Is the AI bubble finally breaking, and how long can it last?

The cracks are not in the middle. They are at the edges, exactly where the money has to come in.

For a while the question was "is AI a bubble?" People argued about it at dinner. This week the tone changed. Fewer people are asking whether. More are asking for how long.

What actually happened this week

  • Hedge funds had one of their worst days since the pandemic-era market rout, as a fresh tech sell-off dragged on.
  • A big memory-chip maker shelved a planned US listing of one of its units, valued around 150 billion dollars, and blamed the chip-stock rout driven by AI spending.
  • A major AI cloud provider's shares fell another 10 percent in a day. Another infrastructure firm pulled its listing. A data centre builder working for OpenAI pushed its own listing to December or later.
  • OpenAI trimmed its valuation target to 1.25 trillion dollars after pitching around 1.4 trillion, while still raising up to 30 billion.
  • After an investor day, some Anthropic backers reportedly doubted that its growth justifies a valuation near 2 trillion, even with quarterly revenue up 45 percent to almost 17 billion dollars.

Now add the bill. The two biggest labs have each signed up for hundreds of billions in computing commitments, more than 850 billion for one and more than 750 billion for the other. Big tech is borrowing and issuing new kinds of shares to keep up with the spending. A chip giant is reportedly standing behind the smaller AI clouds with a safety net worth around 170 billion.

But hold on. Didn't OpenAI just raise money at a price about 50 percent higher than last time? How is that a bubble breaking?

Fair. That is exactly what the optimists say, and some commentators call this week a "bounce back", not a crack. A bubble does not pop the moment prices wobble. It pops when somebody who needs funding cannot get it, and the pattern to watch is the listings: first the weaker names pull back, then the stronger ones have to explain themselves to public investors.

So how long?

Nobody knows, and anyone who sounds sure is selling something. But the next test is already on the calendar. The two biggest private AI companies are heading for the stock market, and the first one is reportedly aiming to list before the end of November. If the public pays a trillion and more, the music keeps playing. If it balks, the edges we just described start looking like the middle.

That is a thesis, not a fact. But it is a thesis with a lot of this week's evidence behind it.