Who Is Warning About an AI Bubble? The Insiders Saying So

 

By Daniel Rao

Here is the strangest chorus in modern markets: the men building the artificial intelligence boom are also the ones warning you about it. The chief executive of OpenAI says investors are overexcited. The chief executive of Google says no company would escape if the bubble bursts, including his own. The International Monetary Fund has put an AI correction on its list of threats to the entire global financial system. And all of them said it while their companies kept spending — hundreds of billions of dollars, at a pace no previous technology boom ever matched.

That contradiction is the story. This is not a warning from the sidelines. The people closest to the money, with the best view of the order books and the data centres, are telling us the excitement has run ahead of reality — and then going back to work on Monday morning to keep building. Either they are wrong, and the spending is justified, or they are right, and we are watching a bubble inflate in real time, narrated by the people inflating it.

So who, exactly, is warning about an AI bubble? What do they actually fear? And who is standing on the other side of the argument, insisting this time is different? The answers matter more than the usual market noise, because the sums involved are now large enough to move the whole economy — and, as the IMF spelled out this month, to drag it down if the bet goes wrong.

Sam Altman: "overexcited about a kernel of truth"

Start with the most unlikely Cassandra of all. In August 2025, OpenAI chief executive Sam Altman sat down to dinner with a group of tech journalists and said the quiet part out loud. "When bubbles happen, smart people get overexcited about a kernel of truth," he said, as reported by The Verge. "Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes. Is AI the most important thing to happen in a very long time? My opinion is also yes."

It was a remarkable pair of sentences, because both halves were true at once. Altman was not disowning his life's work. He was saying the technology is real and the prices are not — or at least, not all of them. He went on to warn that some investors were "likely to get very burnt here", and described startup valuations as "insane", with companies raising hundreds of millions of dollars on little more than an idea. The numbers backed him up: PitchBook counted $104.3 billion raised by US AI startups in the first half of 2025 alone, nearly two-thirds of all American venture funding. His own company had raised $40 billion that March at a $300 billion valuation, the largest private tech round ever.

Photo: Sam Altman in 2009, when he ran the startup Loopt — long before ChatGPT made him the face of the AI boom. Credit: Phil Glockner, CC BY 2.0, via Wikimedia Commons.

Altman's warning carried a particular sting because of who he is. This is not a short-seller talking his book, or a rival executive sniping at a competitor. It is the man whose product ignited the boom, telling the market that the market has lost its head. And note what he did next: nothing. OpenAI kept raising, kept building, kept spending. The warning and the spending coexisted, comfortably, in the same company — a pattern you will see repeated with almost every name in this article.

Sundar Pichai: "no company is going to be immune, including us"

Three months later, the warning came from the other end of the industry's power structure. In an interview with the BBC published in November 2025, Alphabet and Google chief executive Sundar Pichai — a man not known for loose talk — acknowledged "elements of irrationality" in the AI investment boom. Speaking from Google's California headquarters, he compared the moment to the early internet: there had been "clearly a lot of excess investment" then, but the internet itself was "profound". "I expect AI to be the same," he said. "So I think it's both rational and there are elements of irrationality through a moment like this."

Then the BBC asked him the direct question: what happens to Google if the bubble bursts? Pichai did not hide behind corporate reassurance. "I think no company is going to be immune, including us," he said. He added that Google might weather the storm better than most, because it owns the full stack — its own chips, its own data in YouTube, its own models — but weather it is all he claimed. Not escape it. At the time, Alphabet shares had surged about 46 percent in 2025, and the company had just pledged £5 billion over two years to UK AI infrastructure. Pichai was, in effect, warning about a fire in a house he was still furnishing.

Photo: Sundar Pichai, chief executive of Alphabet and Google, in 2023. Credit: European Commission, CC BY 4.0, via Wikimedia Commons.

Pichai's warning also widened the frame. He spoke of the "immense" energy demands of AI — 1.5 percent of global electricity consumption, according to the International Energy Agency, with computing demand potentially reaching 200 gigawatts by 2030 — and conceded that Alphabet's climate targets would slip as it scaled up computing. The bubble debate, in his telling, is not just about share prices. It is about power grids, climate pledges, and jobs: he warned of "societal disruptions" and said people would have to adapt. Those who do, he insisted, "will do better". It was the closest thing to a shrug a CEO can give about a storm he has just forecast.

The IMF: an AI correction is now a systemic risk

If Altman and Pichai were the industry warning about itself, October 2026 brought the institutions. On 13 October, at the IMF–World Bank annual meetings in Bangkok, the International Monetary Fund published its Global Financial Stability Report and named an AI-related market correction as one of three "plausible near-term catalysts for a sharp tightening of financial conditions and market turmoil". The other two were a spike in sovereign bond yields and a major cyberattack. That is the company an AI sell-off now keeps in the mind of the world's financial watchdog.

"Financial stability risks remain elevated," the Fund said. Its deputy director for monetary and capital markets, Athanasios Vamvakidis, put the point with unusual bluntness at the Bangkok briefing: "resilience is not necessarily safety." Markets look strong, he argued, but that strength rests heavily on one technology. "An AI market correction will affect the real economy," he said, through wealth effects, investment pullbacks, and spillovers into broader markets.

What alarmed the IMF was not just prices but plumbing. Hyperscalers — the giant cloud and data-centre operators — now account for 10 percent of global investment-grade bond issuance, up from less than 1 percent in 2024. Hedge funds' gross equity positions and equity margin loans have risen from 10 percent to about 15 percent of advanced economies' GDP since 2024. And single stocks now make up 28 percent of leveraged exchange-traded funds, up from 16 percent — a recipe, the Fund warned, for forced selling if AI investment or profitability disappoints.

Then came the report's most striking piece of detective work. Examining 14 public transactions, the IMF found a circular structure at the heart of the boom: "$300 billion in equity" and "$650 billion in debt, including in off-balance sheet vehicles", handed over "in exchange for $1.4 trillion of future revenue commitments". Chipmakers and hyperscalers invest in AI labs; the labs commit to buying chips and computing power for years ahead; the commitments justify more investment. It works beautifully — until someone asks whether the revenue will actually arrive. Jason Wu, an assistant director in the IMF's Global Markets Analysis Division, spelled out the "triple whammy" at the Bangkok press conference: "If there are questions about the monetization of AI, the equity investments will deplete in value, the debt may be difficult to service, and the revenue commitments in the future may not materialize."

The IMF's separate October outlook added the number that travelled furthest. In a downside scenario — a steep drop in AI investment triggered by disappointment over productivity gains or profitability — global GDP would end up 1.1 percent below baseline by 2028, with US output 1.8 percent lower. That is a scenario, not a forecast, and the Fund was careful to say so. But scenarios are how institutions tell you what keeps them awake. The Fund's managing director, Kristalina Georgieva, gave the image its shape: "We now have an economy that is standing on one leg, the AI leg." Her follow-up question hung in the air in Bangkok: "What if it wobbles?"

The BIS: the central bankers saw it first

The IMF was not even the first institution to sound the alarm. Four months earlier, in June 2026, the Bank for International Settlements — the central bankers' own bank — had published its Annual Economic Report with the AI boom listed as one of four "pressure points" facing the global economy. Its opening chapter, "Progress and Peril", noted that the five largest hyperscalers were "set to spend over a trillion US dollars on AI-related capital expenditure from 2025 through 2026", and that those "commitments are outpacing earnings and the free cash flow of these firms", pushing some of them into debt.

The BIS worried less about the spending than about its structure. It described circular financing deals — chipmakers and hyperscalers taking equity stakes in AI labs or "neocloud" providers, which then sign multi-year purchase commitments — with terms "typically poorly disclosed", raising the risk that the same asset could be pledged more than once. Direct-lending funds, it noted, had roughly quadrupled their AI and IT lending in five years, to about 15 percent of their portfolios. "Disappointment in returns could trigger a sudden pullback in financing," the report warned.

And the BIS reached for history. It compared the AI surge to the canal mania of the 1830s, the railway mania of the 1840s, the electrification craze of the 1920s, and the dot-com bubble of the late 1990s. Every one of those episodes, it argued, began with a genuine breakthrough and ended with capital piled far beyond what commercial returns could justify. For the full comparison, see our earlier piece on how the AI boom stacks up against the dot-com crash. The BIS's point was not that AI is a fad. It was that genuine revolutions are precisely the ones that produce the biggest bubbles, because the story is good enough to suspend disbelief.

Ray Dalio and Ken Griffin: the investors pile in

By October 2026, the warning had migrated from executives and institutions to the investor class itself — and it arrived with a timetable. On 7 October, speaking at the Forbes Global CEO Conference in Singapore, Bridgewater founder Ray Dalio called AI a "classic bubble" and said the market was approaching the point where it bursts. "We're in the part of the cycle that is before that but approaching that," he said. "I think we're close to that."

Dalio's mechanism was specific, and it is worth understanding because it is the one the IMF's numbers support. AI requires enormous capital, and increasingly that capital comes from debt. Higher interest rates raise the cost of carrying that debt. And bubbles, in Dalio's telling, usually pop not when people change their minds but when they are forced to turn paper wealth into cash — to sell, because they need the money. With the S&P 500 and Nasdaq 100 at records that week and bond yields at multi-decade highs, he saw the two trends moving toward each other. His argument, notably, was not that AI is useless. Railroads, cars, and the internet all changed the world, he pointed out, without protecting every investor who financed them.

Thirteen days later, Citadel founder Ken Griffin said it even more plainly. "There's almost certainly bubbles within this moment in time," he told Bloomberg on the sidelines of Citadel's Future of Global Markets Conference in New York on 20 October. "Any radical transformation of our economy is generally intertwined with a bubble." Griffin, who had been dismissive of AI at Davos in January, had changed his mind about the technology — Citadel's own AI tools had cut some high-level research work from six to eight weeks to two or three hours — but not about the prices. His one regret about the last bubble was revealing, and human: looking back at the dot-com era, he said, "I just wish I'd kept my Amazon stock in that moment." The survivors, he argued, "are the market leaders for decades to come" — which is why the question is not whether AI matters, but which companies survive the repricing. We examined that exact question — who survives when the AI bubble pops — in an earlier piece in this series.

The other side: "asking if AI is a bubble is absurd"

Not everyone is warning. The loudest dissenter is SoftBank chief executive Masayoshi Son, and his argument is gloriously, almost comically, large. Speaking at SoftBank's annual corporate conference in Tokyo on 14 July 2026, Son said developing AI "will require investment of $5 trillion each year by 2040". "Every year $5 trillion, or 800 trillion yen, you might think that's a lie, but I am confident that's what it will cost," he said. His logic: if AI revenue reaches 20 percent of global GDP by 2040, then $5 trillion a year is "a rounding error". "Asking if AI is a bubble is absurd," he added. "I don't think people who ask that question know what AI is about." Reuters, reporting his remarks, drily noted that he did not explain how he arrived at either number. Son's record cuts both ways: an early bet on Alibaba made his fortune, while WeWork — the shared-office company he backed to the hilt — went bankrupt. His cumulative investment in OpenAI alone is set to exceed $60 billion before 2026 ends. He is not a neutral observer. He is the boom's biggest cheerleader, and he has bet accordingly.

A subtler defence arrived in late October 2026 from an unexpected quarter: a machine. Yardeni Research unveiled an "AI Bubble Meter" built with Anthropic's Claude model, tracking four bubble conditions (valuations, household equity exposure, market returns, private investment) against two potential triggers (credit growth and Treasury yields). Its first reading was a split verdict: conditions scored "High" at 83 out of 100, but triggers scored "Low" at 34. "The meter says there's a bubble, but it's not about to burst," the firm concluded. The history behind that split is genuinely reassuring for bulls: in 54 quarters since 1965 when conditions were high but triggers were low, the S&P 500 entered a bear market within two years only 9 percent of the time, with a median inflation-adjusted gain of 18.3 percent. When both dials were high, the bear-market frequency jumped to 42 percent, with a median loss of 12.5 percent. In other words, recognising a bubble is not the same as predicting its end. The danger, on Yardeni's own numbers, arrives when the triggers dial starts climbing — which is exactly what Dalio says is now happening.

Then there is Lombard Odier's global chief investment officer, Michael Strobaek, who in October 2026 told reporters that investors should not be "too afraid" of an imminent burst. "We are not worried, right here and right now, that we are in a thoroughly imminent bursting bubble," he said, adding that it was "way too early to call a bubble" and that investors who bet too heavily against the boom "can get very badly burned". Writing in the NZZ am Sonntag in September, Strobaek went further: "We are only at the beginning of a long evolution in AI. I don't see a bubble." His case rests on adoption: AI is being deployed across companies far faster and more broadly than the internet was, and — paradoxically — the fact that everyone is now watching for a bubble may itself prevent the worst excesses. It is a fair point, and a familiar one: the four most dangerous words in finance, as the old joke goes, are "this time is different". Strobaek used them anyway, knowingly.

What the warners actually agree on

Strip away the disagreements, and a surprising consensus emerges. Almost nobody in this debate — not Altman, not Pichai, not Dalio, not even Griffin — says AI itself is a mirage. They agree the technology is real, transformative, and probably underrated in the long run. What they disagree about is narrower and more consequential: whether the prices being paid today, and the debt being taken on to pay them, can survive contact with actual revenues. That is a very different argument from "AI is a fad", and it is why the warnings are hard to dismiss. The people issuing them are not skeptics of the technology. They are insiders pricing its financing.

There is a second consensus, quieter but just as important: nobody knows the timing. Altman warned in August 2025; fourteen months later, markets were higher, spending was larger, and the warnings were louder. Yardeni's meter formalises the point — conditions can stay extreme for years while triggers stay quiet. Bubbles do not burst on a schedule. They burst when something forces selling: a rates shock, an earnings miss, a financing freeze, a revelation that the revenue commitments were softer than advertised. Until then, the rational move for every executive in this story is to keep building, because stopping first means losing the race. That is the trap at the heart of the whole debate, and it is why the warnings and the spending will keep coexisting.

We checked each of the four earlier TechPulse bubble posts cited in this article — all four returned live pages — and cross-read the IMF's October briefing text against three independent reports of it before writing this piece. The pattern held up in every source: the warnings are real, the numbers are large, and the people closest to the boom are the ones most nervous about its financing. If you want the full case for and against, start with our pillar piece, Is AI in a bubble? The honest case for and against, and then read the enterprise side of the story — enterprises bought the AI dream, but the returns never arrived — because that gap between spending and payoff is where any bubble, if it exists, will actually be decided.

Our judgment: believe the warnings, ignore the timetable

Here is where we land, after weighing all of it. The warners are right about the structure and wrong to be treated as a forecast. The financing of this boom — debt-heavy, circular, poorly disclosed in places, and now large enough for the IMF to call it a systemic risk — is genuinely fragile in a way the dot-com boom's equity-funded excesses were not. That is a colossally important difference, and it is the one the optimists keep gliding past. A bubble funded by equity hurts shareholders; a bubble funded by debt, margin loans, and leveraged ETFs can hurt everyone else too. On the other hand, the history Yardeni's meter captures is real as well: expensive markets with quiet triggers have a habit of getting more expensive, and the people who shorted them early got carried out. So believe Altman and the IMF about the risk. Do not believe anyone — Dalio included — who tells you they know when it ends. The honest position is the uncomfortable one: this looks like a bubble, it may not burst soon, and when it does, the damage will depend less on whether you believed in AI than on how much borrowed money was standing behind that belief.

Frequently asked questions

Who is warning about an AI bubble?

OpenAI CEO Sam Altman ("overexcited", August 2025), Alphabet CEO Sundar Pichai ("elements of irrationality", BBC, November 2025), Bridgewater founder Ray Dalio ("classic bubble", October 2026), Citadel CEO Ken Griffin ("almost certainly bubbles", October 2026), and the institutions: the IMF (October 2026) and the BIS (June 2026) both flagged AI investment as a financial-stability risk.

Will the AI bubble burst?

Nobody credible will give you a date. Yardeni Research's AI Bubble Meter scored bubble conditions at 83 out of 100 but burst triggers at only 34 in late October 2026, concluding "there's a bubble, but it's not about to burst". Historically, that combination preceded a bear market within two years only 9 percent of the time. Ray Dalio disagrees, saying the market is "close to" the breaking point as rising rates meet debt-funded spending.

What happens if the AI bubble pops?

The IMF's October 2026 downside scenario puts global GDP 1.1 percent below baseline by 2028, with US output 1.8 percent lower, if AI investment drops sharply on disappointing productivity or profits. The damage would spread through falling share prices, harder-to-service debt, and revenue commitments that fail to materialise — the "triple whammy" the IMF described. Companies with real revenues and strong balance sheets would likely survive; debt-dependent builders and speculative startups would not.

Is AI a bubble, or is it different this time?

Both arguments have force. The bubble case rests on prices, debt, and circular financing deals. The "different this time" case rests on adoption speed and real earnings: unlike 1999, today's biggest AI spenders are profitable companies, and AI is being deployed across industries far faster than the internet was. Lombard Odier's Michael Strobaek argues that breadth of adoption — and widespread awareness of bubble risk — may itself prevent the worst excesses.

Who says AI is NOT a bubble?

SoftBank CEO Masayoshi Son is the most prominent dissenter, calling bubble talk "absurd" and arguing AI will need $5 trillion a year by 2040 — a "rounding error" if AI revenue reaches 20 percent of global GDP. Lombard Odier's Strobaek says "I don't see a bubble", and Yardeni's meter says conditions are bubbly but triggers are not yet in place. The common thread among dissenters: the technology's long-run payoff justifies today's spending.

What do the warners say would actually trigger a pop?

Not a single event: Dalio's mechanism is debt meeting rising rates plus forced selling when paper wealth must turn into cash; the IMF's is a correction in AI-linked asset prices spilling into the wider financial system; Pichai's is that no company escapes, including Google — the trigger differs by warner, which is why the post refuses a single date.

Do the warners agree on timing?

No. Dalio says the market is "approaching" the breaking point; Altman calls investors "overexcited" without a timetable; the BIS and IMF frame it as a systemic risk to manage, not a dated prediction — and the post's judgment is explicit: believe the warnings, ignore the timetable.

Sources: The Verge (Altman, August 2025); BBC via Reuters, The Hindu, and PYMNTS (Pichai, November 2025); IMF Global Financial Stability Report and press-briefing transcript, Bangkok, 13 October 2026; IMF World Economic Outlook downside scenario, October 2026; BIS Annual Economic Report 2026, June 2026; Bloomberg via WealthAdvisor, Stocktwits, and Tekedia (Dalio, 7 October 2026); Bloomberg via CEOWORLD and TradingKey (Griffin, 20 October 2026); Reuters (Son, 14 July 2026); Yardeni Research via Stocktwits, MENAFN, and The WealthAdvisor (Bubble Meter, October 2026); The Edge Singapore and Lombard Odier (Strobaek, September–October 2026).

By Daniel Rao, TechPulse USA. Daniel covers AI, markets, and the business of technology, and has followed the AI investment cycle since the ChatGPT boom began.