The AI Bubble Is Going to Pop
A summary of the links shared in a fascinating Mastodon thread about the current AI hype cycle and the bubble that’s about to pop.
What’ll Happen If We Spend Nearly $3tn on Data Centres No One Needs?
The FT features a ferocious takedown (Elder 2025) (paywalled, open archive version here), of a Morgan Stanley AI booster briefing. It opens with a 1990s-themed web 1.0 graphic featuring a portrait of Sam Altman to set the tone.
FT quotes a wildly optimistic Morgan Stanley briefing:
GenAI revenues could exceed $1tr by 2028, with close to 70% variable margins, compared to just $45bn in 2024. […] ROI of AI should already be positive this year, generating $50bn in revenues, and that this will grow to exceed $1tr/year by 2028
To put that into perspective, that’s in the ballpark of half of all S&P500 investment!
Hyperscaler funding of $300bn to $400bn a year compares with annual capex last year for all S&P 500 companies of about $950bn.
After the telecoms bubble, the money was gone. But at least we still had all the internet infrastructure built with that money. This time may be different. Nvidia GPUs are not general-purpose infrastructure that are useful for any type of compute; they’re specifically optimized for running Large Language Models. And they obsolete really fast.
Morgan Stanley estimates that $1.3tn of data centre capex will pay for land, buildings and fit-out expenses. The remaining $1.6tn is to buy GPUs from Nvidia and others. Smarter people than us can work out how to securitise an asset that loses 30 per cent of its value every year, and good luck to them.
Even worse: those GPUs need a firehose of electricity to run. Even if the current owners go bankrupt, anybody who wants to pick up those assets is going to be faced with the same operational problem: to pay for the electricity. That is, assuming you’ll be able to buy it. Which is a wild assumption in itself. And it’s where the AI hype runs into the unfolding climate catastrophe really hard.
America needs to find an extra 45GW for its data farms, says Morgan Stanley. That’s equivalent to about 10 per cent of all current US generation capacity, or “23 Hoover Dams”, it says.
Build out power capacity to the tune of 23 Hoover Dams in the coming 3 years? Never gonna happen.
Back to the investment. How is all that money supposed to be earned back? It’s worth zooming in on risk embedded in the assumption made by Morgan Stanley, of outrageous revenue growth, from $45bn in 2024 to $1tn in 2028.
When the base case is for 1,900 per cent revenue growth by 2028, isn’t it worth considering the risk of a shortfall?
The article compares the current investment cycle in AI with the telecoms bubble of the late 1990s. Except, this is much bigger. As in: 10 times bigger.
In 2000, at the telecoms bubble’s peak, communications equipment spending topped out at $135bn annualised. The internet hasn’t disappeared, but most of the money did. […] Peak data centre spend this time around might be 10 times higher
Sounds like a repeat of the GFC waiting to happen, to me.
The AI Bubble Is so Big It’s Propping up the US Economy (for Now)
To put that into context: Brian Merchant points out that this investment level is so high, that it’s the main prop keeping the US economy afloat. (Merchant 2025)
He analyses a Wall Street Journal report that makes his head spin.
I’ll just repeat that. Over the last six months, capital expenditures on AI—counting just information processing equipment and software, by the way—added more to the growth of the US economy than all consumer spending combined. You can just pull any of those quotes out—spending on IT for AI is so big it might be making up for economic losses from the tariffs, serving as a private sector stimulus program.
The Hater’s Guide To The AI Bubble
Ed Zitron has been banging this drum for a while. He makes additional points. So much, in fact, that you should read his whole article (Zitron 2025).
He does a drilldown on the financials, and concludes the revenue is puny and partly double-counted. There is no profit.
Yes, generative AI has functionality. There are coding products and search products that people like and pay for. As I have discussed above, none of these companies are profitable, and until one of them is profitable, generative AI-based companies are not real businesses.
I believe that the generative AI market is a $50 billion revenue industry masquerading as a $1 trillion one, and the media is helping.
Zittran concludes, that the stockmarket depends on Nvidia keeping up its rocket growth rate. Which in turn depends on the hyperscalers keeping up their investment. While none of them actually make money on AI.
We are in a bubble. Generative AI does not do the things that it’s being sold as doing, and the things it can actually do aren’t the kind of things that create business returns, automate labor, or really do much more than one extension of a cloud software platform. The money isn’t there, the users aren’t there, every company seems to lose money and some companies lose so much money that it’s impossible to tell how they’ll survive.