TSMC 3 nanometers are in short supply! Has achieved absolute victory over Samsung and Intel (Korean Media)

[Daniel Nenni]

Due to the outbreak of AI craze, large technology companies have introduced 3-nanometer processes. TSMC 's 3-nanometer production capacity is expected to be sold out in 2026. Even if Samsung wants to catch up, it will still be difficult. Samsung's primary goal is to improve 3nm wafer yield in preparation for upcoming Galaxy devices.

Currently, major manufacturers such as Apple, Qualcomm, NVIDIA and AMD all use TSMC's 3nm process. TSMC has said that 3nm production capacity is in short supply. According to industry news, TSMC plans to convert some 5nm equipment to support 3nm within two years to ensure stable production and supply.

TSMC's 3nm production capacity shortage is due to increased demand for AI servers and Apple's upcoming release of a new iPhone. The iPhone 16 series is exclusively supplied by TSMC with processors and is also the largest source of revenue for TSMC’s 3nm products.

Korean media Businesskorea reported that the industry generally believes that TSMC has achieved an absolute victory over Samsung and Intel because TSMC is six months ahead of Samsung in 3-nanometer production, dispelling previous concerns.

Samsung will mass-produce 3nm for the first time in June 2022. The goal is to increase production capacity and vigorously develop Galaxy products using 3nm chips. The Galaxy Watch 7 series will be launched in July, using the 3-nanometer Exynos W1000 chip, and the Galaxy S25 series' 3-nanometer processor Exynos 2500 chip will be mass-produced in the second half of the year.

台積 3 奈米供不應求！韓媒：已取得對三星、英特爾絕對勝利

由於 AI 熱潮爆發，大型科技公司紛紛導入 3 奈米製程，使台積電三奈米產能預計 2026 年售罄，即使三星想追上仍困難重重。三星首要目標是提高 3 奈米晶片良率，為即將推出的 Galaxy 設備做準備。 目前蘋果、高通、NVIDIA 和 AMD 等大廠均採用台積 3 奈米製程，台積電曾說 3 奈米...

technews.tw

 

[Daniel Nenni]

I have been saying this for a while now, TSMC N3 has 99% of the design wins. TSMC knew this as customer N3 prepays were at a record level. Even Intel did a big N3 prepay. All of the top N3 customers will get what they ordered as per the contracts they signed with TSMC. Any wafer demand above and beyond that will be a challenge, a challenge that TSMC has seen before so no worries.

Now that Samsung 5/4nm is yielding and Intel has 4nm in production TSMC would be wise to convert N5 to N3. They are similar processes so that should not be an issue. Especially the new AZ fabs, they can all be N3.

The other interesting comment I made that is coming to fruition is that TSMC N2 will again dominate the market. I am seeing as many, if not more, N2 design starts than N3. I would have guessed otherwise but that is what I am seeing. This puts TSMC in a dominant position for another 5 years, absolutely.

 

[BruceA]

TSM should break 1T valuation very soon, barring a political event

 

[hist78]

Due to the outbreak of AI craze, large technology companies have introduced 3-nanometer processes. TSMC 's 3-nanometer production capacity is expected to be sold out in 2026. Even if Samsung wants to catch up, it will still be difficult. Samsung's primary goal is to improve 3nm wafer yield in preparation for upcoming Galaxy devices.

Currently, major manufacturers such as Apple, Qualcomm, NVIDIA and AMD all use TSMC's 3nm process. TSMC has said that 3nm production capacity is in short supply. According to industry news, TSMC plans to convert some 5nm equipment to support 3nm within two years to ensure stable production and supply.

TSMC's 3nm production capacity shortage is due to increased demand for AI servers and Apple's upcoming release of a new iPhone. The iPhone 16 series is exclusively supplied by TSMC with processors and is also the largest source of revenue for TSMC’s 3nm products.

Korean media Businesskorea reported that the industry generally believes that TSMC has achieved an absolute victory over Samsung and Intel because TSMC is six months ahead of Samsung in 3-nanometer production, dispelling previous concerns.

Samsung will mass-produce 3nm for the first time in June 2022. The goal is to increase production capacity and vigorously develop Galaxy products using 3nm chips. The Galaxy Watch 7 series will be launched in July, using the 3-nanometer Exynos W1000 chip, and the Galaxy S25 series' 3-nanometer processor Exynos 2500 chip will be mass-produced in the second half of the year.

台積 3 奈米供不應求！韓媒：已取得對三星、英特爾絕對勝利

由於 AI 熱潮爆發，大型科技公司紛紛導入 3 奈米製程，使台積電三奈米產能預計 2026 年售罄，即使三星想追上仍困難重重。三星首要目標是提高 3 奈米晶片良率，為即將推出的 Galaxy 設備做準備。 目前蘋果、高通、NVIDIA 和 AMD 等大廠均採用台積 3 奈米製程，台積電曾說 3 奈米...

technews.tw

Foundry business model does work and works very well. This "sharing the challenges and prosperity" spirit propels every company in the ecosystem forward.

 

[Mooredaddy]

TSM should break 1T valuation very soon, barring a political event

Don’t see how TSM doesn’t eventually become the most valuable company in the world at some point. I am more sure about this than I was about nvidia in 2016.

 

[BruceA]

Don’t see how TSM doesn’t eventually become the most valuable company in the world at some point. I am more sure about this than I was about nvidia in 2016.

I am still flabbergasted that Buffett took a 5B stake and than liquidated TSM while still holding Apple citing too much political risk. Apple, Nvidia, Qualcomm, AMD, Intel and so many more all depend solely on TSMC. Strange wall street and others don’t see that value and price LAM, AMAT and ASML at crazy valuations. If and when they realize they should sky but……..

 

[Mooredaddy]

I am still flabbergasted that Buffett took a 5B stake and than liquidated TSM while still holding Apple citing too much political risk. Apple, Nvidia, Qualcomm, AMD, Intel and so many more all depend solely on TSMC. Strange wall street and others don’t see that value and price LAM, AMAT and ASML at crazy valuations. If and when they realize they should sky but……..

It didn’t stand up to any sort of scrutiny. I would have loved to ask Warren when he was asked this exact question at berkshires conference and explain to him how his decision made no sense

 

[Daniel Nenni]

I am still flabbergasted that Buffett took a 5B stake and than liquidated TSM while still holding Apple citing too much political risk. Apple, Nvidia, Qualcomm, AMD, Intel and so many more all depend solely on TSMC. Strange wall street and others don’t see that value and price LAM, AMAT and ASML at crazy valuations. If and when they realize they should sky but……..

Warren has a new team of semiconductor research people, I would not count him out of the semiconductor industry yet. I do agree that his TSM move was not inspiring.

Ten years ago I told a couple of investment firms that TSMC and NVDA would win because of the ecosystems they were building. Unfortunately I also told them AMD was doomed due to the Intel ecosystem.

 

[BruceA]

Warren has a new team of semiconductor research people, I would not count him out of the semiconductor industry yet. I do agree that his TSM move was not inspiring.

Ten years ago I told a couple of investment firms that TSMC and NVDA would win because of the ecosystems they were building. Unfortunately I also told them AMD was doomed due to the Intel ecosystem.

The world was Intel’s to dominated but came down to Andy in some ways. He was a very different kind of manager and personality. That strong and overwhelming personality set the stage for little real strong visionary leaders after him. One could argue Craig was the last leader at at Intel that had some chops. After that the culture, senior leadership and Board was a bunch of zombies, very different stage and company that Morris “Founder” Jensen and Lisa put in place.

https://www.intel.com/content/dam/www/public/us/en/documents/pdf/foundry/sunit-rikhi-keynote-semi-gartner-market%20symposium-presentation.pdf

 

[Mooredaddy]

The world was Intel’s to dominated but came down to Andy in some ways. He was a very different kind of manager and personality. That strong and overwhelming personality set the stage for little real strong visionary leaders after him. One could argue Craig was the last leader at at Intel that had some chops. After that the culture, senior leadership and Board was a bunch of zombies, very different stage and company that Morris “Founder” Jensen and Lisa put in place.

https://www.intel.com/content/dam/www/public/us/en/documents/pdf/foundry/sunit-rikhi-keynote-semi-gartner-market%20symposium-presentation.pdf

Brian K might be up there with one of the worst CEO's of all time. Took a legendary company and stalled it into a nose dive via prioritizing financial engineering.

 

[Fred Chen]

Now that Samsung 5/4nm is yielding and Intel has 4nm in production TSMC would be wise to convert N5 to N3. They are similar processes so that should not be an issue. Especially the new AZ fabs, they can all be N3.

Reducing 5nm capacity seems to be a risk for losing customers. Or do they already see dropping demand there?

 

[hist78]

Brian K might be up there with one of the worst CEO's of all time. Took a legendary company and stalled it into a nose dive via prioritizing financial engineering.

The world was Intel’s to dominated but came down to Andy in some ways. He was a very different kind of manager and personality. That strong and overwhelming personality set the stage for little real strong visionary leaders after him. One could argue Craig was the last leader at at Intel that had some chops. After that the culture, senior leadership and Board was a bunch of zombies, very different stage and company that Morris “Founder” Jensen and Lisa put in place.

https://www.intel.com/content/dam/www/public/us/en/documents/pdf/foundry/sunit-rikhi-keynote-semi-gartner-market%20symposium-presentation.pdf

I started thinking about why Intel's board of directors kept choosing those wrong CEOs and can't help Intel to find a new and better direction. I am also wondering why, despite all its talents and resources, Intel kept missing new hot markets such as mobile, GPU, low power, and AI.

Among several possible reasons, one important cause is Intel's business model. A wrong business model caused Intel to select those wrong board of directors who turned around to choose those wrong CEOs. And then those wrong CEOs under the influence of the wrong business model made the wrong decisions repeatedly.

I'm afraid Intel and Pat Gelsinger's IDM 2.0 business strategy doesn't provide enough changes to alter Intel's failing business model.

 

[Xebec]

Don’t see how TSM doesn’t eventually become the most valuable company in the world at some point. I am more sure about this than I was about nvidia in 2016.

I’ll take the contrarian view here. While I think TSMC is / should be extremely valuable; I don’t see how a company producing commodity+ should be the most valuable.

Examples: In the Apple or AI stack, TSMC is foundational but not offering the majority of “value” in the product. What % of the total profit in an iPhone or AI accelerator goes to TSMC?

 

[Mooredaddy]

I’ll take the contrarian view here. While I think TSMC is / should be extremely valuable; I don’t see how a company producing commodity+ should be the most valuable.

Examples: In the Apple or AI stack, TSMC is foundational but not offering the majority of “value” in the product. What % of the total profit in an iPhone or AI accelerator goes to TSMC?

The way I see it. TSMC wins in every scenario. Because everyone has to go to them, it doesn’t matter who wins either design generation. Also I’m not sure you can call it a commodity if they are the ONLY one capable of delivering at the leading edge on time, yield, and scale. Seems like an oxymoron to call it a commodity, despite technically being an input good by definition. I really feel dram/nand and power controllers fit the commodity classification better but I can see where you are coming of for sure x

 

[Xebec]

The way I see it. TSMC wins in every scenario. Because everyone has to go to them, it doesn’t matter who wins either design generation. Also I’m not sure you can call it a commodity if they are the ONLY one capable of delivering at the leading edge on time, yield, and scale. Seems like an oxymoron to call it a commodity, despite technically being an input good by definition. I really feel dram/nand and power controllers fit the commodity classification better but I can see where you are coming of for sure x

Mostly agree -

I wrote commodity+ because I couldn't think of something above the word "commodity" but below high value "unique" products like Nvidia Blackwell, etc. sorry for confusing wording here.

I just struggle with "to the moon" for TSMC given if you look at a finished product like Blackwell, iPhones, etc, TSMC only captures a small portion of that profit (value) while Nvidia or Apple make a much higher proportion.

Compared to the profit value that Nvidia and Apple get, TSMCs slice isn't nearly as big (hence commodity+).

Thanks

 

[nghanayem]

Mostly agree -

I wrote commodity+ because I couldn't think of something above the word "commodity" but below high value "unique" products like Nvidia Blackwell, etc. sorry for confusing wording here.

I just struggle with "to the moon" for TSMC given if you look at a finished product like Blackwell, iPhones, etc, TSMC only captures a small portion of that profit (value) while Nvidia or Apple make a much higher proportion.

Compared to the profit value that Nvidia and Apple get, TSMCs slice isn't nearly as big (hence commodity+).

Thanks

The devil's advocate argument that TSMC is not capturing very much of the value is that TSMC has better GMs and OMs than most fabless and even some systems companies. It is a very unnatural thing... Which is kind of funny because TSMC wafers aren't really that expensive. But I guess that is just a product of all of the overhead associated with actually designing and then selling your chips that TSMC doesn't have to bear, and TSMC's money printer legacy fabs (TI vs any fabless peer is a great example of that).

Reducing 5nm capacity seems to be a risk for losing customers. Or do they already see dropping demand there?

Since it would seem the toolsets are very similar they can always just flex capacity back in under a year if they see an unforeseen demand increase for N5. Based on the fact they are converting capacity over, similar tool sets, more customer value per wafer, and higher revenue/wafer it seems like long term TSMC wants to have N5 be a lesser used node (similar to how TSMC pushed most of the 40nm customers to 28nm so they could focus on just one "advanced" planar node).

 

[Fred Chen]

Since it would seem the toolsets are very similar they can always just flex capacity back in under a year if they see an unforeseen demand increase for N5. Based on the fact they are converting capacity over, similar tool sets, more customer value per wafer, and higher revenue/wafer it seems like long term TSMC wants to have N5 be a lesser used node (similar to how TSMC pushed most of the 40nm customers to 28nm so they could focus on just one "advanced" planar node).

I'm thinking from the customer's point of view, the big players like Apple, AMD, etc. were already on board with the transition to 3nm. It would be new 5nm customers who would have trouble getting the capacity.

 

[Daniel Nenni]

I'm thinking from the customer's point of view, the big players like Apple, AMD, etc. were already on board with the transition to 3nm. It would be new 5nm customers who would have trouble getting the capacity.

True and TSMC has no competition at N3 so the margins will be higher than previous nodes. I have never heard big TSMC customers openly support TSMC price hikes before. Exciting times.

 

[osnium]

I started thinking about why Intel's board of directors kept choosing those wrong CEOs and can't help Intel to find a new and better direction. I am also wondering why, despite all its talents and resources, Intel kept missing new hot markets such as mobile, GPU, low power, and AI.

Among several possible reasons, one important cause is Intel's business model. A wrong business model caused Intel to select those wrong board of directors who turned around to choose those wrong CEOs. And then those wrong CEOs under the influence of the wrong business model made the wrong decisions repeatedly.

I'm afraid Intel and Pat Gelsinger's IDM 2.0 business strategy doesn't provide enough changes to alter Intel's failing business model.

Could it be that Intel’s talent was simply over-rated and not competent enough to compete in other markets?

 

[hist78]

Could it be that Intel’s talent was simply over-rated and not competent enough to compete in other markets?

Yes, definitely there are some Intel's talent who are not up to the speed of the market competition. But why they can stay at Intel longer than they deserved and sometimes they even became the leaders/executives of Intel?

I believe Intel's business model allows those people to thrive at Intel and cause long term damages without even knowing it.