You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
1. "And 18A, our biggest customer for the next two, three years is still Intel products,"
2. "Now it is about going through the remaining yield challenges, defect density challenges, continuing to improve it, improving process margin and getting it ramped. Will there be challenges? There will be, but I think we are progressing."
Both of these are obvious unless you are a CEO trying to paint a pretty picture.
1) Intel product group will be the largest foundry user for a long time. Probably 4+ years. No one will put their main high volume product on a untested foundry. Intel will be backup and DZ has said that before. Only Pat thought people would jump to Intel
2) you will find 100s of new issues and new testing methods over time and have to solve them one by one. Intel is very capable of this but it is a completely new transistor design AND metallization design. It will take a lot of work to get yields to where they need to be. quoting a D0 number a few months ago with no context doesnt mean much. Getting Panther lake to work and yield is the challenge
Two great examples of Pat's issues with investors vs Naga and Zinsner who are realistic. Pat would have said "we have a whale and will be the largest foundry soon... 2026 will be amazing" and "Panther lake is already yielding and working perfectly, we are ramping it now"
Like I said before. I really like DZ, MJ, Ann, Naga. Being realistic is what is needed right now.... not Cheerleading.
And what are his qualifications to make such a statement? Because he is a successful stock picker? Did he mention the "M" word? Does anyone here see a problem with TSMC being the only leading edge foundry in the world? How about having a US based leading edge semiconductor manufacturer? Is that important to the security of our country?
I realize that dumping manufacturing would help the Intel stock price but I think it would be short term. Intel is not innovating on the design side enough to regain their glory. The competition is too fierce. There are hundreds of companies chomping at the AI pie and the shake-out is coming, absolutely. Having design and leading edge manufacturing is a big differentiator and Intel is the only one to have that today.
Take a look at the Intel IEDM papers. The process innovation continues....
Do you trust a fading power like Intel to provide US based semiconductor manufacturing? Haven’t many/most of the good people there who created the famous technologies in the past gotten let go in the recent cycle of cost cutting and layoffs?
I have heard it called "capital efficiency and cost discipline" which does not accurately describe Intel's standard operating procedure.
Building semiconductor fabs is expensive so you must make sure they are fully utilized. The whole reason the fabless business started was because the IDM fabs were not full so they leased out capacity to small companies like Chips and Technology (the first fabless company which is referenced in my first book and Morris Chang's second book).
The same can be said today, you must fill those fabs. To solve that problem TSMC builds fabs based on customer demand. Some big TSMC customers even pay in advance to build those fabs and guarantee capacity.
IDM foundries (Intel and Samsung) build fabs in hopes customers will come and customers do not come if there is risk of not getting the wafers within the timeframe required. Today the risk is too high for Intel and Samsung foundry for the big fabless customers.
The other risk is TSMC being a monopoly and TSMC being a political pawn in the China vs US chess game.
In business all risks must be balanced out for a more predictable outcome thus the need for second sourcing. My personal mantra is hope for the best but plan for the worst and you will not be disappointed.
Given all of that, Intel Foundry is imperative and Intel Foundry needs Intel design for process development and help filling the fabs. If you split Intel the investors may be happy but it will be short lived because they will both fail to meet investor and industry expectations, just my opinion as a 40 year semiconductor professional and investor. I am not a professional stock picker or armchair analyst. I am one of the many semiconductor professionals that do the work.
Talking about the financial discipline, has Intel hired enough finance MBA, used them appropriately, and listened to them appropriately in recent years? Intel's spending spree and hiring spree should have never happened if those decisions are based on the financial reality.
But TSMC and its people are best in the manufacturing - the foundry business. Those executives who are good in products are most likely working at the fabless companies such as Qualcomm, Nvidia, AMD, MediaTek, and Apple.
If Intel insists to find a new CEO who is good in both products and foundry business, it probably limits the candidate pool to people from Samsung and Intel itself.
If Intel insists to find a new CEO who is good in both products and foundry business, it probably limits the candidate pool to people from Samsung and Intel itself.
I have known a lot of people in the industry from decades of working in it, and I've never met anyone, including from Intel and Samsung, who are suitable to lead and be the final decision-maker in product engineering and chip manufacturing simultaneously. Not one, not ever.
I have known a lot of people in the industry from decades of working in it, and I've never met anyone, including from Intel and Samsung, who are suitable to lead and be the final decision-maker in product engineering and chip manufacturing simultaneously. Not one, not ever.
There are hundreds of competent chip design groups in hundreds of companies on the planet. Many produce very impressive products.
There are on the order of around ten major chip manufacturing companies in the world. Three make mostly memory chips. One dominates in advanced logic chips.
A great chip design team for a complex chip probably costs in the range of $100 million. A new EUV fab costs $25-30B.
I think a reasonable analogy is designing and manufacturing cars versus airliners.
There are hundreds of competent chip design groups in hundreds of companies on the planet. Many produce very impressive products.
There are on the order of around ten major chip manufacturing companies in the world. Three make mostly memory chips. One dominates in advanced logic chips.
A great chip design team for a complex chip probably costs in the range of $100 million. A new EUV fab costs $25-30B.
I think a reasonable analogy is designing and manufacturing cars versus airliners.
Ann Kelleher spoke at an industry event last night. She is definitely still with Intel. I had heard her issue was with Pat G and now that is no longer an issue. I saw Ann earlier this year, tonight she looked different so maybe it is a health issue?
Maybe Intel can have Co CEOs, one for product and one for foundry? It really is a totally different job.
Hock Tan won the Dr. Morris Chang Award of excellence tonight. He did a great (short) speech. Hock should buy Intel. He would right that ship, absolutely. The place was full of semiconductor CEOs and Hock is the only one I think of that can fix Intel.
Pat Gelsinger turned VMware around, during his time there only for Broadcom to buy it later and f'ing it up!!! May be Intel (already messed up) will face similar turbulence.
Cloud provider moved most of its 20,000 VMs off VMware.
arstechnica.com
"Companies have been discussing migrating off of VMware since Broadcom’s takeover a year ago led to higher costs and other controversial changes. Now we have an inside look at one of the larger customers that recently made the move."
Pat Gelsinger turned VMware around, during his time there only for Broadcom to buy it later and f'ing it up!!! May be it Intel (already messed up) will face similar turbulence.
Cloud provider moved most of its 20,000 VMs off VMware.
arstechnica.com
"Companies have been discussing migrating off of VMware since Broadcom’s takeover a year ago led to higher costs and other controversial changes. Now we have an inside look at one of the larger customers that recently made the move."
I agree. PatG's decision at VMware to pursue to a cloud strategy was, IMO, the best strategic direction he ever chose in his career. Broadcom has put VMware into their famous customer harvest mode. After watching them do this to other companies they've acquired, I can't say I'm a Hock Tan fan.
IMHO, there is no (major) design issue with ARL-S. It is their first run at tile architecture in Desktop (same as Meteor Lake in mobile which had similar issues during launch). I agree there are some teething issues like MTL but that will be ironed out with microcode updates. But it has achieved exactly what Intel on desktop lacked good Perf/watt. No more high power draw and run like a room heater as RPL-R did. It draws same power as Zen 5 non X3D chips and runs cooler. ARL-S is competitive with Zen 5 non X3D in perf/watt in gaming and productivity workload (well mostly, there are some wins and some losses too).
The criticism is it has slightly regressed from 14th gen in gaming and is no match for AMD's X3D SKUs. AMD's own 9950X cannot match a 9800X3D in gaming!!! Intel does not have an answer to X3D chips (at least yet- recent leaks indicate Nova lake in 2026 may have a SKU to fight X3D, we will see). Intel has prioritized perf/watt over raw performance and solved the high power draw and overheating issue (the issue everyone was complaining about).
This bodes well for OEM desktops which will usually have the non K SKUs.
Some stuff I have seen online seems to indicate that ARL platform will be good in mobile format (at low power). That is where the money is. We will see that next year if that is the case.
Also some criticisms, I don't understand at all. ARL-S K SKUs are launched for PC enthusiasts who should be able to tinker around the overclocking features to improve performance. Based on the reviews I saw, it is an overclocker's CPU with some performance left on table.
The RPL-S/R issue with degradation is concerning but I don't think we can use that to conclude Intel is underperforming. Also this issue was exacerbated with MB vendors not following Intel Power guidelines (which Intel happily turned a blind eye to compete with AMD's CPUs on better TSM nodes).
The last part is similar to how AMD's X3D chips were blowing up with ASUS motherboards.
Yes, this is the same thing AVGO has done with every company they have acquired. All acquisitions must hit the target margins. If not, prices rise and/or heads roll until margins are met. If still not met, the group is sold or otherwise discarded. Investors love this scheme, other stakeholders not nearly so much.
For each acquisition this juice squeezing works for about 5 years at the max. So AVGO is always lining up the next trophy prey to pull into the processing plant. About a 2 year cadence iirc.
If AVGO bags Intel (I'm not convinced after AVGO pivoted to SW firms) there would be large asset sales and the remaining groups would be squeezed dry. The result would not look like today's Intel.
Maybe that is what Dan expects/wants if AVGO bags Intel?
IMHO, there is no (major) design issue with ARL-S. It is their first run at tile architecture in Desktop (same as Meteor Lake in mobile which had similar issues during launch). I agree there are some teething issues like MTL but that will be ironed out with microcode updates. But it has achieved exactly what Intel on desktop lacked good Perf/watt. No more high power draw and run like a room heater as RPL-R did. It draws same power as Zen 5 non X3D chips and runs cooler. ARL-S is competitive with Zen 5 non X3D in perf/watt in gaming and productivity workload (well mostly, there are some wins and some losses too).
Yeah i don't have issue with the power draw but the regression in core architecture you can read this blog it regressed vs Meteor lake in core capabilities
The criticism is it has slightly regressed from 14th gen in gaming and is no match for AMD's X3D SKUs. AMD's own 9950X cannot match a 9800X3D in gaming!!! Intel does not have an answer to X3D chips (at least yet- recent leaks indicate Nova lake in 2026 may have a SKU to fight X3D, we will see). Intel has prioritized perf/watt over raw performance and solved the high power draw and overheating issue (the issue everyone was complaining about).
This bodes well for OEM desktops which will usually have the non K SKUs.
Some stuff I have seen online seems to indicate that ARL platform will be good in mobile format (at low power). That is where the money is. We will see that next year if that is the case.
Also some criticisms, I don't understand at all. ARL-S K SKUs are launched for PC enthusiasts who should be able to tinker around the overclocking features to improve performance. Based on the reviews I saw, it is an overclocker's CPU with some performance left on table.
The RPL-S/R issue with degradation is concerning but I don't think we can use that to conclude Intel is underperforming. Also this issue was exacerbated with MB vendors not following Intel Power guidelines (which Intel happily turned a blind eye to compete with AMD's CPUs on better TSM nodes).
The last part is similar to how AMD's X3D chips were blowing up with ASUS motherboards.
I just re-read the transcript from Naga and Dave. Again, I need to re-iterate the "realistic" or "open" comments vs Pat. This is a huge improvement.
Naga; "And next year, as I look at it, primarily the first half will be getting the node into engineering samples into our customers' hands and getting the feedback and starting a ramp in Oregon. And the second half of 2025, our milestone is certifying the node, getting it ramped in Arizona and getting the product on the shelves so that customers can buy it. "
This is what is actually expected internally and now its being openly communicated
Pat's hinting that the process and product being ready to go is not helpful to the Intel cause or Pat's career. His quote Dec 4th would have been "18A is Manufacturing ready today and Panther lake is being shipped to customers, 18A is basically mission accomplished"
For each acquisition this juice squeezing works for about 5 years at the max. So AVGO is always lining up the next trophy prey to pull into the processing plant. About a 2 year cadence iirc.
I don't think you are familiar with Broadcom today. For example, ASIC Generative AI revenue is driving growth by Broadcom. The LSI Logic acquisition (closed 11 years ago) is a big piece of driving this growth.
I don't think you are familiar with Broadcom today. For example, ASIC Generative AI revenue is driving growth by Broadcom. The LSI Logic acquisition (closed 11 years ago) is a big piece of driving this growth.
I don't doubt there are groups at AVGO that have adjusted to doing things the AVGO way. One can read of folks there working for the RSUs, or hear similar if they have personal contacts there.
But that does not mean AVGO has moved away from the margin driven playbook they have had so much success with.
Two great examples of Pat's issues with investors vs Naga and Zinsner who are realistic. Pat would have said "we have a whale and will be the largest foundry soon... 2026 will be amazing" and "Panther lake is already yielding and working perfectly, we are ramping it now"
Hmm, I do think Pat had a lot of hubris, but two things he/Intel did stuck out to me as him being a little understated at times:
- During interviews Pat said that he thinks Intel Foundry is already ahead of Samsung if you count Intel's internal business, but that they weren't including that in their projections to become #2 Foundry. (I would have to dig this up but I believe it was during an Ian Cuttress interview with Pat).
- We saw charts published by Intel that admitted they were behind TSMC on certain characteristics with certain nodes. That's pretty humble for a Grove-ite.
