The AI bubble is bigger than you think
prospect.org
prospect.org
The debt securitization could be an issue but one thing that stands out to me is the if GPUs are really being used as the lein or collateral, these are fundamentally depreciating assets and are marked as such even if the depreciation rates are slightly wrong.
Any new tech that renders current build out could dramatically hit this loans of course.
Even then, I don't understand why being a landlord to the place were AI is trained would be financially exciting... Wouldn't investing in NVIDIA make a lot more sense?
As long as the initial value of a blackwell chip derived from correctly forecast and discounted cash flow is sound, this is correct. On the other hand, if the initial value is a function of scarcity/demand and high manufacturing cost that's wildly detatched from actual returns then 'could be an issue' is a significant understatement.
The TFA cites a linked study which states "CoreWeave, for example, depreciates its GPUs over six years" which is way more than 'slightly wrong'. Just mapping that backward, 2020's hot new data center GPU was the A100 and they are just reaching their 5th year of service. How many large customers are lining up to pay top dollar to rent one of those 5 year-olds for the next 12 months? For most current workloads I think A100s are already net negative to keep operating in terms of opportunity cost. That power, cooling and rack space are more profitably allocated toward 2023's now mid-life H100 GPUs.
The rate of data center GPU progress has accelerated significantly in the last five years. I hardly know anything about AI workloads but even I know that newer GPU capabilities like FP8 are recent discoveries which can deflate the value of older GPUs almost overnight. With everyone now hunting for those optimization shortcuts, it's foolish to think more won't be discovered soon. The odds that this year's newly installed H200 GPUs will keep generating significant rental fees for 72 months are, IMHO, vanishingly small. Over a trillion dollars of loans have been secured by assets actually worth maybe half the claimed value. It's like 2009 sub-prime mortgages all over again.
If it's depreciated over 6 years then at 5 years it's valued at 17% of its initial price. That seems kind of reasonable?
Is 15x error in realized GPU + the debt AFTER INFLATION? I suppose but feels less likely except in some tail scenarios that have other interesting properties.
This doesn't mean that there isn't a significant possibility of market correction due to other factors but the GPU factor just seems medium sized compared to other scenarios historically. Am I missing anything in the 1st order thinking?
>I WILL TRY TO EXPLAIN THIS as simply as I can. The build-out of computing power for AI needs about $2 trillion in annual revenue by the end of the decade...
as a source for the it links the Bain article headlined "$2 trillion in new revenue needed..." but reading that their argument is
>... AI’s compute demand grows at more than twice the rate of Moore’s Law ...
>... By 2030, technology executives will be faced with the challenge of deploying about $500 billion in capital expenditures and finding about $2 trillion in new revenue
but demand is a function of price. AI companies could just stop making dumb meme videos, using a lot of compute, for free.
It's like if a food chain gives away a free donut today and two tomorrow it doesn't automatically mean it will give away 2^365 donuts in a years time and crash the economy. They could always stop the free donuts.
Even if the wanted to, they'd run out of donut mix and even if AI companies want to give away infinite compute they'd run out of energy and chips. Energy supplies for AI are pretty maxed out already. No way those are growing at twice the rate of Moore's law.
So far at least. But is this supposed bubble too big for that effect to save it? It feels like a lot of predictions are being made based on historical events like the dotcom crash, the '08 recession,etc.. and maybe those predictions are right but things have changed since. retail investment is a completely different game. "mob psychology" is also a different game with the internet being adapted at a greater scale. Even the tech driving the "AI bubble" would be scifi for someone in '08.
Certainly the variables have changed. But has the formula/game also changed?
As a result, for example xAI now builds a fossil fuel powered data center with a Saudi state-backed firm named Humain.
It will probably end up like the Twitter investment of the Saudis.
https://www.nytimes.com/2025/11/19/business/pif-saudi-arabia...
According to the article, the sovereign wealth fund PIF has many poor/toy investments and is in need of bailouts itself.
The squandering of investment money globally is unprecedented. They could literally just build (not buy!) $200 billion in affordable housing in Berlin or London and rake in 6% annually, with zero risk. Instead they build failing luxury resorts with non-functioning robot servants.