Is the AI Bubble Bursting? Chip Stocks Down 38%, AI Spending Is Not

Is the AI bubble bursting? The most crowded part of it may be: the PHLX Semiconductor Index closed 38.4% below its June record on November 5, 2026, after falling 12.3% in a single session. A burst in the 2000 sense also needs AI buyers to cut spending and customers to walk away, and no major buyer has announced that. So far this is a severe repricing of chip and data center stocks, with financing the part of the story that is deteriorating fastest.

That answer may not last. In 2000 the Nasdaq also had sharp drops and strong rebounds before the decline became clear, as our dot-com bubble timeline shows. What follows sets out six tests that would separate a deep correction from a burst, with the latest number for each. For the day itself, see our report on Thursday's crash in AI stocks.

Prices: a bear market in chips, a correction everywhere else

The damage is uneven, and that is the main reason the word "burst" is still premature. The chip index, Micron and Nvidia have fallen far more than the broad market. The Nasdaq Composite is down 11.9% from its October 6 record, and the S&P 500 only 5.9%. The Nasdaq's loss from March 2000 to October 2002 was 77.9%.

Even after Thursday, most of this year's gains are still in place. The chip index is up 27.2% for 2026, the S&P 500 7.4% and Micron 130%. Nvidia is the exception, down 2.6%. That matters in two ways. It shows how far the trade ran, and it means many holders still have profits to sell. Nomura's Charlie McElligott warned that the unwind is approaching round-trip levels in some trades, Yahoo Finance reported, meaning that some gains are close to being given back in full.

Capex: nobody has cut, but the 2027 number is now in question

The clearest sign of a real burst would be the biggest buyers cutting orders. No major buyer has announced that. Microsoft's shortfall last quarter came from late deliveries, not a decision to spend less, and it guided December-quarter capex above $50 billion, as we covered in our Microsoft capex report. The largest AI spenders outlined as much as $828 billion for 2026, according to Fortune, as set out in our AI capex analysis.

What has changed is confidence in the forecast. Peter Boockvar of One Point BFG Wealth Partners told CNBC that investors now doubt the $1.2 trillion of capex expected next year. Dan Ives of Yorkville Ives contends peak AI capex could arrive in 2027 rather than later in the decade, FXStreet reported. A peak in 2027 is not a collapse in 2026. Stocks trade on the future, though, and a chip supplier valued on years of growth is exposed to any shortening of that runway.

Demand: the numbers from the last two weeks are still strong

Reported results give no sign of customers pulling back. Azure revenue grew 48% last quarter, and Microsoft's commercial backlog reached $677 billion. AMD reported record revenue of $13.4 billion and guided to about $17 billion for the current quarter, per our AMD earnings report. IREN, which reported on Thursday after the close, said contracted prices rose 11% on three-year deals and 12% on five-year deals, and that backlog rose 22% to $20.3 billion.

There are two caveats. This demand is concentrated in a small group of very large buyers, some of them funding each other, as our circular deals analysis describes. And backlog is a promise, not cash. Revenue figures also look backward.

Financing: the test that is moving the wrong way

If one line of the table should worry a reader, it is this one. AI has moved from being paid for out of cash flow to being paid for with outside money, and that money is getting more expensive. In the past week alone, Oracle raised $20 billion in stock and bonds with coupons of up to 8.65%, and S&P revised Meta's credit outlook to negative after its free cash flow turned negative. The Financial Times reported that OpenAI cut its funding round target to about $20 billion.

Rates are not helping. CME FedWatch put the odds of a Fed hike in December at 70.1% at midday Thursday, CNBC reported, and the 10-year Treasury yield was above 5.30% last month before easing to 4.83%. IREN's release shows the dependence: of $15 billion of planned capex left for its fiscal year, $6 billion is target funding that remains subject to market conditions. A day like Thursday makes those conditions worse.

Limits and holders: two new pressures

Voters may have just added a new constraint. Texas holds 27% of planned US data center gigawatts by Wells Fargo's estimate, and its incoming governor campaigned for a pause on new projects, as we reported in our midterms coverage. Construction is slipping too: IREN moved its $4 billion annual run-rate target from year-end to January, and said it may extend to the end of February, citing a customer-requested design change and supplier timing. OpenAI's pledge on Thursday to pay full power costs shows how seriously the industry takes the backlash.

Limits do not simply destroy value, either. Rob Sechan of Newedge Capital Group told CNBC that if new capacity gets harder to build, existing capacity becomes more valuable. Companies that already own working sites could benefit even as forecasts for chip sales fall.

Holders are starting to crack, though only just. Vanda Research counted the largest single-day net retail selling since March 2020, $358 million by mid-afternoon, Yahoo Finance reported, and its strategist Viraj Patel called it capitulation. History offers little comfort on timing. Yahoo Finance found Nvidia had five or more straight down weeks 12 times before this one since its 1999 IPO. It was higher 13 weeks later in only five of those, and higher a year later in eight.

What the AI Bubble Index says about a burst

The AI Bubble Index read 63.3 on November 3, in the Normal band, and that reading predates this week's post-election sessions. Between the Nasdaq's October 6 record and November 3, momentum had already fallen from 82.1 to 51.5 and valuation and concentration from 70.4 to 58.6. Sentiment and attention barely moved, and systemic risk rose. So prices cooled well before people stopped talking about AI. Fragility, at 72.2, sits above Euphoria at 62.8, and the long-run core, which compares today with every day since 1990, is 80.1.

The index describes conditions and does not forecast a date. See the methodology for how each family is built, and the dashboard for daily readings. For the concepts, our guide to what the AI bubble is sets out the warning signs.