SWNFT
Bubbles and Busts

Is AI a Bubble? The Case For and Against

Is AI a bubble? Nobody knows yet. Here are the five signs historians use, with the case for and against as at September 2026, so you can judge it yourself.

James Killick7 min read

TL;DR

  • Nobody can say yet whether AI is a bubble. This post gives both sides, as at 4 September 2026.
  • Historians look for five signs: a real new idea, prices ahead of profit, borrowed money, a crowded trade and first cracks.
  • In December 2025 the Bank of England said US share valuations were close to their most stretched since the dot-com bubble. In January 2026 the IMF said overvaluation was about half the dot-com level.
  • Four big tech firms planned about $724 billion of spending for 2026, on average analyst estimates reported on 26 July 2026.
  • This is education, not financial advice. We make no call.

Is AI a bubble? Short answer: nobody knows yet, and I won't pretend to. Some signs match past bubbles, like stretched prices and huge spending. Others don't, like real revenue and less borrowing. Bubbles usually get their name after they pop. So I'll lay out both sides as at 4 September 2026. You make the call.

Risk note: most AI-linked tokens and many AI-linked shares can lose value fast, and scams love the AI label. Only use money you can afford to lose.

What does "bubble" even mean?

It means prices rose far past what looked sensible, then fell hard. That's the loose version. It's also the honest one.

Two economists, Markus Brunnermeier and Isabel Schnabel, studied 23 bubble episodes over almost 400 years. They admit their data can't truly measure when a price strays from its real value. They use "bubble" for a price move people thought was excessive, which often ended in a sharp fall.

So "bubble" is mostly a label we stick on afterwards. Hold onto that. It matters for everything below.

What are the five signs historians use?

Economists Charles Kindleberger and Hyman Minsky described how booms tend to run. This is our plain-English version of their pattern. It's not an official test.

| Sign | What it looks like | AI as at Sep 2026 | |---|---|---| | 1. A real new idea | A new technology changes what people expect | Yes | | 2. Prices ahead of profit | Valuations rise faster than earnings | Mixed | | 3. Borrowed money | Debt pays for the boom | Rising | | 4. A crowded trade | A few names hold most of the money | Yes, per the Bank of England | | 5. First cracks | Prices wobble on bad news | Early signs |

Let's walk through each one.

Sign 1: Is there a real new idea?

Yes. Here's where AI isn't like tulips. Brunnermeier and Schnabel say bubbles are usually triggered by a "displacement". That's a shock that changes expectations. Railways and the internet were both real inventions. Both still ended in bubbles.

AI is real too. Alphabet's cloud revenue rose 82% in the quarter it reported in July 2026, according to Bloomberg via Fortune. So a real idea isn't proof of a bubble. It isn't proof that prices are fair either.

Sign 2: Are prices ahead of profit?

Mixed. The two main official voices disagree, mate.

The Bank of England said in December 2025 that US share valuations were close to their most stretched since the dot-com bubble. It said this raised the risk of a sharp fall.

The IMF's January 2026 blog took a milder view. It put overvaluation of the broad US index at only about half the dot-com level. It said price-to-earnings ratios had risen less because earnings were stronger.

By July 2026 the Bank of England said valuations had become more stretched on some metrics. It called the earnings forecasts behind them "highly uncertain."

Sign 3: Is borrowed money paying for it?

This is where the risk is growing. Here's the thing: how a boom is paid for matters more than what it buys. Brunnermeier and Schnabel found that Railway Mania in the 1840s came with a lending boom and a banking crisis. The dot-com bubble was paid for more with share money. It ended in a mild recession with no major bank failure.

For AI, the Bank of England's December 2025 report said spending could pass five trillion US dollars over five years. It said about half was expected to be financed from outside, mostly through debt. The IMF noted that debt financing was becoming more common.

The four biggest spenders (Alphabet, Microsoft, Amazon and Meta) were on track for about $724 billion of spending in 2026. That's an average of analyst estimates, compiled by Bloomberg and reported by Fortune. The same report says Alphabet's free cash flow turned negative in the June quarter for the first time since its 2004 listing.

Sign 4: Is it a crowded trade?

Yes, according to the Bank of England. Its July 2026 report said the rally rested on a narrow set of AI companies. That raised concentration in some global indices. It also said hedge fund borrowing in equity markets had risen a lot.

Crowding doesn't tell you when a fall will come. It tells you how far a fall could spread if one does.

Sign 5: Are there first cracks?

A few, yes. Fortune, using Bloomberg, reported on the late July 2026 earnings. Alphabet fell more than 7% in a day after raising its spending plans. That was its worst day in over a year. An index of seven big tech shares fell 4.8%. Chip shares had gained 101% in the first half of 2026, then lost 17% in July.

Investors had started to punish heavy spending instead of rewarding it. But that same index was still only down 3.7% for the year at that point. A wobble isn't a pop.

What does this mean for AI tokens?

AI tokens are crypto coins tied to an AI story. They aren't shares in the big firms above. But they trade on the same mood. If trust in AI spending drops, the story behind AI tokens gets harder to sell.

Want the basics first? Read our guide to what AI tokens are. Most of them have no profits to compare with. So the "prices ahead of profit" test is hard to even run.

So is AI a bubble?

Here's the fair summary as at early September 2026:

  • For: stretched valuations, record spending, rising debt, a crowded trade, a first wobble.
  • Against: real products, real revenue, stronger earnings than in 2000, and an IMF view that overvaluation is far below the dot-com peak.

Both can be true. A technology can change the world and still be priced too high. The internet did exactly that. That's the honest answer, and it's not a tidy one.

Frequently Asked Questions

Is AI a bubble right now?

No one can say for sure. As at 4 September 2026, the Bank of England and the IMF both flagged the risk of a correction. Neither said a crash was certain.

How is AI different from the dot-com bubble?

Large, profitable firms pay for more of the spending. The IMF puts overvaluation at about half the dot-com level. But debt is rising, and that's the part to watch.

What happened to dot-com companies?

Many failed even though the internet was real. We went through 17 of them and checked what killed each one. Our post on what the dot-com bubble can teach us about NFTs covers the NFT side.

Can a bubble be good news?

Sometimes the technology survives and thrives afterwards. That doesn't protect the people who paid the peak price.

This is not financial advice. We don't say what to buy, hold or sell, and we make no price predictions.

About the Author

James Killick

AI Product, Web3 & Sales Automation Specialist

James Killick runs this site →

Digital Product Development

Devwiz is a Sydney-based digital product development agency specialising in launching and scaling SaaS products, web apps and mobile apps. James is a shareholder and advisor, shaping product strategy and PLG approach.

Visit Devwiz

// Related

Keep Reading