The Bank of England used the word "stretched" in October. Sam Altman, the man running OpenAI, has acknowledged a bubble exists. Peter Thiel just sold his entire Nvidia stake. But Jensen Huang insists the skeptics have it all wrong.
If 2025 demonstrated anything, it's that the AI investment debate has calcified into two irreconcilable camps, each armed with compelling data and each convinced the other is catastrophically mistaken.
The January shock nobody predicted
DeepSeek launched a chatbot in late December. By January 27, Nvidia had lost $589 billion in market value, the largest single-day loss in stock market history. The Chinese startup claimed it built a model rivaling ChatGPT for just $5.6 million in training costs.
Marc Andreessen called it "one of the most amazing and impressive breakthroughs I've ever seen." Which is notable given his firm has poured billions into AI infrastructure predicated on the assumption that such breakthroughs required vastly more capital.
The market recovered. Nvidia's share price recovered 8.8% the following day. By July, the company had become the first public company in the world to reach a $4 trillion market cap, bypassing Microsoft and Apple. Jensen Huang's net worth climbed to $140 billion, up $25 billion for the year.
The whiplash was instructive.
What the MIT numbers actually show
In August, a report from MIT's Media Lab landed with the subtlety of a grenade. Despite $30 to $40 billion in enterprise investment into generative AI, 95% of organizations are getting zero return.
The headline went viral. AI skeptics seized on it as validation. But the methodology deserves scrutiny.
The research was based on 150 interviews with leaders, a survey of 350 employees, and an analysis of 300 public AI deployments. The finding of "zero return" came from just 52 interviews that the report itself admits are "directionally accurate based on individual interviews rather than official company reporting."
The methodology didn't account for efficiency gains, cost reductions, customer churn reduction, lead conversion improvements, or sales pipeline velocity. The study measured revenue acceleration specifically, not broader business impact.
This matters. The 95% figure is real. But it's measuring a narrow definition of success during what even AI bulls acknowledge is an infrastructure-building phase.
The concentration problem
Since the October 2022 bear market bottom and the launch of ChatGPT, the S&P 500 has soared 90%, but most of these gains have come from a small group of stocks. The Magnificent Seven, plus another 34 AI-adjacent companies, account for roughly 75% of overall market returns, 80% of earnings growth, and a staggering 90% of capital spending growth in the index.
The top five members of the S&P 500 now command nearly 30% of market share, a record high for any point over the past 50 years.
This is the kind of statistic that can mean different things depending on your priors. Bears see fragility. Bulls see justified concentration around the companies actually building the future.
Morgan Stanley Wealth Management's chief investment officer Lisa Shalett falls into the concerned camp. "At the end of the day … this is not going to be pretty" if the generative AI capital expenditure story falters, she told Fortune. She's watching for a "Cisco moment", referencing the company that was briefly the world's most valuable before losing 80% of its stock price when the dotcom bubble burst.
When asked how close we are to such a moment, Shalett said probably not in the next nine months, but very possibly in the next 24.
The circular financing problem
Paul Kedrosky, an investor and podcaster interviewed by Derek Thompson, describes the web of AI deals as increasingly incestuous.
Take a recent $100 billion deal between Nvidia and OpenAI. Nvidia pumps money into OpenAI to bankroll data centers. OpenAI then fills those facilities with Nvidia's chips. Kedrosky sees a problem: "The idea is I'm Nvidia and I want OpenAI to buy more of my chips, so I give them money to do it. It's fairly common at a small scale, but it's unusual to see it in the tens and hundreds of billions of dollars."
OpenAI has entered deals with CoreWeave worth tens of billions of dollars in which CoreWeave's chip capacity in data centers is rented out to OpenAI in exchange for stock in CoreWeave. Nvidia owns part of CoreWeave and has a deal guaranteeing it will absorb unused data center capacity through 2032.
"The danger," said MIT economist Daron Acemoglu, "is that these kinds of deals eventually reveal a house of cards."
The November exits
The smart money started walking toward the exits.
Peter Thiel's hedge fund Thiel Macro LLC sold off its entire position of 537,742 shares in Nvidia during the third quarter, worth roughly $100 million. At one point, Nvidia represented nearly 40% of Thiel's investment portfolio.
Japan's SoftBank announced it sold off its shares in Nvidia in October for $5.83 billion, though primarily to fund its OpenAI investment.
Michael Burry, who made hundreds of millions betting against the housing market in 2008, disclosed bearish wagers against both Nvidia and Palantir. He posted a cryptic message on X: "sometimes, we see bubbles."
Burry later took issue with the accounting practices of hyperscalers, noting that companies like Microsoft and Alphabet have extended depreciation schedules for their investments in recent years, helping cut expenses and boost profits.
The bulls note that Thiel remains invested in OpenAI and AI startups. SoftBank's sale funded more AI bets. The exits look less like bearish conviction and more like portfolio rebalancing.
What the Bank of England actually said
Central banks typically avoid the word "bubble." The Bank of England got as close as central banks get.
In October, its Financial Policy Committee warned that "equity valuations appear stretched," particularly for AI-focused tech firms. The earnings yield implied by the cyclically adjusted price-to-earnings ratio was close to the lowest level in 25 years, comparable to the peak of the dotcom bubble.
"The risk of a sharp market correction has increased," the FPC stated.
By December, the warnings intensified. The Bank of England warned that a multi-trillion dollar spending boom in artificial intelligence infrastructure financed by debt risks unraveling given "materially stretched" stock market valuations.
The Bank noted early warning signs in credit default swaps of companies leaning on debt to fund AI investments.
The case against panic
Not everyone at the table is nervous.
The S&P 500 Information Technology Index trades around 30x forward earnings today, elevated by historical standards, but well below the 55x multiple reached at the peak of the dot-com era. BlackRock notes that valuations today reflect real revenues, proven business models, and the accelerating adoption of AI across industries.
The Magnificent Seven boast a net negative debt of -22%, meaning they hold more cash than financial debt. They can finance massive investments without heavy borrowing. Their net margin sits around 29%, reflecting extraordinary profitability. The tech leaders during the dotcom bubble had margins of just 16%.
Federal Reserve Chair Jerome Powell has stated that AI differs from other technology bubbles in that the corporations behind it are generating large amounts of revenue and that investment into AI data centres is generating large amounts of economic growth.
Goldman Sachs remains cautiously optimistic. The investment bank's analysis of the Magnificent Seven's median price-to-earnings ratio found it is "roughly half" that of the largest seven companies from the late-1990s dot-com era.
The OpenAI contradiction
OpenAI sits at the center of the debate, embodying both its promise and its problems.
The company reached a $500 billion valuation in October 2025, following a $6.6 billion secondary share sale. The sale cemented OpenAI's status as the world's most valuable privately held company, surpassing SpaceX.
Revenue is growing. Annualized revenue hit $10 billion as of June 2025, up from $5.5 billion in December 2024.
But so are the losses. Deutsche Bank estimates OpenAI could accrue about $143 billion in negative cumulative free cash flow between 2024 and 2029 before turning profitable. The company expects to burn $115 billion in cash through 2029.
OpenAI says it will not make money until 2030 and may have to raise hundreds of billions of dollars to cover losses and investments in AI data centers.
This week, reports emerged that OpenAI is in talks to raise up to $100 billion at a valuation of up to $830 billion. The spread from $500 billion to $830 billion in a matter of months is, as one outlet put it, "pretty wild."
The dotcom comparison, interrogated
The parallels are obvious. The differences might be more important.
In the three years leading up to the March 2000 peak, the Nasdaq 100 climbed more than 500%. The AI rally has delivered more modest gains. From November 2022 to its most recent peak in October 2025, the Nasdaq rose only 125%.
During the dot-com bubble, a record 36% of tech stocks were unprofitable. As of September 2025, that number was only 19% and has actually come down significantly since 2021.
The companies leading this rally generate real revenue. Between September 2021 and September 2025, earnings in the S&P 500 tech sector jumped by 77%. During the internet bubble, profits increased roughly 79% while prices skyrocketed 479%.
The Allianz chief investment officer called the current environment "less a bubble and more a boom underpinned by fundamentals," while advising to balance optimism with caution.
What happens next
McKinsey estimates that companies will invest almost $7 trillion in global data center infrastructure capital expenditures by 2030, a figure equivalent to the combined GDP of Japan and Germany.
Spending from US mega caps is expected to reach $1.1 trillion between 2026 and 2029, and total AI spending is expected to surpass $1.6 trillion.
The infrastructure will get built regardless of whether the bubble thesis proves correct. Past technology-related infrastructure hype cycles suggest that the data centers, electrical infrastructure, and fiber networks being built are unlikely to go to waste. These hard assets will likely form the backbone of a new economy.
"Bubbles always hurt some investors, but the capacity they create endures."
This might be the most honest summary available. AI may be a bubble. The technology is also real. Both things can be true simultaneously.
Jamie Dimon, head of JP Morgan, put it plainly in October: "AI is real" but some money invested now will be wasted. An AI-driven stock crash could result in a lot of invested money being lost, "just like cars in total paid off, and TVs in total paid off, but most people involved in them didn't do well."
The debate will continue. The spending won't stop. And in twelve months, one of these camps will look prescient and the other will look foolish.
We just won't know which until then.
