Intel Reallocates PC Production Capacity to Server CPUs

[erek]

Probably for Ai

“However, hyperscalers have since been ordering larger quantities, leading to a temporary shortage of Xeon processors for the Data Center and AI (DCAI) group. To address this, Intel plans to redirect some resources from the Client Computing Group (CCG) to support Xeon CPU production.

Intel CFO David ZinsnerBoth (DCAI and CCG) will be down as a function of supply. Obviously, we're shifting as much as we can over to data center to meet the high demand, but we can't completely vacate the client market. So we're trying to support both as best we can and obviously work our way out of this supply issue. I do believe that the first quarter is the trough. We will improve supply in the second quarter. Part of the challenge is that in the third and fourth quarter of 2025, we lived off of supply, but we also had a reasonable chunk of finished goods inventory to also work through. Unfortunately, that is now down to kind of 40% of what it was at peak levels.

Intel manufactures its latest Xeon 6 "Granite Rapids" and "Sierra Forest" processors at its Intel Foundry sites, using Intel 7 and Intel 3 nodes. Production on these nodes is now being redirected to the DCAI group, away from the CCG. Interestingly, CCG's newest "Panther Lake" CPU is made on the 18A node, so these supply redirects won't affect the latest product lineup. However, for other lineups, we might see lower availability as inventory decreases. Some client CPUs like "Arrow Lake" are manufactured externally at TSMC's fabs, so they won't be affected by these changes.”

Source: https://www.techpowerup.com/345535/...pacity-to-server-cpus-amid-tight-wafer-supply

 

[erek]

Probably for Ai

“However, hyperscalers have since been ordering larger quantities, leading to a temporary shortage of Xeon processors for the Data Center and AI (DCAI) group. To address this, Intel plans to redirect some resources from the Client Computing Group (CCG) to support Xeon CPU production.


Intel manufactures its latest Xeon 6 "Granite Rapids" and "Sierra Forest" processors at its Intel Foundry sites, using Intel 7 and Intel 3 nodes. Production on these nodes is now being redirected to the DCAI group, away from the CCG. Interestingly, CCG's newest "Panther Lake" CPU is made on the 18A node, so these supply redirects won't affect the latest product lineup. However, for other lineups, we might see lower availability as inventory decreases. Some client CPUs like "Arrow Lake" are manufactured externally at TSMC's fabs, so they won't be affected by these changes.”

Source: https://www.techpowerup.com/345535/...pacity-to-server-cpus-amid-tight-wafer-supply

“Intel Struggles To Meet AI Data Center Demand​ BeauHD 10 hours ago

17

Intel says it struggled to satisfy demand for its AI data-center CPUs while new PC chips squeeze margins. CEO Lip-Bu Tan framed the turnaround as supply-constrained, not demand-constrained, with manufacturing yields (18A) improving but still below targets. Reuters reports: The forecast underscores the difficulties faced by Intel in predicting global chip markets, where the company's current products are the result of decisions made years ago. The company, whose shares have risen 40% in the past month, recently launched a long-awaited laptop chip designed to reclaim its lead in personal computers just as a memory chip crunch is expected to depress sales across that industry. Meanwhile, Intel executives said the company was caught off guard by surging demand for server central processors that accompany AI chips. Despite running its factories at capacity, Intel cannot keep up with demand for the chips, leaving profitable data center sales on the table while the new PC chip squeezes its margins. "In the short term, I'm disappointed that we are not able "to fully meet the demand in our markets," Chief Executive Officer Lip-Bu Tan told analysts on a conference call. The company forecast current-quarter revenue between $11.7 billion and $12.7 billion, compared with analysts' average estimate of $12.51 billion, according to data compiled by LSEG. It expects adjusted earnings per share to break even in the first quarter, compared with expectations of adjusted earnings of 5 cents per share.”

 

[uOpt]

That's not going to work out for them either. True, there will be more music in the server market.

But in the consumer segment Intel is at least competitive with AMD on both performance and price. In the server segment they are not, EPYC spanks them.

 

[erek]

That's not going to work out for them either. True, there will be more music in the server market.

But in the consumer segment Intel is at least competitive with AMD on both performance and price. In the server segment they are not, EPYC spanks them.

you’re right

is interesting the CFO said they’re shifting as much as possible too,

“Responding to a question about weaker-than-expected DCAI outlook, David Zinsner, Chief Financial Officer at Intel, said: "We're shifting as much as we can over to the data center... we can't completely vacate the client market." Intel CEO Lip-Bu Tan also highlighted shortages affecting other areas, saying, "The industry is facing a very big challenge, you know, the memory constraints and the pricing."”
https://www.tomshardware.com/pc-com...ll-on-track-for-late-2026-release-14a-in-2028

“Intel Shares Fall on Swing to Loss​

Troubled chip maker says it hadn’t adequately anticipated demand from AI data centers”​

https://www.wsj.com/business/earnings/intel-intl-q4-earnings-report-2025-9271b096

 

[erek]

“Intel Shares Fall on Swing to Loss​

Troubled chip maker says it hadn’t adequately anticipated demand from AI data centers”​

https://www.wsj.com/business/earnings/intel-intl-q4-earnings-report-2025-9271b096

View: https://www.youtube.com/watch?v=dJSa2BOKrYY

down almost 16% already

 

[Aurelius]

That's not going to work out for them either. True, there will be more music in the server market.

But in the consumer segment Intel is at least competitive with AMD on both performance and price. In the server segment they are not, EPYC spanks them.

Right now it's not so much about competitiveness as simply having parts available. If you need AI computing power in the near term, you're not going to wait for Epyc-based servers to come back in stock; you're going to buy a pile of Xeon servers and make it work.

Intel does have to catch up in performance and value, but it has the luxury of insatiable demand until Clearwater Forest (the first Xeon based on 18A) arrives in the next few months.

 

[LukeTbk]

that a good spin, that one way to present the issue of not enough Intel 3 because intel 20 never came online and bad yield on Intel 18.... we have so much demand that our q4 2025 revenues are down and 2026 revenues will be less than expected...

Can see advanced packaging used for their new laptop using their Intel 3 node wafer for the base and having to fight for capacity, but would not surprise me if Intel 18 yield does not make it that much of an issue.

 

[Zarathustra[H]]

I mean, yes, they are probably going to the AI segment, but I don't think Intel is driving this.

I imagine with enthusiasts unable to afford RAM, GPU's and storage they are likely anticipating shrinking demand for their consumer products and getting ahead of it.

 

[LukeTbk]

If it was not production and yield issue, a shift (toward higher priced item) should not mean reduced revenues.

They are having trouble reaching volume on their new nodes/packaging capacity; it is a painful transition, would be my guess in 2025 they were a bigger client for TSMC in dollars than AMD; that's how much trouble they have making stuff.

Intel revenues could grow by only 3% as a chips and silicon maker during the peak of a silicon boom.... (AMD could do 35% again like last year, TSMC ~30% has well, Nvidia 50% they were ~115% in 2025), Panther Lake numbers look good but if they cost a fortune to make a very small amount of them.. you need to combined both side of the equation at some points

 

[erek]

View: https://www.youtube.com/watch?v=dJSa2BOKrYY

down almost 16% already

View attachment 780921

down almost 17% as of writing this comment

 

[ChadD]

That's not going to work out for them either. True, there will be more music in the server market.

But in the consumer segment Intel is at least competitive with AMD on both performance and price. In the server segment they are not, EPYC spanks them.

Intel has some attractive E core server parts... for AI markets. They aren't looking for RAW CPU compute. They are looking for efficiency. For primarily GPU compute servers pairing them with low power Intel chips is a good value. Don't get me wrong I think AMD has proven to be as efficient as Intel for the most part. Though generally you have to factor in "perf per watt" for AMD to really look good. No doubt AMD has the best performing server offerings PER watt. A lot of AI farms though just need them to drive the GPU compute racks... they aren't looking for ultimate performance. Intel still wins overall power draw if your doing light CPU lifting which is what they are all doing... its why ARM is just as popular as x86 for those use cases. (not just because NV is pushing ARM). Crank the CPUs up AMD wins perf per watt. But if the CPUs are running at 40% load.... Intel sucks less juice.

And of course on top of that no one can meet their orders right now including AMD. For some customers if the choice is take a bunch of 6980P today, or wait 6 months for AMD to ship. Intel becomes a fine choice.

https://www.phoronix.com/review/intel-xeon-6980p-power/7

 

[uOpt]

I mean, yes, they are probably going to the AI segment, but I don't think Intel is driving this.

I imagine with enthusiasts unable to afford RAM, GPU's and storage they are likely anticipating shrinking demand for their consumer products and getting ahead of it.

Sadly, I have to agree with this view.

 

[LukeTbk]

Intel has some attractive E core server parts... for AI markets. They aren't looking for RAW CPU compute. They are looking for efficiency.

e-core is more for raw compute density, not really efficiency; at least historically it was all about putting ~3/4 cores in the space a p-core use, but with a less efficient core (running a p-core at the same frequency at the same speed was usually more efficient), it is space efficacy.

For simple webserver the last version now shift that a little bit, it can still win efficacy by a little bit over pcore but I doubt for the AI type task efficacy.

In the server segment they are not, EPYC spanks them.

for some reason Xeon is still popular for raking GPU instead of epyc, for one they often offer higher bandwith per memory channel (even if they have less of them) at lower latency which sometime is important not just total for example, Intel AMX, their PCIe controller long history and validated use of PLX switch low latency versus epyc io die chiplet, big clients run their own affair with their own benchmark, software stack/tooling, looking at generic tooling/workload benchmark on phoronix would not tell all the story.

when you look at multi-gpu benchmark, Xeon seem still popular:
https://mlcommons.org/benchmarks/training/

and that would not necesasrily be just for legacy, they are quite new workload and system in some case, Intel 72% share in the server market is not just some big market failure, they are still the best option in many case we should presume.

 

[philb2]

That's not going to work out for them either. True, there will be more music in the server market.

Meaning?

In the server segment they are not, EPYC spanks them.

How do you know that?

 

[HockeyJon]

“Intel Shares Fall on Swing to Loss

Troubled chip maker says it hadn’t adequately anticipated demand from AI data centers”​

https://www.wsj.com/business/earnings/intel-intl-q4-earnings-report-2025-9271b096

This is probably the most Intel thing to ever Intel. Just totally underscores how the company has been run for the past decade. There's been an AI boom going on for a couple of years, and here's Intel sitting there being like "I think we have enough datacenters already, who could possibly want another one?".

 

[erek]

This is probably the most Intel thing to ever Intel. Just totally underscores how the company has been run for the past decade. There's been an AI boom going on for a couple of years, and here's Intel sitting there being like "I think we have enough datacenters already, who could possibly want another one?".

“Why Did Intel Lose Its Crown?​

Intel’s Data Center & AI share fell for one primary reason amidst repeated delays in its 2021 and 2022 CPU chip iterations.

Intel was CPU-focused while competition intensified, and after ChatGPT’s launch (Q4 2022), data-center spending shifted toward GPU-heavy AI systems.

The company failed to adapt and scale, as its AI-chip deals fell short of initial expectations.

Management even dropped its 2024 target of $500M+ in AI-accelerator revenue, citing a software platform transition.”


https://www.visualcapitalist.com/charted-the-battle-for-ai-data-center-revenue-2021-2025/

 

[LukeTbk]

This is probably the most Intel thing to ever Intel. Just totally underscores how the company has been run for the past decade. There's been an AI boom going on for a couple of years, and here's Intel sitting there being like "I think we have enough datacenters already, who could possibly want another one?".

could be more a PR spin, Intel 18 having good yield, 20 being an mature node right now with 14 right around the corner:

There was an ambitious risking to have low volume but trying to get to the world best wafer quick that looked like this:

Intel 20A RibbonFET / PowerVia Late 2024
Intel 18A Refined Ribbon FET 2025
Intel 14A High-NA EUV 2026

Intel 20 never worked, intel 18A has terrible yield, Intel 3 that used for advanced packaging and data center Xeon is overcrowded as they only working advanced node and can imagine they underbuild it overscoring its successor not underscoring the obvious datacenter build up.

Intel 3 that should be used for panther lake (laptop cpu) packaging, is overcrowded to make xeons, not because there is special amount of them, but because they were able to make less wafer in 2025 than in 2022.

Making the conversation less about deep manufacturing issue (that are extremelly hard to turn around) to be about ultra easy take a day to change demand prognostic look much better for the future.

Not even sure the high xeon demand is true to start with, Data center sales last quarter were only 4.7 billion; that's not a business week of business for Nvidia.

When you look at this:

vs now

Are they really hit by a special and surprising amount of orders while having way less revenues than 4 year ago in that category?

Could have been like if last year they would have tried to say they are spending a fortune on TSMC capacity because they did not anticipate the giant demand for their gpu or laptop cpu.... people would have rapidly thought that it was because their own attempt at making volume state of the art production had issues and failed, not some surprise level of demands of Intel products.

 

[Lakados]

could be more a PR spin, Intel 18 having good yield, 20 being an mature node right now with 14 right around the corner:

Intel 20A RibbonFET / PowerVia Late 2024
Intel 18A Refined Ribbon FET 2025
Intel 14A High-NA EUV 2026

Intel 20 never worked, intel 18A has terrible yield, Intel 3 that used for advanced packaging and data center Xeon is overcrowded as they only working advanced node and can imagine they underbuild it overscoring its successor not underscoring the obvious datacenter build up.

Intel 3 that should be used for panther lake (laptop cpu) packaging, is overcrowded to make xeons, not because there is special amount of them, but because they were able to make less wafer in 2025 than in 2022.

Making the conversation less about deep manufacturing issue (that are extremelly hard to turn around) to be about ultra easy take a day to change demand prognostic look much better for the future.

Not even sure the high xeon demand is true to start with, Data center sales last quarter were only 4.7 billion; that's not a business week of business for Nvidia.

When you look at this:
View attachment 780991 vs now View attachment 780992

Are they really hit by a special and surprising amount of orders while having way less revenues than 4 year ago in that category?

Intel also failed to remember about a refresh cycle, AI is all hot and heavy sure but Intel straight up forgot about the datacenter renewals that aren’t AI, they do large purchases every 4-5 years as they upgrade entire cluster rows, Intel failed to allocate enough capacity for them.

 

[HockeyJon]

could be more a PR spin, Intel 18 having good yield, 20 being an mature node right now with 14 right around the corner:

Intel 20A RibbonFET / PowerVia Late 2024
Intel 18A Refined Ribbon FET 2025
Intel 14A High-NA EUV 2026

Intel 20 never worked, intel 18A has terrible yield, Intel 3 that used for advanced packaging and data center Xeon is overcrowded as they only working advanced node and can imagine they underbuild it overscoring its successor not underscoring the obvious datacenter build up.

Intel 3 that should be used for panther lake (laptop cpu) packaging, is overcrowded to make xeons, not because there is special amount of them, but because they were able to make less wafer in 2025 than in 2022.

Making the conversation less about deep manufacturing issue (that are extremelly hard to turn around) to be about ultra easy take a day to change demand prognostic look much better for the future.

Not even sure the high xeon demand is true to start with, Data center sales last quarter were only 4.7 billion; that's not a business week of business for Nvidia.

When you look at this:
View attachment 780991 vs now View attachment 780992

Are they really hit by a special and surprising amount of orders while having way less revenues than 4 year ago in that category?

Yes, that's a very good point indeed.

 

[erek]

could be more a PR spin, Intel 18 having good yield, 20 being an mature node right now with 14 right around the corner:

Intel 20A RibbonFET / PowerVia Late 2024
Intel 18A Refined Ribbon FET 2025
Intel 14A High-NA EUV 2026

Intel 20 never worked, intel 18A has terrible yield, Intel 3 that used for advanced packaging and data center Xeon is overcrowded as they only working advanced node and can imagine they underbuild it overscoring its successor not underscoring the obvious datacenter build up.

Intel 3 that should be used for panther lake (laptop cpu) packaging, is overcrowded to make xeons, not because there is special amount of them, but because they were able to make less wafer in 2025 than in 2022.

Making the conversation less about deep manufacturing issue (that are extremelly hard to turn around) to be about ultra easy take a day to change demand prognostic look much better for the future.

Not even sure the high xeon demand is true to start with, Data center sales last quarter were only 4.7 billion; that's not a business week of business for Nvidia.

When you look at this:
View attachment 780991 vs now View attachment 780992

Are they really hit by a special and surprising amount of orders while having way less revenues than 4 year ago in that category?

“Intel Confirms 14A Wafer Capacity Expansion Only Follows Customer Commitments​

by AleksandarK Today, 04:00 Discuss (5 Comments)
Intel's concluded its Q4 and full-year 2025 earnings call, revealing interesting things about its foundry operations and node selection. When talking about the upcoming 14A node, Intel confirmed that the node is on track for 2027, with yields that are in good shape. However, the bulk of the 14A external capacity Intel will dedicate depends on customers willing to sign manufacturing contracts first, before Intel spends insane amount of funds required to expand manufacturing capacity. Intel CEO Lib-Bu Tan noted "Engagements with potential external customers on Intel 14A are active. We believe customers will begin to make firm supplier decisions starting in the second half of this year and extending into the first half of 2027. We also have the opportunity to provide strong differentiation in advanced packaging, particularly with EMIB and EMIB-T."

Intel CFO David Zinsner has stated "On 14A, Lip-Bu has been very direct with us on all of this. He does not want to spend on capacity on 14A, only spend on the kind of TD spend or R&D spend associated with 14A, even in the fab until we have customers secured. We've talked about the likelihood is our customers on 14A, their window to secure, or for us to secure them will be in the back half of this year and in the first half of next year. And so once visibility improves there, we'll start to unlock the spend on 14A. I can just add a little bit more. I think on the yield improvement, we see 7%, 8% yield improvement per month." Intel has previously mentioned that while there will be a base capacity for internal use, any future expansion for third-party services will depend on customers' willingness to invest capital with Intel Foundry.“

 

[erek]

 

[erek]

That's not going to work out for them either. True, there will be more music in the server market.

But in the consumer segment Intel is at least competitive with AMD on both performance and price. In the server segment they are not, EPYC spanks them.

when i was saying you’re right, it’s this article i was looking for

“Intel Server & Client CPU Market Share Has Seen A Huge Drop Since The Launch of AMD Ryzen & EPYC Chips​

AMD's EPYC processors have helped the company move forward, driving strong adoption among cloud providers and enterprise customers that are seeking higher core counts and stronger performance-per-watt. That said, the client segment for Intel looks even weaker. After gaining a slight edge in the market share in 2016-2017, Intel has been going down since the release of the Zen architecture. With the release of newer Zen architectures and X3D chips, Intel has lost significant CPU shares in both the desktop and notebook segments.

You can see Intel's market share starting to fall around 2017, which is when the first Ryzen CPUs were introduced. Intel did manage to crunch away some share back with its 12th Gen Alder Lake and the 13th Gen Raptor Lake lineup, but due to the degrading issues on 13th and 14th gen chips, a lot of DIY & OEM markets switched over to Ryzen, which is why AMD's share in the desktop segment has seen a major uplift.“

https://wccftech.com/intel-server-c...ge-drop-since-launch-of-amd-ryzen-epyc-chips/

 

[Balkroth]

Intel has some attractive E core server parts... for AI markets. They aren't looking for RAW CPU compute. They are looking for efficiency. For primarily GPU compute servers pairing them with low power Intel chips is a good value. Don't get me wrong I think AMD has proven to be as efficient as Intel for the most part. Though generally you have to factor in "perf per watt" for AMD to really look good. No doubt AMD has the best performing server offerings PER watt. A lot of AI farms though just need them to drive the GPU compute racks... they aren't looking for ultimate performance. Intel still wins overall power draw if your doing light CPU lifting which is what they are all doing... its why ARM is just as popular as x86 for those use cases. (not just because NV is pushing ARM). Crank the CPUs up AMD wins perf per watt. But if the CPUs are running at 40% load.... Intel sucks less juice.

And of course on top of that no one can meet their orders right now including AMD. For some customers if the choice is take a bunch of 6980P today, or wait 6 months for AMD to ship. Intel becomes a fine choice.

https://www.phoronix.com/review/intel-xeon-6980p-power/7

The Intel 6 Sierra Forest parts don't particularly get used in a lot of the AI racks/Designs I've seen. I really do like them , and they are great for shoving a lot of compute in a single slot. See them a lot on the networking side actually. I use the 6980P processors in a single socket 1 RU server configuration for vCMTS/vBNG stuff, where you can basically replace a 2 socket Sapphire rapids system, get better performance (Bandwidth in this instance) at less power. There are some really good use cases for the SRF chips for workloads that can be split up between cores properly where they can take advantage of it being a real core and not just a HT. It's actually to the point with them that it just doesn't make sense to use QAT offload for encryption with the cores any more.

 

[sleepeeg3]

I imagine with enthusiasts unable to afford RAM, GPU's and storage they are likely anticipating shrinking demand for their consumer products and getting ahead of it.

It's probably more enterprise business, before consumer / retail, but you're probably right about the pricing. FWIW, enthusiasts aren't their market. We are influencers - not near-term market movers.

FWIW, I almost bought an Arrow Lake over Ryzen. Price/performance was superior to AMD (at least the deal at Microcenter - bundle was $170 cheaper than the next best from AMD). Even if you're gaming, AMD's X3D cache benefit doesn't do much at high/gaming resolutions - overrated. The only reason I went AMD was being able to run 2 x NVME's in PCI-E 5.0.

 

[uOpt]

Intel has some attractive E core server parts... for AI markets. They aren't looking for RAW CPU compute. They are looking for efficiency. For primarily GPU compute servers pairing them with low power Intel chips is a good value. Don't get me wrong I think AMD has proven to be as efficient as Intel for the most part. Though generally you have to factor in "perf per watt" for AMD to really look good. No doubt AMD has the best performing server offerings PER watt. A lot of AI farms though just need them to drive the GPU compute racks... they aren't looking for ultimate performance. Intel still wins overall power draw if your doing light CPU lifting which is what they are all doing... its why ARM is just as popular as x86 for those use cases. (not just because NV is pushing ARM). Crank the CPUs up AMD wins perf per watt. But if the CPUs are running at 40% load.... Intel sucks less juice.

And of course on top of that no one can meet their orders right now including AMD. For some customers if the choice is take a bunch of 6980P today, or wait 6 months for AMD to ship. Intel becomes a fine choice.

https://www.phoronix.com/review/intel-xeon-6980p-power/7

Yeah, you have a point there. Software tooling is still better on Intel. Especially profiling tools.

To the best of my knowledge, for example, there is no tool on any AMD platform to display the PCIe bus utilization. Intel provides that in their toolset.

I suppose this particular detail can be important when you want to run multiple GPUs.

 

[LukeTbk]

when i was saying you’re right, it’s this article i was looking for

well like the article say, down but still 72% of the x86 server share and 61% of the revenues, would they ever get there manufacturing cost side of things in orders... they should be profitable with that market share of a growing market, they do not need to gain anything back.

 

[philb2]

Yes, that's a very good point indeed.

And very predictable, if you sold the customer the current CPUs. Plus for non-Intel CPUs, it would be easy enough to ID a decent share of that. So you could easily build a forecast of future orders, based on the current installs, plus some assumptions about refresh cycles, for which you would also have history. Did Intel not do that? That should be grounds for a shareholder lawsuit. And it's marketing malfeasance.