TL;DR
The short version
Most of this month's AI bubble warnings are stand-ins for one question: will demand grow faster than the build-out? The signal most worth watching is how concentrated AI spending is, and the Anthropic and OpenAI IPOs are about to make it public.
On The AI Daily Brief, Nathaniel Whittemore takes the warnings one at a time: neocloud distress, GPUs waiting in Amazon warehouses, data-center debt on insurer and pension balance sheets, and revenue concentration. For each, he says where it stands and what would make him more worried.
Built on Nathaniel Whittemore's episode (24 October 2026). Figures checked against Gartner, Ramp, Business Insider and reporting on Anthropic's draft prospectus.
One question under all the noise
Tech stocks have slid for three weeks, the Anthropic IPO is close, and bubble talk is back. Nathaniel Whittemore goes through the warnings one at a time. For each one he asks where it stands today and what would make him more worried.
In many ways, everything else is just a proxy fight for the big question of how much AI demand is there going to be and whether the value of that demand is ultimately higher than the amount being spent on the infrastructure.
Nathaniel Whittemore, The AI Daily Brief
He also asks for precise words. A 20–30% bear market isn't a dot-com bust, which took more than 75% off the Nasdaq. A stock crash isn't a credit crisis, where defaults spread through banks and need a central bank to stop them. Much of the bubble talk mixes the three up.
Neoclouds: real stress, small sector
The loudest stress is in neoclouds, the GPU-rental specialists. Firmus withdrew its IPO. Microsoft pulled out of an Nscale site in Essex over power. Gartner says AI data-center owners need $8.5 trillion in revenue through 2030 to justify their plans, against no more than $3.6 trillion on offer. It expects more than 80% of neoclouds to stop operating independently, which is not the same as going bankrupt.
Gartner (14 Oct 2026) and The Information
Whittemore's point is scale. These are small companies next to the biggest AI firms, and the industry is propping them up. Nvidia alone has put more than $170 billion on the line through guarantees and backstops.
What would worry him: a neocloud fails and nobody rescues it. Stress spreads to debt on data centers leased to Meta, Microsoft or Google. Or a hyperscaler walks away from capacity because it doesn't need it, not because the grid isn't ready.
Warehoused GPUs and hidden debt
Critics seized on a Business Insider report that AWS is taking delivery of billions of dollars in Nvidia GB300 racks before it has the power to run them. Whittemore reads it the other way round. Dot-com dark fiber was "too little demand for already installed infrastructure. This is too little infrastructure to service demand."
Debt is the scarier question after 2008. Insurers bought more than $740 million of CoreWeave debt tied to its Meta contract. Thirteen public pension funds made $138 billion in gains from AI stocks since 2022. Exposure is real, but $740 million is small next to those portfolios.
The subprime comparison misses what made subprime dangerous. Mortgage slices were rated as safe as Treasuries. Data-center debt, Whittemore notes, trades as ordinary corporate bonds, and "no one is pretending" it's that safe. He would worry if exposure jumped to tens of billions in two or three PE-owned insurers, or if the debt got repackaged with ratings above the loans underneath.
The signal that matters: who's paying
Revenue quality is the warning he takes most seriously. About a fifth of Anthropic's first-half revenue came from two customers. Some of that is likely hyperscalers reselling Claude, which is sturdier than it sounds.
The bigger issue is on Ramp: the top 1% of firms drive 80% of spending on OpenAI and Anthropic, at about $9,600 per employee per month. Whittemore cites Ramp data that 80% of that 1% are VC-backed startups. If venture money dries up, much of that spend could go with it.
So the real question is whether this concentration lasts. Either it's how AI demand will always look, or the rest of the economy is earlier on the same curve. Ramp's own analyst says per-employee spend is "stuttering" below its July high, so the answer isn't in yet.
The IPO ends the guessing
Right now, this all rests on samples and leaks. "Once Anthropic and OpenAI go public, they'll have nowhere to hide," Whittemore says. Audited quarterly numbers will show whether revenue is broadening.
Valuations are already softening. OpenAI's round is landing at $1.25 trillion, down from $1.4 trillion in earlier reports. US IPOs raised under $600 million this month. Whittemore's test is simple: Anthropic pricing at $1.5 trillion after months of $2 trillion talk is fine. Pricing there and still trading badly would confirm the bears.
Keep the incentives in view. Blackstone's Jon Gray, who says AI is "hoping you can keep up with demand," holds neocloud positions. Citadel's Ken Griffin, who calls it a bubble, profits from volatility. Whittemore is a self-described long-term optimist.
