here is the Intel High Volume Manufacturing (HVM) timeline. By the end of 2025, Intel 4 and Intel 3 had been in HVM for about two years. Why didn’t they provide enough help for Intel to significantly increase revenue in such a booming market?

• Intel 7: Began HVM in 2020
• 
  
• Intel 4: Formally entered HVM in September 2023 at Fab 34 in Leixlip, Ireland
• 
  
• Intel 3: Entered HVM in late 2023/early 2024 at Fab 34 in Leixlip, Ireland

 

[siliconbruh999]

Yes, Intel 7 may be part of the reason, but it cannot be responsible for all the problems Intel is facing. I remember Intel’s former CEO Pat Gelsinger once saying that Intel 7 was neither technologically nor cost competitive. He expected that the arrival of Intel 4 and Intel 3 would resolve those issues. So why couldn’t Intel’s customers, Intel itself, or both transition to Intel 4 and Intel 3 as planned?

Is it because the yields of Intel 4 and Intel 3 were much lower than expected, the cost of Intel 4 and Intel 3 was even higher than the already expensive Intel 7, the performance and pricing of Intel 4/3 based products were not attractive, or something else?

Here is the Intel High Volume Manufacturing (HVM) timeline. By the end of 2025, Intel 4 and Intel 3 had been in HVM for about two years. Why didn’t they provide enough help for Intel to significantly increase revenue in such a booming market?

• Intel 7: Began HVM in 2020
• 
  
• Intel 4: Formally entered HVM in September 2023 at Fab 34 in Leixlip, Ireland
• 
  
• Intel 3: Entered HVM in late 2023/early 2024 at Fab 34 in Leixlip, Ireland

Intel 7 is a derivative of Intel 10nm which began HVM in 2019 with Ice Lake Intel had like years to ramp it into production compared to Intel 4/3

 

[MKWVentures]

MKWVentures said:
LBT has figure out how to deal with "what if neither happens?"

What could his approach be?

The team has looked at options.... so far none of them came to fruition. LBT is a smart and pragmatic man, I think he will find a way. I also still think looking at the IBM semiconductor manufacturing history for parallels may be insightful.

 

[siliconbruh999]

The team has looked at options.... so far none of them came to fruition. LBT is a smart and pragmatic man, I think he will find a way. I also still think looking at the IBM semiconductor manufacturing history for parallels may be insightful.

there are difference though Intel Yield and IBM's Yield are in different class that alone makes the process sellable

 

[Xebec]

But Intel’s Q1 2026 Client Computing Group (CCG) revenue is essentially flat compared to Q1 2025, and it has been declining since Q2 2025.

If the PC market is as hot as Intel describes, that strength is not reflected in CCG revenue. Intel attributed the weak results to capacity constraints at Intel Foundry. But there is another possibility: the PC market demand may not be as strong as Intel claims.

I think we're in an "Intel is stretching the truth" scenario - they're not completely wrong, but their claim is at least very misleading/verifiably false:

- Desktop Motherboard makers are reporting a drop in sales of boards in Q1, and are projecting drops between 22-37% YoY (overall 28%).

- Intel has produced more Arrow Lake-S chips than the market wants -- 270K Plus is down to $259 at Microcenter - and this is the full yield (24 core) ARL-S compute die.

- Arrow Lake-H also appears to be unavailable for corporate mobile mostly due to lack of RAM availability. (The OEMs are pushing Lunar Lake hard since each Lunar Lake chip already has RAM included, presumably bought at a lower price than current market).

- However, they *are* capacity constrained for Panther Lake on 18A, and Arrow Lake-U on Intel 3. This is based on what I've heard from a friend at a major corporation trying to buy PCs from major OEMs. Panther Lake appears to be 18A yield/ramp related with only limited availability in Q2, Arrow Lake-U's shortage appears to be Intel biasing wafers to Server.

..

Motherboard sales 'collapse' by more than 25% as chipmakers strangle enthusiast PC market to build more AI chips — Asus projected to sell 5 million fewer boards than 2025, Gigabyte, MSI, and ASRock also expected to see reduced sales numbers

Fewer people are buying parts and building new PCs from scratch.

www.tomshardware.com

(Source on mobile is "self" - via a friend who has been shopping around for Mobile Corporate PCs for a large corporation recently and working/quoting with 3 major OEMs)