Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/china-has-spent-billions-of-dollars-building-far-too-many-data-centers-for-ai-and-compute-could-it-lead-to-a-huge-market-crash.22511/page-4
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021770
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

China has spent billions of dollars building far too many data centers for AI and compute - could it lead to a huge market crash?

The foundry model works wonder in semiconductor because one foundry fab can serve many fabless companies on the same production line, thus scalable in volume and huge capital cost burden shared by many fabless companies.

Car/EV manufacturing takes up huge space and similarly huge capital but cannot mix run different models or brands on the same production line, making the foundry model and cost sharing unfeasible.

Until someone figures out that, IDM model probably will continue to dominate in auto industry.
 
Jert said:
The foundry model works wonder in semiconductor because one foundry fab can serve many fabless companies on the same production line, thus scalable in volume and huge capital cost burden shared by many fabless companies.

Car/EV manufacturing takes up huge space and similarly huge capital but cannot mix run different models or brands on the same production line, making the foundry model and cost sharing unfeasible.

Until someone figures out that, IDM model probably will continue to dominate in auto industry.
Click to expand...
Mostly agree on this point. Stamping/body shop tooling is very specific to the model being made. General assembly is also pretty specific to the model being made.

Battery/drivetrain you can support a few different models, and paint is somewhat flexible.

You can have different variants of a car on the same line. But you can't for example make an SUV and a compact car on the same line.
 
Jert said:
The foundry model works wonder in semiconductor because one foundry fab can serve many fabless companies on the same production line, thus scalable in volume and huge capital cost burden shared by many fabless companies.
Click to expand...
Real men have fabs until it's too expensive.
 
Daniel Nenni said:
The automotive industry itself is a cyclical industry with consolidation cycles. The US is in a cooling period so consolidation is probably coming. China is still growing but tariffs may stop that. Hard to predict with tariffs looming.
Click to expand...
Thought this was an interesting view on the China EV market.

Jato Dynamics data show that 169 domestic and foreign auto brands are fighting it out in China, but only 14 have a market share higher than 2%. Last year, 86 brands produced a total of 327 EV models, not including hybrids.
The competition has depressed prices to levels unfathomable outside China. As Tesla and other automakers struggle to produce EVs to sell for less than $30,000 in the United States, BYD's entry-level Seagull electric hatchback - the star of the 2023 Shanghai auto show - is priced below $10,000.

 
KevinK said:
Thought this was an interesting view on the China EV market.

Jato Dynamics data show that 169 domestic and foreign auto brands are fighting it out in China, but only 14 have a market share higher than 2%. Last year, 86 brands produced a total of 327 EV models, not including hybrids.
The competition has depressed prices to levels unfathomable outside China. As Tesla and other automakers struggle to produce EVs to sell for less than $30,000 in the United States, BYD's entry-level Seagull electric hatchback - the star of the 2023 Shanghai auto show - is priced below $10,000.

Click to expand...
Size of Subaru outback
PHEV 85km ev range
3-4L/100km fuel efficiency when battery is depleted
One tank of fuel could reach 2000km
Priced less than 20k USD

This product can bring meaningful savings to users.

https://twitter.com/x/status/1914999155166265759
 
KevinK said:
Thought this was an interesting view on the China EV market.

Jato Dynamics data show that 169 domestic and foreign auto brands are fighting it out in China, but only 14 have a market share higher than 2%. Last year, 86 brands produced a total of 327 EV models, not including hybrids.
The competition has depressed prices to levels unfathomable outside China. As Tesla and other automakers struggle to produce EVs to sell for less than $30,000 in the United States, BYD's entry-level Seagull electric hatchback - the star of the 2023 Shanghai auto show - is priced below $10,000.

Click to expand...

As I suspected, most of them outsource their manufacturing to bigger automakers. Tiny sales volumes can't feed a full sized car factory.
 
Could the time latency severely limit the value of the Chinese data centers as distributed compute power will be the future as the cost of memory comes down and speed increases?
 
Arthur Hanson said:
Could the time latency severely limit the value of the Chinese data centers as distributed compute power will be the future as the cost of memory comes down and speed increases?
Click to expand...
As a rule, mainland China has terrible internet speeds, regardless of how many gigabit their telecom providers claim to provide. These gigabits only work until the next telecom closet, and local network exchange.
 
Arthur Hanson said:
Could the time latency severely limit the value of the Chinese data centers as distributed compute power will be the future as the cost of memory comes down and speed increases?
Click to expand...
Even in Bitcoin mining, a purely distributed and decentralized computation that doesn't require memory and bandwidth, data centers are still far more efficient than individual miners. This is even more true for AI computing, which requires memory and bandwidth.
 
You must log in or register to reply here.
Back
Top