Investors poured hundreds of billions into artificial intelligence over the past three years. Tech giants raced to build data centers and train ever-larger models. Stock prices soared. Yet signs of strain have multiplied. Returns on those massive outlays remain uncertain. Limits of the technology grow clearer by the quarter. And prominent voices now warn the entire surge could unwind faster than many expect.
The Warnings Mount
Jeremy Grantham doesn’t mince words. The longtime investor and GMO founder told The Guardian he plans to sell his tech shares soon. He sees AI repeating a familiar pattern. Everyone overinvests at first. Then reality sets in. The invention becomes a utility. Real money flows to services built around it, not the core technology itself. “AI was similar to the invention of railways or the internet in that everyone overinvests and, when they realise it is a utility, such as electricity, they understand that there is not much money to be made from the invention itself, except for the companies that build services around it,” Grantham said.
His view finds company among other heavyweights. Ray Dalio, Bridgewater Associates founder, told Bloomberg the AI market shows clear bubble signals. It will burst as paper wealth converts back into actual money. Chinese hedge funds have grown even louder. Shanghai Banxia Investment Management Center declared in a June letter that “the trigger for the AI bubble to burst has already appeared.” They pointed to slowing revenue growth at key players like Anthropic, per a separate Bloomberg report.
But the data tells a more layered story. The five largest hyperscalers — Amazon, Alphabet, Meta, Microsoft and likely one more — stand on track to spend over $1 trillion on AI capital expenditures in 2025 and 2026 alone. That’s according to the Bank for International Settlements’ latest annual report, analyzed in The Wall Street Journal. The BIS, which flagged risks before the 2008 crisis, fears fierce competition will drive spending to unsustainable levels. Profitability could suffer. A sharp reversal might tip economies into recession.
Manufacturing offers a concrete example of the gap between hype and application. Companies have automated simple, repeatable tasks for decades. AI promised to tackle complexity. Yet production floors face late supplier deliveries, machine breakdowns, demand swings and regulatory hurdles. Current systems can’t handle that variability well. As Felicity Bradstock reported in Yahoo Finance on July 12, firms now recognize the enduring value of human workers. AI can’t replace them in messy, real-world settings. Confidence erodes. Deployment slows.
Productivity numbers back the skepticism. A National Bureau of Economic Research study from February 2026 found 90% of firms reported no measurable AI impact on workplace output. Executives still projected gains of 1.4% in productivity and 0.8% in overall output. The disconnect echoes the old productivity paradox. Billions spent. Minimal visible return so far. Wikipedia’s entry on the AI bubble, updated through mid-2026, captures these debates in detail.
Meanwhile, the spending war continues. Four hyperscalers alone planned $670 billion in AI infrastructure for this year. J.P. Morgan analysts forecast $5 trillion total through 2030. That scale exceeds every major U.S. capital project in history except the Louisiana Purchase, adjusted for GDP share, The Wall Street Journal noted earlier. OpenAI and Anthropic generate annualized revenue around $25 billion and $19 billion respectively. The math doesn’t add up without explosive adoption. Time highlighted this mismatch in March, urging preparation for the fallout.
IPOs add fuel. SpaceX eyes a record $1.77 trillion valuation. Anthropic and OpenAI plan massive public debuts this summer. David Wallace-Wells and Natasha Sarin dissected the risks in a New York Times opinion piece. The public will soon buy into trillion-dollar promises. Whether those valuations hold once retail investors bear the weight remains an open question. Retirement accounts could take the hit if optimism fades.
Yet not everyone sees doom. David Rubenstein pushed back in a Bloomberg video interview, arguing he doesn’t expect the stock bubble to burst anytime soon. Nvidia’s forward price-to-earnings ratio sits near 23 times, its lowest since 2019. Revenue hit $216 billion, up 65%, with 88% growth forecast next year. Some analysts on X, including traders sharing charts today, call the “bubble” talk overblown. Price has trailed earnings acceleration, they argue.
Still, token costs tell another tale. Chamath Palihapitiya posted on X this weekend about real-world deployment at his firm 8090. Early gains from frontier models on complex enterprise systems have shrunk. Costs doubled. The last 5% of performance demands massive context, endless A/B tests and more tokens. “The fully loaded cost of the model plus the engineer will not pay for itself,” he concluded. Engineers risk treating models like slot machines, turning off their own thinking while OpEx balloons. Many observers on the platform today echoed the theme. “The AI bubble is popping,” several users wrote.
The BIS report spells out broader ripple effects. An AI investment bust could disorder the global financial system. Banks hold exposure through loans and securities tied to tech. Pension funds and insurers chase yields in the same names. A synchronized selloff would hurt. European markets, left behind in the AI trade, might look attractive as a hedge, suggested one Wall Street Journal column from mid-June. All that money flooding into AI serves as its own warning sign, per a separate Journal analysis.
History offers lessons without guaranteeing outcomes. The dot-com bubble inflated on similar excitement over transformative technology. Many firms vanished. Survivors like Amazon and Google reshaped commerce. AI could follow suit. But the current debt load, concentrated among fewer players, raises different systemic questions. Gary Marcus, AI researcher, declared on his Substack that the bubble is “all over now, baby blue.” Clues appear everywhere, from slowing returns to mounting costs.
So what happens next? Companies may dial back capex if revenue growth disappoints. Hyperscalers could rent out excess compute, signaling overbuild. Token prices might fall as competition intensifies. Or the technology could break through barriers in reasoning and reliability, justifying the spend. Few predict the latter will arrive fast enough. Manufacturing plants still need humans. Complex software systems resist full automation. Users notice when outputs falter.
The frenzy built quickly. ChatGPT’s 2022 launch ignited it. Trillions in market value followed. Now the infrastructure bill arrives. Whether markets digest it calmly or seize up will shape the next decade of tech investing. Grantham, Dalio and the BIS have sounded the alarm. Industry insiders would do well to listen. The data centers keep rising. The questions about payback only grow louder.
WebProNews is an iEntry Publication