Samsung talking smack about passing TSMC.

[Daniel Nenni]

第一季前十大晶圓代工業者營收季減近二成，第二季仍將持續下滑

TrendForce 研究顯示，受終端需求持續疲弱以及淡季效應加乘影響，第一季全球前十大晶圓代工業者營收季跌幅達 18.6%，約 273 億美元。本次排名最大變動為格羅方德（GlobalFoundries）超越聯電（UMC）拿下第三名，以及高塔半導體（Tower）超越力積電（PSMC）及世界先進（V...

technews.tw

No need 2025, tsmc has achieved 60%+ in 2023 Q1.
2019 Q1 foundry market share: TSMC 48.1%; Samsung 19.1%
2023 Q1 foundry market share: TSMC 60.1%; Samsung 12.4%

I may have to up my prediction to 70% market share for TSMC at 2025. But then again you are quoting from a very inaccurate source. TFARCE.....

 

[delong.height]

I understand your logic but I look at it differently. QCOM and NVDA bounce between foundries, that is part of their company culture. Today they bounce between TSMC and Samsung. Right now they are at TSMC due to Samsung PDK failures but could bounce back at anytime. Next bounce, in my opinion, would be to Intel at a loss to Samsung.

Intel and Apple have a checkered past so I don't think they will bounce again. Apple had bounced between Samsung and TSMC but have been at TSMC exclusively since the iPhone 6+. AMD/Xilinx will stay at TSMC as well. AMD due to the Intel competition and Xilinx due to Intel competition. That is the problem with being an IDM foundry, you compete with your customers.

Intel and Samsung both have a foundry day coming up this month. It will be interesting to see the positioning. Intel had said they will be the #2 foundry in 10 years. Samsung has said they will be #1 on different occasions. Last I read it was by 2030. Of those two things happening I would bet it would be IFS as #2 by 2030. Any takers?

I agree with majority of these. They are very valid.

AMD/Xilinx will stay at TSMC as well. AMD due to the Intel competition and Xilinx due to Intel competition. That is the problem with being an IDM foundry, you compete with your customers.

I don't think acquiring AMD as a foundry customer would be that important to Intel since Intel and AMD have too many overlaps, and Intel controls majority of laptop, desktop, enterprise, cloud CPU markets today. With rising competition in FPGA and GPU markets. They also compete in other markets as well, like edge, IPU, etc. They probably don't want to offer AMD with any more helps to their internal manufacturing capabilities.

I don't think AMD will go to Intel anytime soon. And I don't think Intel will be happy to take AMD as a customer either.

Forcing AMD to have a little choice when it comes to manufacturing is extremely beneficial when Intel can benefit either way. If IFS sucks, internal product teams can switch to TSMC, with some chiplets done at IFS, and if IFS succeeds, they can control their own supply and benefit from vertical integration.


For others fabless, I don't think it'll matter much to them since Intel only participates in PC, server and edge. They don't participate in mobile, automotive (although they do with their Mobileye), networking.

 

[lefty]

I don't think AMD will go to Intel anytime soon.

First sign that this is happening is when we see the devil ice skating

 

[ta152h]

I agree with majority of these. They are very valid.

I don't think acquiring AMD as a foundry customer would be that important to Intel since Intel and AMD have too many overlaps, and Intel controls majority of laptop, desktop, enterprise, cloud CPU markets today. With rising competition in FPGA and GPU markets. They also compete in other markets as well, like edge, IPU, etc. They probably don't want to offer AMD with any more helps to their internal manufacturing capabilities.

I don't think AMD will go to Intel anytime soon. And I don't think Intel will be happy to take AMD as a customer either.

Forcing AMD to have a little choice when it comes to manufacturing is extremely beneficial when Intel can benefit either way. If IFS sucks, internal product teams can switch to TSMC, with some chiplets done at IFS, and if IFS succeeds, they can control their own supply and benefit from vertical integration.


For others fabless, I don't think it'll matter much to them since Intel only participates in PC, server and edge. They don't participate in mobile, automotive (although they do with their Mobileye), networking.

I actually think Intel HAS to offer IFS to AMD, not because they need them, but because it's a terrible look if they turn customers away for competition.

As I mentioned, I think AMD is safe, but as I think about it, possibly less safe than I initially mentioned.

Yeah, Intel probably doesn't want them, would definitely accept them if they were willing to pay same as customers. Intel own desktops and mobile, over 80%, and it's less clear if their higher performance nodes would be as attractive in servers. But, either way, yes, they'd be tossing an advantage away, which I believe currently gives them the performance crown in single-threaded performance. But, they'd look like clowns if they said, nope, only so and so can do business with us, because we compete. No way they can that. Plus, Mr. Pat has said he'd be interested in their business.

So, it really falls to AMD. And AMD might want to use them, for exactly the reasons why you say Intel shouldn't let them use their fabs. Not only would AMD have access to potentially superior nodes to TSMC in some metrics, Intel would find it less profitable to engage in price wars, as they would make money either way (more on their own stuff, but it would change the calculation as to where the most advantageous price would be). Not only that, it would keep their fabs busy, again making it less imperative to drop prices as much.

I am not suggesting it will happen, but you and I both know crazy stuff happens all the time in this world of ours (and it keeps getting crazier), and this really isn't that out of the question. I think the odds are low in the near term though, so I am with you on that.

 

[Daniel Nenni]

I actually think Intel HAS to offer IFS to AMD, not because they need them, but because it's a terrible look if they turn customers away for competition.

As I mentioned, I think AMD is safe, but as I think about it, possibly less safe than I initially mentioned.

Yeah, Intel probably doesn't want them, would definitely accept them if they were willing to pay same as customers. Intel own desktops and mobile, over 80%, and it's less clear if their higher performance nodes would be as attractive in servers. But, either way, yes, they'd be tossing an advantage away, which I believe currently gives them the performance crown in single-threaded performance. But, they'd look like clowns if they said, nope, only so and so can do business with us, because we compete. No way they can that. Plus, Mr. Pat has said he'd be interested in their business.

So, it really falls to AMD. And AMD might want to use them, for exactly the reasons why you say Intel shouldn't let them use their fabs. Not only would AMD have access to potentially superior nodes to TSMC in some metrics, Intel would find it less profitable to engage in price wars, as they would make money either way (more on their own stuff, but it would change the calculation as to where the most advantageous price would be). Not only that, it would keep their fabs busy, again making it less imperative to drop prices as much.

I am not suggesting it will happen, but you and I both know crazy stuff happens all the time in this world of ours (and it keeps getting crazier), and this really isn't that out of the question. I think the odds are low in the near term though, so I am with you on that.

With chiplets AMD can use multiple foundries but with packaging it is more complicated. I was told that TSMC will not package non TSMC chiplets. This would be a problem for AMD, Apple, NVDA, and others. Intel on the other hand said they would package other foundry chiplets (as they do for internal product) so this could be a competitive advantage for IFS. Samsung is still working on their packaging offering. We will know more at the Samsung Foundry Forum on 6/27.