The coming global recession, and overspending by semi industry

[Paul2]

So, I lived long enough to see first major war in my lifetime. Something the last 2-3 generations were fortunate not to see after the end of WW2.

The enormous, industry-wide cap-ex earmarked for the coming decade will have to go somewhere, but we already see very inevitable consumption decline for the decade to come, and market for ultra fancy chips topping out, while 200mm fabs having better runs money-wise than at the time they were originally launched.

The culmination of the decade long lemming run led us to very expensive, low volume chips made for consumption of 2-3 clients globally, and assumptions that there absolutely will be buyers for such ridiculously expensive process.

But if you scratch deeper, you see very shaky foundations for this assumption:

1. 1. Apple — consumer market products: fluctuations on the trends/fashions/whims are just natural for it.
2. 2. Terrabit switches — network chips don't really need enormous transistor counts, they are long past the point diminishing returns on transistor counts. The overall economic gloom will not favour new products.
3. 3. Crypto miners — these chips are fully driven by economics, and cost/performance ratios which will never favour latest nodes.
   4. "Artificial intelligence," super GPUs, another novel applications — as the hype vanes, so do the investor moneys.
   
   There were extremely profitable companies who used Wall Street play to buy competitors, and who can pay for that latest node capacity, but, now, they will no longer do so. They will rather go for Wall St. pleasing entrenchment in uncertain times.
   
   ------------------------------------------------------
   
   Separately, I will underline that China is nearing to be a done for as a market, and a manufacturing location. China is more than half of world's semi market, and indirectly more than 70%.
   
   If companies can no longer manufacture in China, there is simply nowhere else for half of world's chips to go at all. The moment biggest brands like Apple will face distress, they will take their enormous supply chains with them (Apple is already panicking as they can lose the 2022 iphone 14 season because their Russian glass supplier is owned by Russian military.)
   
   Besides leading edge logic, DRAM market is critically dependent on consumer goods, sensors are also now driven by consumer sensor enriched widgets — Veblen goods.
   
   Only analog, and power electronics may somehow emerge as victors in the coming decade as the push for energy efficiency will intensify, but companies making that are all very remote from stock market, and such (which is good!)
   
   
4. ------------------------------------------------------
   
   Another black swan is the Taiwan strait crisis 2.0. If this happens, it may well be a nail in the coffin for half the industry...

 

[benb]

Best of times, worst of times scenario.
Best of times: Unlimited demand for 14nm for 10 years, 28nm demand continues, and there are plausible but boutique businesses in < 14nm which TSMC is dominating
Worst of times: Moore's law at 28nm+ provided cost reduction and performance improvements. < 28nm provides performance but cost rises, so Moore's law is now akin to inflation, you pay 20% more for 20% improvements every year.
Predictions: Yangze Memory will push out either Samsung, Hynix or Micron in DRAM. Intel foundry will dominate 22nm, Samsung Foundry 14nm, TSMC 28nm and < 14nm.

 

[jms_embedded]

Crypto miners — these chips are fully driven by economics, and cost/performance ratios which will never favour latest nodes.

Interesting... do you know of any public sources that would support that? I would have thought that total cost of ownership was more important, meaning energy-efficient was a priority (gigahashes per joule or whatever)

 

[jms_embedded]

Only analog, and power electronics may somehow emerge as victors in the coming decade as the push for energy efficiency will intensify, but companies making that are all very remote from stock market, and such (which is good!)

Um.... there are major semiconductor firms in those industries: TI, Analog Devices, Infineon, Microchip, ON Semiconductor, etc. (GaN power transistors have spawned a number of some relatively small companies in the last decade, but that's the exception.)

 

[Paul2]

Interesting... do you know of any public sources that would support that? I would have thought that total cost of ownership was more important, meaning energy-efficient was a priority (gigahashes per joule or whatever)

Biggest hash chips makers will mine a bit for themselves, and then sell once they can afford a newer node.

 

[Paul2]

Um.... there are major semiconductor firms in those industries: TI, Analog Devices, Infineon, Microchip, ON Semiconductor, etc. (GaN power transistors have spawned a number of some relatively small companies in the last decade, but that's the exception.)

Well, I am not talking about them, but countless Asian MOSFET, PMIC, and low-end discretes makers.

 

[jms_embedded]

Well, I am not talking about them, but countless Asian MOSFET, PMIC, and low-end discretes makers.

Oh, I see, you meant expansion into China's producers.

 

[Arthur Hanson]

So, I lived long enough to see first major war in my lifetime. Something the last 2-3 generations were fortunate not to see after the end of WW2.

The enormous, industry-wide cap-ex earmarked for the coming decade will have to go somewhere, but we already see very inevitable consumption decline for the decade to come, and market for ultra fancy chips topping out, while 200mm fabs having better runs money-wise than at the time they were originally launched.

The culmination of the decade long lemming run led us to very expensive, low volume chips made for consumption of 2-3 clients globally, and assumptions that there absolutely will be buyers for such ridiculously expensive process.

But if you scratch deeper, you see very shaky foundations for this assumption:

1. 1. Apple — consumer market products: fluctuations on the trends/fashions/whims are just natural for it.
2. 2. Terrabit switches — network chips don't really need enormous transistor counts, they are long past the point diminishing returns on transistor counts. The overall economic gloom will not favour new products.
3. 3. Crypto miners — these chips are fully driven by economics, and cost/performance ratios which will never favour latest nodes.
   4. "Artificial intelligence," super GPUs, another novel applications — as the hype vanes, so do the investor moneys.
   
   There were extremely profitable companies who used Wall Street play to buy competitors, and who can pay for that latest node capacity, but, now, they will no longer do so. They will rather go for Wall St. pleasing entrenchment in uncertain times.
   
   ------------------------------------------------------
   
   Separately, I will underline that China is nearing to be a done for as a market, and a manufacturing location. China is more than half of world's semi market, and indirectly more than 70%.
   
   If companies can no longer manufacture in China, there is simply nowhere else for half of world's chips to go at all. The moment biggest brands like Apple will face distress, they will take their enormous supply chains with them (Apple is already panicking as they can lose the 2022 iphone 14 season because their Russian glass supplier is owned by Russian military.)
   
   Besides leading edge logic, DRAM market is critically dependent on consumer goods, sensors are also now driven by consumer sensor enriched widgets — Veblen goods.
   
   Only analog, and power electronics may somehow emerge as victors in the coming decade as the push for energy efficiency will intensify, but companies making that are all very remote from stock market, and such (which is good!)
   
   
4. ------------------------------------------------------
   
   Another black swan is the Taiwan strait crisis 2.0. If this happens, it may well be a nail in the coffin for half the industry...

 

[Arthur Hanson]

So, I lived long enough to see first major war in my lifetime. Something the last 2-3 generations were fortunate not to see after the end of WW2.

The enormous, industry-wide cap-ex earmarked for the coming decade will have to go somewhere, but we already see very inevitable consumption decline for the decade to come, and market for ultra fancy chips topping out, while 200mm fabs having better runs money-wise than at the time they were originally launched.

The culmination of the decade long lemming run led us to very expensive, low volume chips made for consumption of 2-3 clients globally, and assumptions that there absolutely will be buyers for such ridiculously expensive process.

But if you scratch deeper, you see very shaky foundations for this assumption:

1. 1. Apple — consumer market products: fluctuations on the trends/fashions/whims are just natural for it.
2. 2. Terrabit switches — network chips don't really need enormous transistor counts, they are long past the point diminishing returns on transistor counts. The overall economic gloom will not favour new products.
3. 3. Crypto miners — these chips are fully driven by economics, and cost/performance ratios which will never favour latest nodes.
   4. "Artificial intelligence," super GPUs, another novel applications — as the hype vanes, so do the investor moneys.
   
   There were extremely profitable companies who used Wall Street play to buy competitors, and who can pay for that latest node capacity, but, now, they will no longer do so. They will rather go for Wall St. pleasing entrenchment in uncertain times.
   
   ------------------------------------------------------
   
   Separately, I will underline that China is nearing to be a done for as a market, and a manufacturing location. China is more than half of world's semi market, and indirectly more than 70%.
   
   If companies can no longer manufacture in China, there is simply nowhere else for half of world's chips to go at all. The moment biggest brands like Apple will face distress, they will take their enormous supply chains with them (Apple is already panicking as they can lose the 2022 iphone 14 season because their Russian glass supplier is owned by Russian military.)
   
   Besides leading edge logic, DRAM market is critically dependent on consumer goods, sensors are also now driven by consumer sensor enriched widgets — Veblen goods.
   
   Only analog, and power electronics may somehow emerge as victors in the coming decade as the push for energy efficiency will intensify, but companies making that are all very remote from stock market, and such (which is good!)
   
   
4. ------------------------------------------------------
   
   Another black swan is the Taiwan strait crisis 2.0. If this happens, it may well be a nail in the coffin for half the industry...

The automation of literally everything has be taken into account and is not even mentioned. Automation is growing vertically and horizontally in not only the world of physical pursuits, but mental pursuits as well. The automation of research, management and marketing are just a few of the non physical areas that are undergoing an accelerating revolution of automation of the ecosystem that even encompasses R and D in ever more areas. This trend is in its very early stages and will cause social and commercial changes that most can't even imagine. This is not a casual observation for me, but a well executed investment strategy. Harnessing this great power is the only way we will be able to deal with a vastly over populated planet without having to resort to deadly conflicts and famine. Collaboration and cooperation will be key to dealing with the challenges we face and it we don't the consequences will be beyond the imagination of most.

 

[bill01]

Interesting take! I agree completely on your take of the Chinese market, however I would be wary in discounting future innovation in terms of applications. Specifically referencing AI, while it may seem like "hype" at the moment, there is massive potential utility (and a market to go with it) in the further development of AI.

 

[VCT]

Best of times, worst of times scenario.
Best of times: Unlimited demand for 14nm for 10 years, 28nm demand continues, and there are plausible but boutique businesses in < 14nm which TSMC is dominating
Worst of times: Moore's law at 28nm+ provided cost reduction and performance improvements. < 28nm provides performance but cost rises, so Moore's law is now akin to inflation, you pay 20% more for 20% improvements every year.
Predictions: Yangze Memory will push out either Samsung, Hynix or Micron in DRAM. Intel foundry will dominate 22nm, Samsung Foundry 14nm, TSMC 28nm and < 14nm.

"Yangze Memory will push out either Samsung, Hynix or Micron in DRAM."

Do you mean push out or being pushed out?
How do Yangze Memory beat current players in best scenario?
Yangze Memory looks ok in the past few years because it's the longest bull market for DRAM.

 

[hist78]

Best of times, worst of times scenario.
Best of times: Unlimited demand for 14nm for 10 years, 28nm demand continues, and there are plausible but boutique businesses in < 14nm which TSMC is dominating
Worst of times: Moore's law at 28nm+ provided cost reduction and performance improvements. < 28nm provides performance but cost rises, so Moore's law is now akin to inflation, you pay 20% more for 20% improvements every year.
Predictions: Yangze Memory will push out either Samsung, Hynix or Micron in DRAM. Intel foundry will dominate 22nm, Samsung Foundry 14nm, TSMC 28nm and < 14nm.

I'm wondering why Intel foundry will dominate 22nm?

 

[tooLongInEDA]

I'm wondering why Intel foundry will dominate 22nm?

Perhaps no one else actually has 22nm ! Remember, TSMC 20nm was a very little used node - I think most designs went straight from 28nm to 16nm. That's assuming Intel 22nm aligns with the TSMC numbering (which it almost certainly doesn't).

 

[Paul2]

Perhaps no one else actually has 22nm ! Remember, TSMC 20nm was a very little used node - I think most designs went straight from 28nm to 16nm. That's assuming Intel 22nm aligns with the TSMC numbering (which it almost certainly doesn't).

The question is tools vs. process. TSMC "16"nm largely used same setup as 28nm. The 28 to 16 transition was mostly by perfecting advanced multiple patterning techniques, a process thing.

 

[Daniel Nenni]

The question is tools vs. process. TSMC "16"nm largely used same setup as 28nm. The 28 to 16 transition was mostly by perfecting advanced multiple patterning techniques, a process thing.

TSMC N16 was the same set-up as TSMC 20nm. Same fabs.... TSMC 20nm was the first to use double patterning. Apple used TSMC 20nm for the iPhone 6. The 6+ was N16 and Samsung 14nm.