Hype Index: 'AI Is a Bubble'

TL;DR: "AI is a bubble" is the rare claim where both sides show up with receipts. The bubble case: circular financing measured in hundreds of billions, hyperscaler capex nearly doubling to ~$750B in guidance, and the two most famous crash-callers alive both positioned. The other side: revenue that dot-com never had, and valuations at roughly half the multiples of 2000. I read the top 20 published verdicts: 7 yes, 1 no, 12 hedge. We score it 50% noise / 50% signal, the most annoying score this series will ever give and the only honest one.

What is the claim?

Stated the way it circulates:

AI is a speculative bubble, and it is going to pop like the dot-com crash did.

Search interest in the question spiked 950% year over year in late 2025. (A wilder "1,581%" version of that stat circulates too; I could not trace it to any real source, which tells you something about the information diet on both sides of this debate.)

What do the bubble-callers have?

More than vibes. Ray Dalio wrote in January that "the AI boom that is now in the early stages of a bubble had a big effect on everything," separately putting the euphoria at about 80% of the run-up to 1929. Michael Burry disclosed puts on Nvidia and Palantir with over a billion dollars in notional value, then wrote the line the whole bear case fits inside: "I am not claiming Nvidia is Enron. It is clearly Cisco."

The structural evidence is the part that survives skepticism about famous bears. Hyperscaler capex ran roughly $410B in 2025 and current 2026 guidance sums to about $750B after the late-July raises, with Meta's free cash flow already compressing under it. The financing is going circular: Nvidia is in talks to support up to $600 billion of OpenAI-related financing, the chipmaker backstopping its own biggest customer, a structure Bloomberg now covers with a standing "AI Circular Deals" graphic. The Bank of England warned of a "sharp market correction" with index concentration the highest in half a century, and OpenAI's leaked audited 2025 loss of roughly $38.5B is the number Ed Zitron builds his "Lehman Brothers of the AI bubble" thesis on. And every price in the industry is ratcheting one direction because the current ones lose money.

What does the other side have?

The thing dot-com never did: revenue with a slope. OpenAI crossed $25 billion annualized by February 2026; Anthropic announced a $47 billion run rate in its own May funding release (a run rate, self-reported, but a number Pets.com never had a genre for). The companies funding the capex are the most profitable in history, Magnificent Seven net margins around 25.8%, roughly double the index. And the valuation math refuses to cooperate with 2000 nostalgia: Goldman's Peter Oppenheimer counts today's AI leaders at an average P/E near 25 versus 52 for the giants at the dot-com peak, Cisco peaked around 131x forward earnings versus Nvidia's ~30x, and the IMF notes AI investment is running under 0.4% of US GDP versus 1.2% during 1995-2000, funded by cash rather than debt. Jensen Huang's dismissal is exactly what he would say either way, but Jefferies' Brent Thill said the quiet bull part with a straight face: "The AI economy is healthy. The bear thesis is garbage."

What should you do instead?

Stop asking the binary and watch the three gauges that actually distinguish the outcomes. One: circular deal volume; vendor-financed demand is how real technology booms manufacture fake revenue, and it is the single cleanest 2000 rhyme on the board. Two: the gap between capex growth (~77%) and revenue growth at the companies spending it; that scissors closing is soft landing, widening is bubble. Three: whether the stated business results underneath hold up, because the failure-rate folklore and the layoff attributions show how much of the AI economy's narrative layer is already fiction in both directions, and how fast the fiction moves: AI-attributed job cuts fell 72% between May and July 2026. A bubble in financing structures can coexist with a real technology; 1999 proved the crash and the internet were both real.

Verdict

My read of the top 20 published answers, 7 yes, 1 no, 12 hedge, is not evidence of cowardice; it is evidence the record is genuinely split, and this series scores records, not moods. The capex slope, the circularity, and the concentration are real bubble machinery, assembled and running. The revenue, margins, and comparatively sane multiples are real non-bubble facts, also on the record.

Verdict: 50% noise / 50% signal. The only wrong positions are certainty in either direction, and both are selling something.