Lip-Bu Tan Update on 18A and 14A

[siliconbruh999]

Zinsner said Intel 7 cost and 18A cost are the same a while ago... things didnt work out like he planned.. Lets see what he says next week. 18A wafer cost is not I7 wafer cost

Also the 18A volume is still low .... it is not any where near 40K wafer per month at Fab 52. and then we have the whole "Intel is only 51% owner of Fab 52"... I have no idea how they will report this "non controlling income"

the real question will be how much of the 18A financial issue will go to product group and how much will go to IFS. OR will intel change the accounting yet again???? looking forward to see what happens.

Upto Intel where they want it to be.

If you can ask questions of Intel at earnings:
" are 18A PRODUCT margins better or worse than Intel 7 in 2026"
"are panther lake product margins higher or lower than Arrow lake and Lunar Lake"

Well for the first part of question it will be obvious answer of Intel 7 product being a better margin overall at Intel level cause the IP and both the process have all the juice extracted from them
For 2nd part I would like to know as well.

I have my Intel unit and wafer cost spreadsheet ready to be updated after the earnings come out..... cant wait!

I am waiting for earnings as well.

 

[Brady]

The funny thing is, even if Intel knows that trying to do 'dibs and dabs' together is what made them stumble before, they still have to do the exact same thing to regain process leadership. Now they're integrating Nanosheets and BSPDN simultaneously, while trying to bring up 2.5D packaging along with them. They have to learn them fast because now High NA(and stitching) is wating for them...

My impression is that they’re actually technically pulling it off with all these at once (Panther Lake and Clearwater Forest is proof), but “at what cost” is the question? e.g. yield, wafer cost, packaging cost

 

[hist78]

the real question will be how much of the 18A financial issue will go to product group and how much will go to IFS. OR will intel change the accounting yet again???? looking forward to see what happens.

If my observation of Intel’s past behavior is accurate, I expect the company to attribute as many costs and expenses as possible to IFS. That approach makes the narrative easier to explain to analysts and the general public. Everyone already knows that IFS is losing money and facing significant challenges, so adding a few more items to IFS' list of problems doesn’t meaningfully change the overall picture.

 

[hist78]

If you can ask questions of Intel at earnings:
" are 18A PRODUCT margins better or worse than Intel 7 in 2026"
"are panther lake product margins higher or lower than Arrow lake and Lunar Lake"

In June 2024, Intel’s then CEO, Pat Gelsinger, said during an interview in Taiwan:

"Q: Intel's foundry business has suffered huge losses over the years. What’s the future for the company’s foundry business?

A: Today, our wafers are largely driven by Intel 10 and Intel 7. Those technologies are not that competitive in terms of processes nor costs. But as we get to the five nodes and four years to Intel 3 and Intel 18A, these are very competitive and much more cost-effective. So, as those technologies ramp up, we will go to break even and then be in a very comfortable profitable business later in this decade. We feel very comfortable with the profitability strategy plan that we've laid out."

Source: https://english.cw.com.tw/article/article.action?id=3709

Now that we are in 2026 and Intel has Intel 3 and 18A in volume production, I wonder if Pat Gelsinger’s prediction has come true.

Another unusual situation is that, despite the high cost of Intel 7, Intel may prefer it in some cases because it may generate more profit for Intel than Intel 3/4 or Intel 18A. The reason is that the Arizona fabs (responsible for 18A manufacturing) and the Ireland fabs (responsible for Intel 3 and Intel 4 manufacturing) are 49% owned by Brookfield (Arizona fabs) and Apollo (Ireland fabs). Intel has guaranteed certain investment returns (not publicly disclosed) to Brookfield and Apollo. As a result, Intel 7 may ultimately deliver higher profit to Intel because there is no outside party with whom profits must be shared.

 

[Artificer60]

If my observation of Intel’s past behavior is accurate, I expect the company to attribute as many costs and expenses as possible to IFS. That approach makes the narrative easier to explain to analysts and the general public. Everyone already knows that IFS is losing money and facing significant challenges, so adding a few more items to IFS' list of problems doesn’t meaningfully change the overall picture.

They may, indeed, make that choice. However, I would caution against setting expectations based on Intel's past behavior. Intel has had two CEO's that weren't steeped in the Intel culture.

The first was Bob Swan who, if rumors are to be believed, was looking to divest Intel of its fabs. Heresy to the old guard who bleed blue. He was also making significant changes to the culture that were designed to move Intel away from the twisted version of the Grovian culture that had grown up over time. Just my opinion, but I think Andy Grove was rolling over in his grave when he saw what the culture he had fostered had become. A culture where data was ignored and people were afraid to question decisions wasn't what Grove advocated for.

The second is Lip-Bu Tan. Like Swan, he is an outsider and less interested in doing things the Intel way. From all I hear, he is doing far more than giving lip service to the idea of listening to the customer. It will be a long journey (I've seen statements that he committed to 10 years in the CEO role), but sees a very different company at the end of the road than his predecessor. Gelsinger seemed to want to raise the old Intel from the ashes, but Lip-Bu Tan seems like he wants to fundamentally change the company. I believe his approach is far more pragmatic and the changes he is driving in Intel culture will result in a very different company from the Intel of the past.

 

[semiman]

My impression is that they’re actually technically pulling it off with all these at once (Panther Lake and Clearwater Forest is proof), but “at what cost” is the question? e.g. yield, wafer cost, packaging cost

Yes exactly. It's not like it's mission impossible or anything. If they manage to pull everything(or most of things) at once, and if it helps their products team to regain "product" leadership(compared to mixing TSMC silicons) before they burn all cash available.

I think Intel is sending mixed signals now. Panther lake showed superior PPA in their presentations, very first debut of BSPDN but where are 18A HX products and desktop variants? Maybe they simply lacked capital to ramping up due to COVID? or Do they bleed capital if they use more Intel silicons than TSMC's?

One interesting opponents here is Samsung. Samsung Foundry acquired non-conventional(?) leading edge customers. Samsung Memory(HBM base die, high volume and easy to pattern) and Tesla(Non-mobile leading edge launch customer). It'll be really interesting to see how this market unfolds for 2~3 years.

 

[dkr1986]

Also the 18A volume is still low .... it is not any where near 40K wafer per month at Fab 52. and then we have the whole "Intel is only 51% owner of Fab 52"... I have no idea how they will report this "non controlling income"

Exactly as you say I suppose. It will be under "Income attributable to non-controlling interest" and will be deducted before the "Net income attributable Intel investor" is calculated. Net income, Earnings per Share and Operating & Free Cash Flow metrics will be lower. Also I would not assume that since Brookfield owns 49% of the Fab & tools, they would be entitled to 49% of the profit. It will be much lower in my opinion as Intel operates the fab & owns the IP being produced in the fab. It is like leasing a co-owned property. I don't know if that is 20% or 30% but definitely cannot be 49% (if it is then the CFO is a fool).

the real question will be how much of the 18A financial issue will go to product group and how much will go to IFS. OR will intel change the accounting yet again???? looking forward to see what happens.

We need to keep in mind Intel Product is already paying high wafer prices to TSMC (including 50% TSMC gross margin). So high wafer price for a leading-edge node is already accounted for in recent times for Intel Products team. It is a matter of ramping down Arrow lake+ Lunar Lake volume and increasing Panther Lake volume to offset. Considering 18A based products are much better than Arrow Lake & lunar lake (yes, all things indicate PTL is a better product than the ARL & LNL so far), ASP increase for 18A based products can be incremental. So, I think as 2026 progresses, Intel Product margins should stay flat. The threshold is Intel 18A wafers starting to eat into Intel 7 wafers volume (as opposed to N3), then product margin will start taking some hit. is the 18A based product ASP increase enough to offset 18A wafer ASP increase is a question, I don't have answer for yet. I have not accounted to demand destruction due to high memory prices.

On the foundry side, the 18A ramp cost would eat into margins but 18A wafer price would help margins. If yield is decent, the more they use foundry, the more the economics of foundry look good. If the cost of 18A wafer is higher than 18A wafer price due to low yield, then nothing changes and foundry bleeds money. But even these changes over time as yield improves. I only see margin improvement for Intel Foundry from here after some startup costs during initial ramp. The bigger problem is idle older node fabs that not producing anything or being underutilized.

 

[hist78]

We need to keep in mind Intel Product is already paying high wafer prices to TSMC (including 50% TSMC gross margin). So high wafer price for a leading-edge node is already accounted for in recent times for Intel Products team

Are we really sure that Intel Foundry’s wafer costs are lower than TSMC’s? When we consider all the direct and indirect cost tied to Intel’s manufacturing, Intel’s actual costs might be higher than TSMC’s.

I’ve heard both current CEO Li‑Bu Tan and former Intel CEO Pat Gelsinger acknowledge Intel’s high cost structure. But I haven’t heard either of them suggest that TSMC’s pricing is the source of the problem.

John Pitzer, Intel’s Vice President of Corporate Planning and Investor Relations, stated during an investor conference with Morgan Stanley in March 2025 that Intel intends to keep outsourcing up to 30% of its wafer needs. He explained that continuing to use TSMC creates “healthy competition” between TSMC and Intel Foundry, describing TSMC as a “great supplier.” I don’t think Intel would deliberately rely on a higher cost TSMC wafer just to keep Intel Foundry in check; that doesn’t seem logical.

 

[siliconbruh999]

Are we really sure that Intel Foundry’s wafer costs are lower than TSMC’s? When we consider all the direct and indirect cost tied to Intel’s manufacturing, Intel’s actual costs might be higher than TSMC’s.

I’ve heard both current CEO Li‑Bu Tan and former Intel CEO Pat Gelsinger acknowledge Intel’s high cost structure. But I haven’t heard either of them suggest that TSMC’s pricing is the source of the problem.

John Pitzer, Intel’s Vice President of Corporate Planning and Investor Relations, stated during an investor conference with Morgan Stanley in March 2025 that Intel intends to keep outsourcing up to 30% of its wafer needs. He explained that continuing to use TSMC creates “healthy competition” between TSMC and Intel Foundry, describing TSMC as a “great supplier.” I don’t think Intel would deliberately rely on a higher cost TSMC wafer just to keep Intel Foundry in check; that doesn’t seem logical.

They also said they are unsure about where this would stand they are currently at 30% would they keep at 30% or it would be more like 20% we don't know

 

[lispwright]

I don’t think Intel would deliberately rely on a higher cost TSMC wafer just to keep Intel Foundry in check; that doesn’t seem logical.

After the 10 nm debacle and particularly how it was handled internally it strikes me as entirely logical.

I believe CFO prior to the purge of BK Bob Swan realized Intel Foundry was broken, but I can see him realizing he didn't have the knowledge or stature to fix the broken culture of a technical company. Neither he nor PG fired the numbers of managers that as another crude measure would be a necessary step in fixing the culture.

Lip-Bu Tan is reported here on the forum to be trying to do that, and also was able to say severe layoffs including cutting management layers was financially necessary. But he was hired the same month as the statement by John Pitzer.

It's also insurance against another node transition failure. Plus there's the tools issue, having the product side widely use standard ones for TSMC could help Intel Foundry's struggle to create PDKs whales are wiling to use.

 

[siliconbruh999]

After the 10 nm debacle and particularly how it was handled internally it strikes me as entirely logical.

I believe CFO prior to the purge of BK Bob Swan realized Intel Foundry was broken, but I can see him realizing he didn't have the knowledge or stature to fix the broken culture of a technical company. Neither he nor PG fired the numbers of managers that as another crude measure would be a necessary step in fixing the culture.

Bob swan made the decision what was right according to him he can't fix Intel foundry so he decided to shut it down.

Lip-Bu Tan is reported here on the forum to be trying to do that, and also was able to say severe layoffs including cutting management layers was financially necessary. But he was hired the same month as the statement by John Pitzer.

It's also insurance against another node transition failure. Plus there's the tools issue, having the product side widely use standard ones for TSMC could help Intel Foundry's struggle to create PDKs whales are wiling to use.

Well i am pretty sure Intel would have learnt the lesson after IFS 1.0

 

[hist78]

After the 10 nm debacle and particularly how it was handled internally it strikes me as entirely logical.

I believe CFO prior to the purge of BK Bob Swan realized Intel Foundry was broken, but I can see him realizing he didn't have the knowledge or stature to fix the broken culture of a technical company. Neither he nor PG fired the numbers of managers that as another crude measure would be a necessary step in fixing the culture.

Lip-Bu Tan is reported here on the forum to be trying to do that, and also was able to say severe layoffs including cutting management layers was financially necessary. But he was hired the same month as the statement by John Pitzer.

It's also insurance against another node transition failure. Plus there's the tools issue, having the product side widely use standard ones for TSMC could help Intel Foundry's struggle to create PDKs whales are wiling to use.

So do you think Intel wafer cost is higher or lower than TSMC's, or at least be more competitive than TSMC?

 

[hist78]

Bob swan made the decision what was right according to him he can't fix Intel foundry so he decided to shut it down.

Well i am pretty sure Intel would have learnt the lesson after IFS 1.0

After the 10 nm debacle and particularly how it was handled internally it strikes me as entirely logical.

I believe CFO prior to the purge of BK Bob Swan realized Intel Foundry was broken, but I can see him realizing he didn't have the knowledge or stature to fix the broken culture of a technical company. Neither he nor PG fired the numbers of managers that as another crude measure would be a necessary step in fixing the culture.

Lip-Bu Tan is reported here on the forum to be trying to do that, and also was able to say severe layoffs including cutting management layers was financially necessary. But he was hired the same month as the statement by John Pitzer.

It's also insurance against another node transition failure. Plus there's the tools issue, having the product side widely use standard ones for TSMC could help Intel Foundry's struggle to create PDKs whales are wiling to use.

Are we placing too much expectation on Li‑Bu Tan and hoping he is a “superman” who can turn Intel into the only advanced‑logic IDiM in the world, and in history, capable of beating the foundry/fabless business model?

Or do we believe Intel’s problems have nothing to do with the IDM business model? Instead, is the real issue that over the past 20 years Intel consistently hired poor engineers, ineffective managers, weak boards of directors, and bad CEOs who made bad decisions?

 

[dkr1986]

Are we really sure that Intel Foundry’s wafer costs are lower than TSMC’s? When we consider all the direct and indirect cost tied to Intel’s manufacturing, Intel’s actual costs might be higher than TSMC’s.

I’ve heard both current CEO Li‑Bu Tan and former Intel CEO Pat Gelsinger acknowledge Intel’s high cost structure. But I haven’t heard either of them suggest that TSMC’s pricing is the source of the problem.

I am just saying that If Intel Foundry charges market price for 18A wafers to Intel Products, there is not going to be any immediate impact on Intel Product's margins because they have been paying that kind of wafer pricing for N3 for the last couple of years.
Intel's wafer cost being higher or lower than TSMC is not a factor here for Intel Products team. Products team will only pay a fair price for a wafer. That is the whole point of reporting Intel Foundry & Intel Products team financials separately. At least that is how it should be.

For eg. Say N3B wafer pricing was 20k per wafer (made up number). Considering 18A is N3P equivalent or better, Intel Foundry can charges 22k for an 18A wafer. Intel products team incurs 2K (10%) over what they were paying TSMC but they can increase sale price of these new CPUs to recover the lost margin or may even price it more to expand margin as these are better products. So there is not going to be huge impact on Intel Products margin as long they ramp down N3B based products & replace it with 18A products in 2026. It is only when 18A based products volume increase so much, that it eats into their Intel 7 based products volume, they (Products team) will feel a significant impact on margins. But it is not going to evident in 2026 is my point as Intel product team has been paying similar wafer pricing for N3B.

Now, considering TSMC has >50% gross margins, this means it only costs about 10K to make a N3B wafer. What portion of that gross profit goes to Intel Foundry now due to using 18A is function of the cost of Intel vs TSMC. This affects the margins for the Intel Foundry team.

So far what we know per Intel management is Intel 7 wafer cost is at similar ball bark to Intel 18A wafer cost (due to DUV multipatterning vs EUV + PowerVia), Intel 7 wafers are sold to Intel products team at cost (meaning 0% gross margin) and when they move to Intel 18A increase in wafer ASP is 3x the increase in cost.

 

[tomatoma]

High-NA EUV is garbage

 

[tomatoma]

Intel always fails

 

[tomatoma]

As I, very much not a fab guy, understand it, Intel's 10 nm was very roughly equivalent to TSMC's N7 in what it targeted, but a bit more aggressive.

I strongly suspect Intel's legendary high level bad management not only seeped down to its crown jewel of fab technology, but was responsible for those multiple failed rescues. Did the company even internally acknowledge it was a failure until the single SKU Cannon Lake launch, with i's biggest piece of silicon, the iGPU fused off, with essentially no laptops using it for sale? BK was purged from Intel three months later.

So neither company planned on using EUV for their first iteration of this general node, but TSMC's success allowed it to ease into EUV use with N7+ for "up to four of its critical layers." This was not much of an option for Intel until the base node worked well enough (and even then, there was a terrible mistake made with a clock tree circuit for Intel 7+ higher end Raptor Lake SKUs).

I personally like the explanation about Raptor Lake in the second half, as I understand it to be a failure phenomenon unrelated to the process node, but I've never heard of an IGPU failure related to Cannon Lake in the first half, so it's a bit of a negative.
In fact, there are various models of Raptor Lake that use Intel 7+, but there are many models that do not have any problems.

 

[tomatoma]

I think you are confounding two separate issues here. The negative effect of Copy Exact have to do with what happens after the transfer. The 10nm face plant occurred in Oregon before it was transferred so Copy Exact had nothing to do with it.

I'm told TSMC also practices a Copy Exact methodology. When the process is transferred to a new fab, they have to match the performance of the transferring fab without process modification to qualify the fab. The real difference happens once the new fab is qualified. Once the qualification process is completed those fabs are free to work on process improvement.

Copy exact really became toxic for Intel during the 10nm days. Prior to this the development team disengaged from a process after transfer as soon as the new technology began ramping. That left the factories running the older technology free to make improvements (at least in a limited fashion). However, since Intel was stuck in 14nm for so long, the development team didn't disengage and the other fabs were prevented from making any process improvements.

That still put Intel's non-development fabs at a disadvantage because Intel cycled through process nodes so quickly, but the idea that Copy Exactly prevented improvement after the development team left the process is simply not true. The other place where Copy Exact as practiced by Intel failed was in fostering a "not invented here" mentality among the development team and things that were learned in the HVM fabs running the previous process were not fed forward to the development team. This resulted in less than optimal processes being transferred.

Yes, almost all companies develop their products in a similar way to Copy Exact.
It's not that Copy Exact is inferior.

 

[tomatoma]

This is the fundamental problem. Times change. people need to adapt. Intel did not (well they did in 2020 but Pat reversed the decision).

When I started (late 80s). Greate compute companies had to have fabs. IBM, DEC, AMD, Intel and others. Their technologies helped differentiate. the margins paid for the RnD. IBM and DEC were ahead of Intel.

IF you have scale, and a technical lead, and high margins, and no serious competitors. It works. They all learned their lesson and TSMC delivered and the rest is history. Some smart man wrote a book "Fabless"

Today, all great compute companies outsource manufacturing. Advanced technology, tons of flexibility.

We will soon learn: "sometimes you are so busy seeing if you CAN do a new technology, you didnt stop to say whether you SHOULD do a new technology". Intel CAN do 18A and 14A.


Side note: I love the numbers in the Charts. Intel outspent ALL other companies in RnD in 2020 (TSMC, AMD, Nvidia combined). Intel didnt spend too little... , thanks @hist78

So Intel has to go bankrupt before it can be separated.
Before IDM is outdated, Intel itself is outdated.
18a/14a is a waste. It's like garbage.

 

[tomatoma]

Pat did agree to reverse a possible design only course Swan opened and so tried to rescue the company as an IDM+Foundry. Besides financial forecasting missing the shift to AI datacenter spending, his biggest failure was not landing any 18A whales, right? Something he bet the company on.

For GPGPUs, I start with the paragon of Nvidia, which under one man's leadership decided on an adequate at minimum total ecosystem strategy with the start of the CUDA software project in 2004. And they stuck to that strategy and executed well enough for two decades and eventually reaped wild rewards with the AI boom.

Intel did try its hand in the then hot discrete graphics chip market with an i740 project that launched in early 1998 and quickly failed, but was followed by other projects based on it if I read Wikipedia's history article correctly. They count 12 generations through Intel Xe, but it looks like after gen 4 they went from embedding in the Northbridge, which probably counts as an iGPU, to iGPUs in the CPU.

I'm not directly familiar with this because my requirements have been well satisfied by an OK 2D (i)GPU for decades. But I can't forget the Larrabee, started in one form in 2005 per an IEEE article, was canceled before launch after much hype, Gelsinger played a major role in it, and many think he got fired for that in the year before. For GPGPU type usage without any GPU features Xeon Phi followed, and shipped product but failed and was discontinued in 2020 per Wikipedia.

So that's two strategies but without Intel trying to move to GPGPU type market until Larrabee? Then they launched a "real" GPGPU project in 2017, but it was led by a guy who had been let go from AMD. One of BK's last major hires I assume, but then again who good would sign on to work on this project at Intel? It was a rolling disaster until pretty recently.

Enough so Lip-Bu Tan has admitted Intel has lost its chance at the current AI training market, and will now try for the inference market.

No, Gelsinger himself didn't cancel the Larrabee project.
He was in a position to lead the project, and he was in a position to be disappointed when the project was cancelled.