Is the AI bubble bursting? What the buildout data says

Not by the measure that matters most: spending. The seven largest AI buyers spent $657 billion on capex over their last four reported quarters, and none cut its guidance this summer. What has broken is sentiment. Supercycle's AI Fear & Greed Index has read Fear or Extreme Fear for six straight weeks, led by power, grid and data center stocks.

That is as of October 8, 2026, when the index closed at 33 (Fear). It has not been above 44 in its last 30 readings. The part of the story that usually gets skipped is where the fear is. The layers that receive the AI money are the most afraid, while the software layers are not afraid at all.

What does "the AI bubble bursting" actually mean?

Two different things get called the AI bubble. One is stock prices running ahead of the earnings that would justify them. The other is spending running ahead of the revenue it will produce. They can break at different times, and a fall in prices does not mean the spending has stopped.

This post tests both. Spending comes from the Buildout Tracker, which records capex guidance, data center projects and power deals with a source on every row. Prices come from the AI Fear & Greed Index, which reads sentiment across the tracked AI universe every night. The capex cycle entry explains why the two can drift apart.

Is AI spending slowing down?

Not in the latest reports. These are the seven buyers the tracker follows, with their most recent quarter as of October 4, 2026.

Company	Latest quarter	Capex, latest quarter	Capex, last four quarters	Change vs a year earlier	Latest guidance
Amazon	Q2 2026	$53.1B	$173.0B	+68%	Raised: about $220B for 2026
Microsoft	FY26 Q4	$41.0B	$145.3B	+69%	Over $50B for FY27 Q1, full year to grow
Alphabet	Q2 2026	$44.9B	$132.4B	+100%	Raised: $195B to $205B for 2026
Meta	Q2 2026	$31.1B	$92.4B	+83%	Floor raised: $130B to $145B for 2026
Oracle	FY27 Q1	$28.5B	$75.7B	+235%	Reiterated: up to about $95B gross for FY27
CoreWeave	Q2 2026	$9.4B	$27.2B	+224%	Raised: $35B to $39B for 2026
Nebius	Q2 2026	$5.7B	$11.1B	+1008%	Reaffirmed: $20B to $25B for 2026

Together that is $213.7 billion in the latest quarter and $657.1 billion over four quarters. Four of the seven raised their guidance in the July to September reporting round, and the other three held or extended it. None cut.

Where the data stops: these are the second-quarter reports. The third-quarter round lands in the last days of October, and that is the first chance for any of the seven to change course. Definitions also differ by company, for example Meta's guidance includes finance lease payments. The September capex report has the notes for each row.

Is the buildout still adding projects and power?

Yes. The tracker holds 104 data center projects. 54 are under construction, 29 are announced, 16 are partly running and 5 are fully operational. The 91 that disclose their size add up to 104.5 gigawatts of planned capacity.

It also holds 52 power deals between AI buyers and power suppliers, 44 signed and 8 at the letter-of-intent stage. The 47 that disclose capacity total 70.3 gigawatts. 14 of the 52 are dated in the third quarter of 2026, more than any earlier quarter in the dataset. Treat that count as a floor rather than a trend, because the tracker's coverage of older deals is thinner.

Then why are AI stocks in Fear?

The index's five components on October 8 explain the reading:

• Momentum scores 10. The AI composite is flat against its 125-day average.
• Breadth scores 23. Only 41% of tracked stocks are above their own 50-day average.
• 52-week position scores 4. More stocks sit near their 52-week lows than near their highs.
• Volatility scores 65. The composite is moving about 18% annualized, calmer than usual for this group.
• Exposure tilt scores 62. Pure-play AI names are beating peripheral ones by about 4% over 20 days.

The selling is broad and slow. It is not a collapse in the purest AI names, which are still doing better than the rest of the universe.

Which parts of the AI trade are most afraid?

This is the part the bubble debate leaves out. The index scores each buildout layer separately, and on October 8 the spread was wide.

Layer	Score	Band
Utilities & Grid	11	Extreme Fear
Robotics & Automation	13	Extreme Fear
Power & Energy	22	Extreme Fear
Raw Materials & Mining	22	Extreme Fear
Neoclouds	24	Extreme Fear
Networking & Optics	25	Fear
Data Centers & Hardware	27	Fear
Semiconductors	38	Fear
AI Applications	57	Greed
Cybersecurity AI	63	Greed

Power, grid, data centers and neoclouds get paid when the buyers above spend. They are the most fearful layers in the index. AI software and applications, which spend less on the buildout and sell to its users, are in Greed. Stock prices often move before company guidance does, so one reading of the table is an early bet that the spending will slow. The other is that investors are selling these layers for reasons that have little to do with AI demand, such as interest rates or the politics of electricity prices. The index measures price behavior, not causes, so it cannot tell the two apart.

Is this like the dot-com bubble?

One similarity shows up in the data. A small number of buyers carries the buildout. Google is the buyer on 13 of the 52 tracked power deals and Meta on 10. If one or two of the large buyers slowed, their suppliers would feel it quickly.

The comparison also has limits the data cannot settle. How the 2026 spending is funded, and how that compares with the late 1990s, needs each company's cash flow statement, which this post does not cover. Whether the spending pays off depends on AI revenue over the next several years, which no capex table can show.

What would change the read?

Three things would move this from a sentiment problem to a spending problem:

1. A cut to capex guidance from any of the seven buyers in the late-October reports.
2. Tracked projects slipping from under construction to delayed or cancelled on the Buildout Tracker.
3. The index's exposure tilt turning negative, meaning pure-play AI names start to lag the rest of the universe.

The first of those can arrive within three weeks. This post will be updated with the third-quarter capex figures when they land.