AI Bubble Warnings: What Dalio, Burry and JPMorgan Actually Said

Three famous warnings, three completely different arguments#

On 14 September 2026, Nasdaq 100 futures fell 1.72% before the American market opened. No earnings miss, no rate decision. The trigger was a set of comments about AI safety from Anthropic's Dario Amodei, echoed by Elon Musk and Sam Altman, and in premarket trading Intel, AMD and Marvell were each down by 5% to 6% (Yahoo Finance markets report).

Days like that have become routine, and so has the ritual that follows. Three names get pulled into the conversation: Ray Dalio, Michael Burry and JPMorgan. They are treated as one chorus shouting the same word.

They are not. One is measuring crowd psychology. One is disputing an accounting entry. One is a bank whose research desks disagree with each other in print. Only one of the three claims is easy for an ordinary investor to check.

What a bubble actually means, in plain terms#

A bubble is not simply "prices are high". Dalio has spent decades making the word measurable, and published his method: a six-part gauge asking how high prices are against traditional measures, whether prices assume conditions that cannot last, how many brand-new buyers have arrived, how bullish sentiment is, how much borrowed money is funding purchases, and whether buyers have locked in heavy forward commitments (Dalio's bubble indicator).

Two other terms matter here. Capital expenditure, or capex, is money spent on long-lived assets such as servers and chips. Companies do not charge that cost against profit all at once; they spread it over the asset's assumed useful life through depreciation. Stretch the assumed life from four years to six and this year's reported profit rises, without anything changing in the real world.

The other is notional value. When a fund discloses put options, which pay off if a share price falls, filings report the market value of the shares the contracts cover, not the premium actually paid. A "$900 million position" can cost a small fraction of that.

Dalio is measuring euphoria, and says that is not a sell signal#

Dalio's position has been consistent and widely misreported. In a CNBC interview on 20 November 2025 he said "there is definitely a bubble in markets", then added that the signs were present but it might not pop soon (Seeking Alpha). Reporting of those remarks put his estimate of the AI mania at about 80% of the euphoria seen before the 1929 crash and the 2000 dot-com peak, a paraphrase rather than a quoted figure (Benzinga). On 6 January 2026 he wrote that "the AI boom that is now in the early stages of a bubble had a big effect on everything" (Fortune).

The sharpest version came on 4 August 2026, on the Diary of a CEO podcast. Asked whether Jeremy Grantham was right to call this the biggest investment bubble in American history, Dalio said "He's right", then added: "Classic signs that we're in [one]." His framing was about paper gains. "Wealth is not the same as money," he said. "You see a lot of people getting wealthy but you can't spend the wealth." He named two things that usually burst bubbles, rising interest rates and a surge in new share issuance, pointing to the coming flotations of SpaceX, Anthropic and OpenAI (Fortune).

The part that rarely makes the headline: Dalio no longer runs Bridgewater's investments, and the firm's first-quarter 2026 filing went the other way, adding roughly 827,800 Nvidia shares, 670,000 Broadcom, 586,000 Micron and a new holding of about 1.077 million TSMC shares (Bridgewater Q1 2026 13F). Those are the firm's trades, not his personal book, but the gap is worth holding onto.

Burry's case is about accounting, not valuation#

Burry's quarrel is narrower than "AI stocks are expensive". He argues that the biggest AI spenders overstate profits by assuming their chips and servers keep earning for longer than they really do.

He put money behind it. Scion Asset Management's third-quarter 2025 filing, released on 13 November 2025, showed put options covering about 5 million Palantir shares, notional value $912 million, and about 1 million Nvidia shares at $187 million notional. No strike price, expiry or premium was disclosed, and Sherwood News later corrected its own framing to stress that the notional figure is not the size of the bet (Sherwood News). Days later he deregistered Scion as an investment adviser with the SEC (Bloomberg) and began publishing on Substack as Cassandra Unchained.

His framing was blunt. Understating depreciation by artificially extending asset lives, he wrote, is "one of the more common frauds of the modern era", and chips on a two to three year product cycle should not be getting longer assumed lives (reported by Yahoo Finance). In a post dated 9 July 2026 he set out the mechanism: depreciation is "the recovery of capitalized sunk cost over a defined window of earnings", making it "an economic lever, not a physical measurement". His evidence was the inconsistency between companies. Amazon shortened its server lives to five years in 2025, at a cost of $677 million. Meta extended its assumed server lives to 5.5 years and saved $2.9 billion (Cassandra Unchained).

JPMorgan analysts pushed back, arguing that corporations rather than consumers will carry most of the infrastructure cost through productivity gains, and that short positions "risk missing out on this momentum" (reported by AOL).

The market has not settled it cleanly. Nvidia was up 21.5% for 2026 as of 4 September, at $230.36. Palantir was down 2.8% for the year as of 17 September, at $176.24 (StatMuse: NVDA, PLTR). One leg roughly worked. The other did not.

JPMorgan is not one voice. It is at least three.#

Michael Cembalest, chairman of market and investment strategy for the asset and wealth management arm, published his 2026 outlook on 1 January under the title Smothering Heights. His central figure is concentration. In his own wording, "a broader group of 42 AI related companies has generated 65%-75% of S&P 500 earnings, profits and capital spending since ChatGPT's launch in November 2022". Strip those companies out and the index would have trailed Europe, Japan and China. The market value of four hyperscalers and the semiconductor ecosystem firms they depend on has gone "from $3 trillion to $18 trillion in just a few years", which summaries of the note put at roughly 20% of developed-market equities. Meta's capex plus research runs at 70% of revenue, against 10% for the median S&P 500 company. His question is deliberately left open: will all this investment eventually produce matching profits? He floats a possible "Metaverse moment", recalling the Magnificent Seven's 50% fall in 2022, while noting that tech valuations look less extreme once growth is factored into the PEG ratio (Eye on the Market Outlook 2026).

By 22 July 2026 he had turned cautious on American equity technicals, for a specific reason: "The worrisome part of a boom cycle is when companies closest to final demand roll over, even as capital spending beneficiaries of the cycle keep on thriving." Semiconductor stocks had gained about 90% over the prior year while Alphabet, Amazon, Meta, Microsoft and Oracle went sideways (Motley Fool).

JPMorgan technical strategist Jason Hunter made the same point with harder numbers in late August 2026. The Philadelphia Semiconductor Index was up 87% for the year and had just posted its best ever quarter, while Microsoft was down 18% year to date. The four biggest spenders, Meta, Microsoft, Amazon and Alphabet, were on track for $725 billion of AI infrastructure spending in 2026, up 77% on the $410 billion of 2025. Hunter's verdict: "The growing divergence that exists now and the outright negative hyperscalers price performance are reminiscent of the 1999-2000 dynamic" (TheStreet, 25 August 2026). In an earlier note he had flagged Microsoft's June fall as its worst month since 2000 (InvestingLive, 2 July 2026).

Then there is the bank's Global Research desk, which is bullish. In November 2025 it put global data centre and AI infrastructure investment at $5 trillion over five years, requiring roughly $650 billion of annual revenue to earn a 10% return, plus 122 gigawatts of new capacity between 2026 and 2030 and a $1.4 trillion funding gap (Techstrong.ai). By its midyear outlook on 25 June 2026 the desk had raised total capex to $5.5 trillion through 2030, with $4.1 trillion debt financed, and called the boom "not only durable, but increasingly profitable". The caveat sat in the last line: "AI-driven capital spending is scaling rapidly, and the economics are holding, for now" (Fortune).

Where the three claims overlap, and where they do not#

Who	The specific claim	What would confirm it	What would refute it	Where it stood in September 2026
Ray Dalio	AI equities show classic bubble signatures at roughly 80% of 1929 and 2000 extremes, but the pop needs a trigger	Rising long rates plus a wave of mega-IPOs absorbing cash	Rates fall, issuance stays modest, earnings catch up	Signals flashing; SpaceX, Anthropic and OpenAI flotations pending
Michael Burry	Hyperscalers inflate profits by assuming servers and GPUs last longer than they earn	Companies shorten useful lives or write down GPU fleets	Older GPUs keep earning across full five to seven year rental terms	Disputed. Amazon shortened lives, Meta lengthened them
Cembalest and Hunter, JPMorgan	Chip sellers rising while chip buyers fall is the 1999 pattern	Hyperscaler share prices and capex guidance both weaken	Buyers and sellers re-converge upward	Divergence persists: semis up 87%, Microsoft down about 18%
JPMorgan Global Research	The build-out is profitable and debt financed capex is serviceable	AI revenue approaches $650 billion a year	Revenue stalls while the $4.1 trillion debt load matures	Bullish, with an explicit "for now"

Two of these overlap more than they appear to. Burry's depreciation argument and Hunter's divergence signal are really about the same worry: the companies selling shovels book revenue now, while the companies buying them must eventually show a return. Cembalest's open question is that worry stated politely. Dalio sits on a different axis, because his gauge measures the crowd rather than the cash flows, and he is open that it offers no timing.

What none of them claims is that AI is worthless, or that a crash is scheduled. That reading belongs to headline writers.

The figures and quotations above are facts drawn from filings, published notes and interviews. How the three arguments relate to each other is my interpretation. None of this is investment advice, and nobody cited here claims to know when a bubble ends.

Key takeaways#

Dalio's warning is a measurement of crowd psychology, not a forecast. Reporting put his reading of the AI trade near 80% of historic bubble extremes, and he said in the same breath that it need not pop soon.
Burry's warning is an accounting dispute about depreciation schedules. It is testable in company filings and separate from any claim about valuation.
JPMorgan holds at least two opposing views at once. Cembalest and Hunter warn about a 1999-style divergence while Global Research calls the build-out durable and profitable.
Everyone agrees on the concentration numbers. Cembalest counts 42 AI-related companies behind 65% to 75% of S&P 500 earnings, profits and capital spending since late 2022, and four firms plan $725 billion of capex in 2026.
Scoreboards are humbling. Burry's Palantir leg was roughly flat in 2026 while his Nvidia leg was well offside, and Dalio's old firm was buying chip stocks while he warned about bubbles.

Frequently asked questions#

Did Michael Burry bet $1.1 billion against Nvidia and Palantir? No. The filing disclosed put options covering shares with a combined market value of about $1.1 billion. The premium he paid is not disclosed and would be far smaller (Sherwood News).

Why does depreciation matter so much for AI companies? Because the sums are enormous. When a company spends hundreds of billions on chips, the assumed useful life decides how much of that cost hits this year's profit. Meta's move to 5.5 year server lives saved $2.9 billion; Amazon's move the other way cost $677 million (Cassandra Unchained).

Has the AI bubble burst? Not on the evidence available in September 2026. Semiconductor indices were up sharply for the year while several hyperscalers were flat or down. That is a divergence, not a collapse (TheStreet).

What is the $650 billion figure people keep quoting? JPMorgan's estimate of the annual revenue AI infrastructure would need to generate to earn a 10% return on roughly $5 trillion of investment. It is a sizing exercise, not a prediction of failure (Techstrong.ai).

Does Ray Dalio still manage Bridgewater's money? No. He founded the firm but no longer runs its investment decisions, which is why its filings can show chip buying while he talks about bubbles (Bridgewater Q1 2026 13F).

Is the dot-com comparison fair? Partly. Cembalest and Hunter both invoke 1999, but the analogy is narrow: equipment suppliers thriving while the companies closest to end demand stall. Today's spenders also have far larger operating cash flows than the telecoms of 1999 did (Fortune).

How can a non-professional check any of this? Read the depreciation note in a company's annual report and look for changes in assumed useful lives. It is a single disclosed number, and it moves reported profit.

Glossary#

Bubble indicator. Dalio's six-part gauge of valuation, sustainability of assumed conditions, new buyers, sentiment, leverage and forward purchases.

Capex. Capital expenditure. Money spent on long-lived assets such as data centres, servers and chips.

Depreciation and useful life. The accounting method that spreads an asset's cost across the years it is expected to earn. The assumed number of years is the useful life, and management sets it.

Hyperscaler. A very large cloud operator. In practice Alphabet, Amazon, Meta, Microsoft and increasingly Oracle.

Put option. A contract giving the holder the right to sell shares at a set price. It gains value when the share price falls.

Notional value. The market value of the shares an options contract covers, which is not the amount paid for the contract.

13F filing. The quarterly disclosure of American equity holdings that larger institutional managers must file with the SEC.

PEG ratio. A price-to-earnings ratio divided by the expected earnings growth rate, used to compare valuations across companies growing at different speeds.

References#

Yahoo Finance, "Wall St falls as AI anxiety batters Nvidia, chipmakers", 14 September 2026: https://finance.yahoo.com/technology/ai/articles/ai-warnings-knock-nasdaq-futures-092329455.html
Ray Dalio, "The 6 Measures Of My Systemized Bubble Indicator", originally published on LinkedIn, February 2021: https://acquirersmultiple.com/2021/02/ray-dalio-the-6-measures-of-my-systemized-bubble-indicator/
Seeking Alpha, "Ray Dalio: Signs of a bubble are there, but it might not pop soon", November 2025: https://seekingalpha.com/news/4509546-ray-dalio-signs-of-a-bubble-are-there-but-it-might-not-pop-soon
Fortune, "Ray Dalio says AI is in the early stages of a bubble, so watch out for 2026", 6 January 2026: https://fortune.com/2026/01/06/ray-dalio-says-ai-is-in-the-early-stages-of-a-bubble-so-watch-out-for-2026/
Fortune, "Ray Dalio on the AI bubble nearing 1929, 2000 levels", reporting his 4 August 2026 Diary of a CEO interview: https://fortune.com/2026/08/04/ray-dalio-ai-bubble-1929-2000-ipos-wealth-is-not-money/
Bridgewater Associates Q1 2026 Form 13F holdings summary: https://www.kucoin.com/blog/bridgewater-13f-q1-2026-ray-dalio-s-firm-boosts-chip-stocks-nvidia-broadcom-micron-tsmc-and-exits-software-names-like-salesforce
Sherwood News, on Scion Asset Management's Q3 2025 Form 13F, released 13 November 2025: https://sherwood.news/markets/michael-burry-big-short-discloses-1-1-billion-options-bet-against-nvidia-palantir-puts/
Bloomberg, "Michael Burry's Scion Fund Is Deregistered, SEC Filing Shows", 13 November 2025: https://www.bloomberg.com/news/articles/2025-11-13/michael-burry-s-scion-hedge-fund-deregistered-sec-filing-shows
Michael Burry, Cassandra Unchained, "Short Thoughts: NVDA, Neos, Hyperscalers, Jevons Paradox, and Compression", 9 July 2026: https://michaeljburry.substack.com/p/short-thoughts-july-8-2026-nvda-neos
Scion Asset Management 13F filing history, SEC Form 13F records: https://13f.info/manager/0001649339-scion-asset-management-llc
Yahoo Finance, reporting Michael Burry's Cassandra Unchained claims on hyperscaler depreciation: https://finance.yahoo.com/news/michael-burry-just-exposed-big-162617167.html
Michael Cembalest, J.P. Morgan Private Bank, Eye on the Market Outlook 2026: Smothering Heights, 1 January 2026: https://privatebank.jpmorgan.com/content/dam/jpm-pb-aem/global/en/documents/eotm/smothering-heights.pdf
The Motley Fool, on Michael Cembalest's Eye on the Market note of 22 July 2026: https://www.fool.com/investing/2026/08/03/top-jpmorgan-strategist-michael-cembalest-cautious/
TheStreet, "JPMorgan sends another strong message to stock market investors", on Jason Hunter's note, 25 August 2026: https://www.thestreet.com/investing/stocks/jpmorgan-warns-investors-about-artificial-intelligence-ai-stocks
TheStreet, "JPMorgan sends stark warning on AI stocks, cites dotcom worries": https://www.thestreet.com/investing/jpmorgan-ai-stocks-dotcom-bubble-warning
InvestingLive, on Jason Hunter's earlier note, 2 July 2026: https://investinglive.com/stock-market-update/jpmorgan-sees-dot-com-era-warning-as-ai-hardware-stocks-diverge-from-spenders-20260702/
Benzinga, on Ray Dalio's about-80% comparison to the 1929 mania, January 2026: https://www.benzinga.com/news/topics/26/01/50001526/ray-dalio-flags-ai-as-early-stages-of-a-bubble-comparing-todays-optimism-to-about-80-of-1929-mania
CNBC, "Michael Burry of 'Big Short' fame deregisters Scion Asset Management", 13 November 2025: https://www.cnbc.com/2025/11/13/michael-burry-of-big-short-fame-deregisters-scion-asset-management.html
J.P. Morgan Asset Management, Eye on the Market Outlook 2026 landing page: https://am.jpmorgan.com/us/en/asset-management/institutional/insights/market-insights/eye-on-the-market/outlook-2026/
Techstrong.ai, on J.P. Morgan's November 2025 AI infrastructure analysis: https://techstrong.ai/agentic-ai/ai-infrastructure-demands-650-billion-in-annual-revenue-for-viable-returns-jpmorgan-analysis/
Fortune, on J.P. Morgan Global Research's Midyear Outlook, 25 June 2026: https://fortune.com/2026/06/25/what-bubble-jpmorgan-5-5-trillion-ai-capex-explosion-profitable-for-now/
AOL, reporting JPMorgan's response to Michael Burry's depreciation claims: https://www.aol.com/articles/jpmorgan-says-michael-burry-dead-161646229.html
J.P. Morgan Global Research, 2026 Market Outlook: https://www.jpmorgan.com/insights/global-research/outlook/market-outlook
StatMuse, NVDA 2026 year-to-date return, as of 4 September 2026: https://www.statmuse.com/money/ask/nvda-stock-performance-returns-for-2026
StatMuse, PLTR 2026 year-to-date return, as of 17 September 2026: https://www.statmuse.com/money/ask/pltr-stock-performance-returns-for-2026