The Wonderland of Intel Stock

[dkr1986]

3. From 2020 to 2024, Intel spent roughly $109 billion in CapEx. Did Intel allocate that money to the wrong areas?

Most of the money went into expanding the R&D facility in Oregon (D1X Mod 3), ramping Intel 7 in Fab 42 & Fab 28 (2020-2022), finishing Fab 34 in Ireland and ramping Intel 4 & 3 (2021-2023) and building Fab 52 & 62 in Arizona (2021-2025), finish the shell of Fab 38 in Israel (current status unknow - news is shell is finished) and starting the Ohio shells. Out of this it looks like part of the money was prematurely used up to order up tools for Arizona Fab 52 & 62 (all locked up in Assets under Construction) and on Israel's Fab 38.

It is just a timing issue in my opinion. their capital is locked up in nodes either in the process of ramping or will ramp in near future. Unfortunately, as I said earlier, they also took down some Intel 7 capacity which turned to be mistake now.

I would also highlight here that both Fabs 34 & 38 are announced under Bob Swan but scope was expanded & completed under Pat.

Intel to upgrade new Kiryat Gat fab with $15b investment

en.globes.co.il

https://www.hpcwire.com/off-the-wire/intels-ireland-milestone-first-tool-roll-in-at-fab-34/#:~:text=It%20is%20the%20first%20chipmaking,Credit%3A%20Intel%20Corporation

 

[MKWVentures]

That is because they took down Intel 7 capacity in Q3'24 to upgrade Fab 42 to support 18A (that upgrade I know didn't materialize under the Interim CO-CEO & new CEO). Intel 3 is capacity constrained because Lip Bu does not want to spend the necessary capex to add capacity without seeing demand increasing. It increased now out of nowhere, but it takes a while from ordering tools to installing tools to produce wafers. Explains the current predicament of Intel's capacity constraints. And I bet even Fab 52 is only partially tooled to save cost on the ramp of 18A. So even if CWF launched in 1Q'26, it will be capacity constrained for couple of Qtrs due to this.

"We will only build capacity if customer shows up" works for onboarding new foundry customers as they will commit at least 1 to 1.5 years ahead enough time to tool out & ramp a fab but that strategy does not work for sudden increase in demand like we are seeing for server CPUs now. It also does not work if there is no physical fab space (you know, the one that take 3+ years to build).


It kind of makes sense as most of their Intel 3 (same fab line as Intel 4 I suppose) is dedicated to GNR & SRF (also base die for CWF) which is large die size and low volume. So, on wafer volume basis EUV usage is still low, but it doubled from last year and so I guess we can expect it to more than double next year (Fab 34 only in 2025 vs Fab 34 & Fab 52 in 2026). On one hand Bob Swan's plan to outsource to TSMC bought them valuable time to tread water on Intel Product side while on the other hand it also is destroying the fab economics on the foundry side.

Good input on Intel 7. so what is Fab 42 doing?

18A loses money if you don't ramp it and loses even more if you do ramp it. It is a very cool technology with a high wafer cost and low yields. There is no answer to this paradox which is why successul companies outsource to TSMC.
CWF will take a long time to ramp. GR/SF wont peak until 2027.

Its not like Demand went up and Intel missed it and did not invest. Intel is not growing yet is constrained. Amazing

 

[MKWVentures]

10K has some good info on Capex at Fab 52/62 and the exact amount Intel spent. The partner has a say in whether to ramp or not on that as well. Intel pays fees if underloaded or if they hold inventory.

Which brings us back to the 2020 Corporate Strategic Discussions. Would the Intel BUs rather work with Intel Fabs or TSMC?

 

[dkr1986]

Good input on Intel 7. so what is Fab 42 doing?

I have no idea what is going on there now. In the last Q ER call, Dave was saying they are not looking to expand capacity on Intel 7 but this ER he said they are looking to expand capacity on Intel 7 too. So, I guess they may reequip it with Intel 7 again to improve Intel 7 capacity? I honestly don't know.

18A loses money if you don't ramp it and loses even more if you do ramp it. It is a very cool technology with a high wafer cost and low yields. There is no answer to this paradox which is why successul companies outsource to TSMC.

So only TSMC can & is capable of making these advanced node process in this world? Are you saying LBT & DZ are lying now on investor calls and conferences when they say 18A has better cost structure than their older nodes? That they are doing federal securities fraud?

CWF will take a long time to ramp. GR/SF wont peak until 2027.

I agree but CWF could have been incremental capacity to service this sudden jump in demand.

Its not like Demand went up and Intel missed it and did not invest. Intel is not growing yet is constrained. Amazing

Isn't that exactly what happened? DCAI revenue is up 14.6% QoQ and it was up 5% QoQ on last quarter too. That shows accelerating demand for Intel CPUs.
Intel is saying for Q1'26 they don't have the finished server CPU inventory (they had in 2025) to meet this increased demand & they will have to reallocate some of the client CPU capacity to server (although not fully vacate it). So DCAI still decreases next Q as they don't have that inventory buffer. Before someone here brings up the $11B inventory in BS, only $4B is finished goods and could mostly be client related CPUs.

 

[hist78]

Most of the money went into expanding the R&D facility in Oregon (D1X Mod 3), ramping Intel 7 in Fab 42 & Fab 28 (2020-2022), finishing Fab 34 in Ireland and ramping Intel 4 & 3 (2021-2023) and building Fab 42 & 52 in Arizona (2021-2025), finish the shell of Fab 38 in Israel (current status unknow - news is shell is finished) and starting the Ohio shells. Out of this it looks like part of the money was prematurely used up to order up tools for Arizona Fab 52 & 62 (all locked up in Assets under Construction) and on Israel's Fab 38.

It is just a timing issue in my opinion. their capital is locked up in nodes either in the process of ramping or will ramp in near future. Unfortunately, as I said earlier, they also took down some Intel 7 capacity which turned to be mistake now.

I would also highlight here that both Fabs 34 & 38 are announced under Bob Swan but scope was expanded & completed under Pat.

Intel to upgrade new Kiryat Gat fab with $15b investment

en.globes.co.il

https://www.hpcwire.com/off-the-wire/intels-ireland-milestone-first-tool-roll-in-at-fab-34/#:~:text=It%20is%20the%20first%20chipmaking,Credit%3A%20Intel%20Corporation

Those are valid reasons. But after spending a massive $109 billion in capital expenditures (2020–2024) and $78.87 billion on R&D over the same period, Intel still cannot generate net profit, cannot stop its annual revenue from declining, and cannot even meet the internal needs from Intel Products. If that’s the case, what is the point of discussing IFS serving external customers?

In the semiconductor industry, being six months late often means losing product viability and losing customers or market share. Intel’s customers and competitors did not wait, and will not wait, for Intel to resolve its timing, execution, and products problems.

 

[MKWVentures]

So only TSMC can & is capable of making these advanced node process in this world? Are you saying LBT & DZ are lying now on investor calls and conferences when they say 18A has better cost structure than their older nodes? That they are doing federal securities fraud?

18A does not have a better margins today or throughout most of 2026. Margins go down the more 18A ramps. Margins on 18A are negative today, and much worse than other nodes throughout most of 2026. In theory, if they get the outs per tool, to goal get yields on target, and fully ramp 2 fabs.... maybe GMs are 50%... that is 2028 timeframe best case. all wafers costs at Intel are >30% higher than prediction once they start running in HVM. Also: IFS will not break even in 2027, Intel will not be number 2 external foundry in 2030, Intel did not launch Panther lake in 2025. Alternative Facts....

Isn't that exactly what happened? DCAI revenue is up 14.6% QoQ and it was up 5% QoQ on last quarter too. That shows accelerating demand for Intel CPUs.
Intel is saying for Q1'26 they don't have the finished server CPU inventory (they had in 2025) to meet this increased demand & they will have to reallocate some of the client CPU capacity to server (although not fully vacate it). So DCAI still decreases next Q as they don't have that inventory buffer. Before someone here brings up the $11B inventory in BS, only $4B is finished goods and could mostly be client related CPUs.

Sort of: DCAI went up... But as Stacy Rasgon mentioned.... lets push on that a little:
1) So revenue is dropping Q1 QoQ. but you are moving wafers to DCAI which should have higher margins and Revenue/wafers. But revenue is lower and lower than average of last 4 quarters? and Margins are lower? was the previous forecast that revenue would drop 10% in 2026? Reality: Intel 7 is what people want, they do not want GR, SF, Meteor lake. Panther lake was delayed and ramp slowed due to "issues". Intel did not see that coming. So now there is blow back to older nodes and as you stated, Intel 7 is limited.

Lets see what happens to AMD, units sold, and DCAI market share

 

[dkr1986]

18A does not have a better margins today or throughout most of 2026. Margins go down the more 18A ramps. Margins on 18A are negative today, and much worse than other nodes throughout most of 2026. In theory, if they get the outs per tool, to goal get yields on target, and fully ramp 2 fabs.... maybe GMs are 50%... that is 2028 timeframe best case. Also: IFS will not break even in 2027, Intel will not be number 2 external foundry in 2030,

I guess we have to wait and see how this unfolds.

Intel did not launch Panther lake in 2025.

Intel did deliver the Panther Lake SKUs to their OEMs in 2025 so that Panther Lake laptops can be pre-ordered in Jan 6, 2026, and that is what they promised imo. There are recordings going back to early 2023 of Pat saying panther lake is mostly 2026 volume even though they expect it to launch in late 2025. A little delayed, I agree.

1) So revenue is dropping Q1 QoQ. but you are moving wafers to DCAI which should have higher margins and Revenue/wafers. But revenue is lower and lower than average of last 4 quarters? and Margins are lower? was the previous forecast that revenue would drop 10% in 2026?

If you look at the OM for both the Client & DCAI, it is actually CCG that commanded better margins so far. With AMD's competitiveness in DC space, I don't think its too out of the ordinary to expect Intel margins & revenue will be pressured even if the units shipped are increasing. This is about making sure to capture the socket not near-term revenue & margins. If they lose this socket wins now, AMD will have more leverage on the next upgrade cycle.

 

[hist78]

18A loses money if you don't ramp it and loses even more if you do ramp it. It is a very cool technology with a high wafer cost and low yields. There is no answer to this paradox which is why successul companies outsource to TSMC.

Is Intel pushing both Backside Power Delivery and GAA into 18A a major factor for this paradox?

 

[hist78]

DCAI revenue is up 14.6% QoQ and it was up 5% QoQ on last quarter too. That shows accelerating demand for Intel CPUs.

18A does not have a better margins today or throughout most of 2026. Margins go down the more 18A ramps. Margins on 18A are negative today, and much worse than other nodes throughout most of 2026. In theory, if they get the outs per tool, to goal get yields on target, and fully ramp 2 fabs.... maybe GMs are 50%... that is 2028 timeframe best case. all wafers costs at Intel are >30% higher than prediction once they start running in HVM. Also: IFS will not break even in 2027, Intel will not be number 2 external foundry in 2030, Intel did not launch Panther lake in 2025. Alternative Facts....


Sort of: DCAI went up... But as Stacy Rasgon mentioned.... lets push on that a little:
1) So revenue is dropping Q1 QoQ. but you are moving wafers to DCAI which should have higher margins and Revenue/wafers. But revenue is lower and lower than average of last 4 quarters? and Margins are lower? was the previous forecast that revenue would drop 10% in 2026? Reality: Intel 7 is what people want, they do not want GR, SF, Meteor lake. Panther lake was delayed and ramp slowed due to "issues". Intel did not see that coming. So now there is blow back to older nodes and as you stated, Intel 7 is limited.

Lets see what happens to AMD, units sold, and DCAI market share

The revenue growth and possibly volume growth of DCAI are stalled, considering the current high demand market situation. I consolidated data from multiple Intel quarterly earnings materials to create the following tables. Something terrible seems to have happened at Intel DCAI division.

Comparing DCAI 2025 revenue with 2024, Intel was only able to grow about 4%. How can this happen in such a hot market?

If we compare each quarter with the same quarter in the previous year, the revenue growth is again stalled or very limited. This should not be happening under such hot market conditions.

DCAI’s operating margins did improve compared with 2024. Does that mean Intel intentionally chose not to make or ship too many DCAI products (in terms of units)? Or customers hesitate to buy more from Intel? Or Intel just can't make what customers want to buy?

 

[MKWVentures]

The revenue growth and possibly volume growth of DCAI are stalled, considering the current high demand market situation. I consolidated data from multiple Intel quarterly earnings materials to create the following tables. Something terrible seems to have happened at Intel DCAI division.

I think @dkr1986 hit on some things and Aaron Rakers published a summary. DC demand grew a lot in the 2H2025 and I guess it was a surprise. Aarons take is that YoY growth went from like 2% to 15%. Then there are some pricing changes

My concerns are:
Intel should have capacity for everything. they are literally the slowest growing company in Compute.
I heard that Intel was chopping pricing to get share (rumor). I can still remember a BU manager telling me "we are doing a price cut to grow agressively, you better get me my units" ... sounds like maybe they couldnt get the units.
Then there is the whole margins are not higher on DCAI problem. It could be LBT wants DCAI to get future sockets so in theory you could redirect wafers even though margins/revenue are worse.

Back to my original comment: When you are struggling, and growth suddenly shows up, it is unacceptable not to have capacity.

AMD might give us info on what the truth is.

 

[blueone]

Comparing DCAI 2025 revenue with 2024, Intel was only able to grow about 4%. How can this happen in such a hot market?

Cloud computing companies represent about 40% of server CPU volume annually. All three major companies (AWS, Google Cloud, and Azure) are producing their own CPUs in increasing volumes, mostly for internal applications, but these still represent millions of units in a total server CPU volume that is probably about 25 million units. (The cloud companies don't publish volume numbers for their internal CPU designs.) This means the merchant CPU market is not growing as fast.

Oracle is a big user of Ampere CPUs. AMD currently has 28% market share, and growing faster than Intel. Cloud computing is growing faster than enterprise servers, so the market for Intel server CPUs is probably shrinking or only growing very slowly.

While Nvidia connects to Intel CPUs, most of the CPU volume they drive is in their own designs.

In the current environment, 4% growth doesn't surprise me at all.

 

[Xebec]

They are short on Intel 7 somehow, which actually has very good yields. Only 10% of their output is EUV.

They are replacing a decent amount of Raptor Lake silicon for warranty..

 

[MKWVentures]

They are replacing a decent amount of Raptor Lake silicon for warranty..

I have seen estimates on this, I didn't think it was material and OEMs didnt seem concerned.

how many units do you think are being replaced per quarter. ... what is a decent amount.

 

[blueone]

Cloud computing companies represent about 40% of server CPU volume annually. All three major companies (AWS, Google Cloud, and Azure) are producing their own CPUs in increasing volumes, mostly for internal applications, but these still represent millions of units in a total server CPU volume that is probably about 25 million units. (The cloud companies don't publish volume numbers for their internal CPU designs.) This means the merchant CPU market is not growing as fast.

Oracle is a big user of Ampere CPUs. AMD currently has 28% market share, and growing faster than Intel. Cloud computing is growing faster than enterprise servers, so the market for Intel server CPUs is probably shrinking or only growing very slowly.

While Nvidia connects to Intel CPUs, most of the CPU volume they drive is in their own designs.

In the current environment, 4% growth doesn't surprise me at all.

The Wall Street Journal has a different take on this 4% issue than I did. They propose it was due to not being able to meet demand for older server CPUs:

https://www.wsj.com/tech/intel-problems-trump-bump-17d2c941?st=w9gagx&reflink=desktopwebshare_permalink

The company has also been seeking to cut costs by capping spending on older technology and being more cautious about how it expanded manufacturing capacity for its latest chips.

To date, the focus of much AI spending has been graphics processing units, or GPUs—the type of processors that designers like Nvidia and Advanced Micro Devices specialize in—which allow software developers to accelerate computing by performing billions of tasks in parallel.

Over the past year, Intel has retired expensive tools used to make its older generations of data center CPUs, including those from its Emerald Rapids and Granite Rapids series.

But over the course of the second half of 2025, companies such as OpenAI, Amazon Web Services and Google began to realize that deploying generative AI models required more and better CPUs, which act as the central computing brain of most servers, than they had initially thought.

Suddenly, Intel was fielding requests to buy thousands of older CPUs. But because it had taken so much manufacturing capacity offline, the company was unable to meet that demand.

OK, this is weird. Since when do cloud computing companies buy "more and better CPUs", but they placed orders for "thousands of older CPUs"? Why would anyone want previous generation CPUs in quantity? And "thousands" of CPUs wouldn't move the revenue needle at all, so the orders must have been at least hundreds of thousands of CPUs, per company, meaning there's some sort of common demand... based on what? A shortage of AMD supply, and the need for x86 CPUs?

The financial press continues to amaze me with their lack of coherence in the semiconductor field.