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!
Agree to disagree. TSMC's best customers are exclusive to TSMC and there are quite a few. Intel also competes with TSMC on the foundry side. Intel is certainly a big TSMC customer but I would not say best. Lip-Bu may change that, we shall see, but I can assure you CC Wei did not like Pat Gelsinger so some relationship repair must be done.
I still have not heard of an N2 wafer agreement with Intel but I will know more next week. In my opinion N3 will suffice for the supporting chiplets Intel will surround 18A die with. I am really hearing good things about 18A so in my opinion Intel will start moving away from TSMC in the very near future. Intel based N3 revenue however will have a long life, absolutely. TSMC N3 really is a record breaking node, absolutely.
As of rn Yes due to the N3B/E contracts but over time it's due to reduced I don't think anyone would like to keep wafers external while having internal wafer shop and loosing money on Internal wafer shop.
Intel is one of TSMCs top 6 customers in 2025 I believe
Intel external wafer sales are <1B per year in 2025. not a top 10 foundry in 2025
Intel 18A works very well from what I hear. The problem I am guessing is that the MOR output/tool is lower than expected, yields are not there yet (we are still 6 months from Product qual so not major) and as a result the 2026 product costs are higher than originally planned. So it might not be profitable in 2026.
If this were to be true (who knows). then the response would be to not ramp like planned. There are lots of Rock/hard place trade offs due to financing so we will see how it comes out in 2026. This will show how Tan's mind is thinking
Simple metric: in 2026 How many wafers are running on 18A, How many are being sold to Intel product group, How many to External foundry customer. Then ask IFS leader "what is the Operating Margin of the 18A product line" .... the answer will begin with "well, its complex and we have some unexpected ....."
I'm surprised. AMD was always slower moving to new TSMC nodes. Launching N2 products in 2026 alongside first Apple products is a big shift. And a welcome one, because I feel like their old strategy of not using TSMC's best was a disadvantaged strategy.
and you will not see any chiplets. This is a full N2 wafer. Chiplets use 2.5/3D integration techniques since chiplets are individual small chips that are used with interposers. Depending on how this N2 wafer was designed, the chiplets could be attached. But this N2 wafers has many individual (non chiplet) die that need to be scribed/broken from the whole wafer.
I mean there weren't exactly any bad signs in my book. This is a better sign for AMD then TSMC in my book since TSMC was always getting the buisness. IMO the question was always when AMD N2 products would launch not if.
Absolutely not. Unless TSMC has invented time travel and took the wafers rolling off the line at EOQ1'26 back to last month launching an N2 iPhone would literally be impossible.
I really think that Intel's best hope on the design side is to pull another Netburst --> Core transition out of their hat. Your point about being tied to what is already in the pipeline is a good one, but if Lip-bu Tan is to be believed then he is looking for ways to disrupt the design cycle. Core has been a good architecture, but it is getting a bit long in the tooth and it is time to move on.
Even world class design teams take 4+ years to design an SOC, and the IP situation has even longer time horizons. It is also by definition impossible to change how you design anything after said design work has done. Even if Tan overnight made Intel products the best design house in the world (which he definitely didn't do) there is nothing that can be done about IPs product definitions and other work that has already been done. The only way to "get out of it" is to throw it all away and start from scratch which still doesn't "fix" the issue of not having products from Tan's Intel launching for 5 years. Now obviously Novalake next next is going to have a lot of influence from the work Tan did since almost all of the development would have happened under him and any reforms he started would have been implemented for some parts of the product development cycle. But other things like how IP is done, what IPs to use, what processes to support, how to define the product, what do our customers want? That will have either happened under Pat or under systems that were created back when he was CEO. But something like Pantherlake or Novalake are 100% Pat's babies. Something like Novalake next would be like 60/40 Pat/Tan if Tan completely reformed Intel over night (but realistically we are likely talking 90/10 because even at great companies reforms take lots of time and effort). And I think realistically Novalake next next would be like 50/50 with planning/foundational work being all done under the Pat regime and execution being all done under Tan (assuming Tan completes at least substantial reforms of the design side no later than his 2 year mark). And before you ask even if he does it in one I don't think that greatly influences his impact on the prior projects because so much of the work is already done or at least will be in a year.
I also have to believe that listening to your customers and designing what they need would go a long way towards addressing some of Intel's gaps as well. This is another area Lip-bu Tan promises to address.
But I'm not a design guy, so maybe I'm just being completely delusional here.
While not wrong, talk is cheap and actions speak louder than words. Pat loved talking about "going back to delighting our customers" and "improving our say do ratio" but just because he said that was the goal didn't mean it happened. While much of what Pat gets flack about are inherited problems from the prior yahoos, he definitely had some failings on the design side. Sorting out the inherited problems and being overoptimistic on three separate counts thus getting in over his skis because he thought Intel products was hotter than it was, Pat didn't seem to be doing enough fast enough to stabilize Intel products or to reform them into a world class fabless organization. I think there were some decent ideas he pushed towards that end, but the reforms just weren't happening fast enough. Saying "this is no good, if we want to be better we should do this" doesn't mean anything if the changes take a year or more to implement. This is why I don't totally get the "Tan is already so much better than Pat" stuff because they have literally the exact same talking points. I need to see results and timely rather than plodding reforms before I say I believe in Intel products turning around like foundry is. Just a shame that a clear assessment of how good of a job Tan did won't be possible for many years. Hopefully the board doesn't try another 5 year turnaround plan in 3.5 years. For that is the garuenteed path for complete anhiliation for all parts of Intel.
I think Lip-Bu is quite efficient. He replaced (voluntary or involuntary) two executives within 1 month mark, Sandra L. Rivera and Christy Pambianchi. Also the head of TD.
Even world class design teams take 4+ years to design an SOC, and the IP situation has even longer time horizons. It is also by definition impossible to change how you design anything after said design work has done. Even if Tan overnight made Intel products the best design house in the world (which he definitely didn't do) there is nothing that can be done about IPs product definitions and other work that has already been done. The only way to "get out of it" is to throw it all away and start from scratch which still doesn't "fix" the issue of not having products from Tan's Intel launching for 5 years. Now obviously Novalake next next is going to have a lot of influence from the work Tan did since almost all of the development would have happened under him and any reforms he started would have been implemented for some parts of the product development cycle. But other things like how IP is done, what IPs to use, what processes to support, how to define the product, what do our customers want? That will have either happened under Pat or under systems that were created back when he was CEO. But something like Pantherlake or Novalake are 100% Pat's babies. Something like Novalake next would be like 60/40 Pat/Tan if Tan completely reformed Intel over night (but realistically we are likely talking 90/10 because even at great companies reforms take lots of time and effort). And I think realistically Novalake next next would be like 50/50 with planning/foundational work being all done under the Pat regime and execution being all done under Tan (assuming Tan completes at least substantial reforms of the design side no later than his 2 year mark). And before you ask even if he does it in one I don't think that greatly influences his impact on the prior projects because so much of the work is already done or at least will be in a year.
Not saying you are wrong here, because I honestly don't know. But I keep hearing one of the advantages of chiplets is that they speed up the design cycle, because you don't have to re-spin all of the IP blocks for the new process (my interpretation anyway). Am I missing something here?
While not wrong, talk is cheap and actions speak louder than words. Pat loved talking about "going back to delighting our customers" and "improving our say do ratio" but just because he said that was the goal didn't mean it happened. While much of what Pat gets flack about are inherited problems from the prior yahoos, he definitely had some failings on the design side. Sorting out the inherited problems and being overoptimistic on three separate counts thus getting in over his skis because he thought Intel products was hotter than it was, Pat didn't seem to be doing enough fast enough to stabilize Intel products or to reform them into a world class fabless organization. I think there were some decent ideas he pushed towards that end, but the reforms just weren't happening fast enough. Saying "this is no good, if we want to be better we should do this" doesn't mean anything if the changes take a year or more to implement. This is why I don't totally get the "Tan is already so much better than Pat" stuff because they have literally the exact same talking points. I need to see results and timely rather than plodding reforms before I say I believe in Intel products turning around like foundry is. Just a shame that a clear assessment of how good of a job Tan did won't be possible for many years. Hopefully the board doesn't try another 5 year turnaround plan in 3.5 years. For that is the garuenteed path for complete anhiliation for all parts of Intel.
I guess in this case I'm looking at Tan's track record at Cadence. From every account I can find he seems to have taken the criticism to heart and made the changes to give his customers what they needed. I don't thing that was Gelsinger's strong suit. He had too much of the Intel of old in his blood.
I think Lip-Bu is quite efficient. He replaced (voluntary or involuntary) two executives within 1 month mark, Sandra L. Rivera and Christy Pambianchi. Also the head of TD.
Ann Kelleher had announced her plans to retire before Gelsinger was replaced. I doubt that had anything to do with Tan, but he did change the succession plans from what I can tell.
Ann Kelleher had announced her plans to retire before Gelsinger was replaced. I doubt that had anything to do with Tan, but he did change the succession plans from what I can tell.
I think Lip-Bu is quite efficient. He replaced (voluntary or involuntary) two executives within 1 month mark, Sandra L. Rivera and Christy Pambianchi. Also the head of TD.
I fail to see how any of this helps CCG, DCAI, or NEX become world class. In fact you just proved my point by putting Tan on a pedestal despite not having done anything yet. Other than Sandra none of the people you mentioned even have any indication they have been sacked.
Not saying you are wrong here, because I honestly don't know. But I keep hearing one of the advantages of chiplets is that they speed up the design cycle, because you don't have to re-spin all of the IP blocks for the new process (my interpretation anyway). Am I missing something here?
Initial work is far far far more work than monolithic. And even when you do get to reuse content I would say easier not faster. To design that new CPU die was always going to take 4+ years. But if you are using an old I/O die hey you now don't need to also spend 4+ years on that and can work ahead on the next Gen or maybe just have fewer people working on that. But someone who actually has chip design experience rather than just watching it from the fab side like blueone or I IanD would have a more informed opinion on the matter than I.
I guess in this case I'm looking at Tan's track record at Cadence. From every account I can find he seems to have taken the criticism to heart and made the changes to give his customers what they needed. I don't thing that was Gelsinger's strong suit. He had too much of the Intel of old in his blood.
FWIW I don't think Tan is a disaster in the making and I think he might be the best compromise we realistically could have gotten between getting a foundry focused CEO and a design focused CEO. It also doesn't hurt that he seemed to have ran cadence well. I just don't like making buisness or engineering decisions on non results based "feelings". And at this point all we got to work with is feeling and potentially unrealistic expectations for how much Tan (maybe even anybody) can do. Because the way people talk is like Tan is going to overnight fix Intel. Which strictly speaking is impossible. Nobody can do that, and even if they could the results wouldn't show overnight. So I say all of that to say way to early to have any verdict on Tan and a full picture will be impossible until basically the end of the decade (provided the board doesn't feel like sabotaging their company anymore than they already have). Sort of like how Roy did most of the work fixing AMD (like 80%) but gets almost 0 credit for AMD's miraculous turnaround.
Saying "this is no good, if we want to be better we should do this" doesn't mean anything if the changes take a year or more to implement. This is why I don't totally get the "Tan is already so much better than Pat" stuff because they have literally the exact same talking points. I need to see results and timely rather than plodding reforms before I say I believe in Intel products turning around like foundry is. Just a shame that a clear assessment of how good of a job Tan did won't be possible for many years. Hopefully the board doesn't try another 5 year turnaround plan in 3.5 years. For that is the garuenteed path for complete anhiliation for all parts of Intel.
I did hear Pat mention half way through his tenure that he said "I wouldn't give Intel products an A rating, but at least the pipeline [for new products] is no longer broken". That tells me that the problems Intel was facing in designing and developing products were/are much harder to fix than Pat expected, even after being in the position for a bit. Of course, it's hard to tell how much effort Pat actually put into fixing this, though if you have a large layer of toxic middle management, sometimes you can only fix it by starting fresh..
Hopefully Tan applies a stronger skeptics hat than Pat did when engaging his engineering design [leadership] teams for change. (to be clear - I'm assuming Intel has a strong talent pool in the trenches ready to do the work if led properly).
.. I think Intel is at the point where the only way it's going to get measurably better is if their competitors mis-step at this point. They were handed a golden opportunity with Zen 5 being underwhelming and Intel didn't have the products to capitalize on that, even though timing gave them a (paper) tiger to launch. Nvidia literally can't make enough chips to satisfy the GPU and AI markets and Intel had nothing at all ready.
