Perhaps nobody embodies artificial-intelligence mania quite like Jensen Huang, chief executive of chip giant Nvidia, whose stock has surged roughly 300% in two years. On a recent earnings call, Huang moved quickly to deflate bubble talk: “There’s been a lot of talk about an AI bubble,” he told investors. “From our vantage point, we see something very different.”
That defensive posture is common: many with the most to gain from perpetual AI spending insist critics are wrong. White House AI adviser and venture capitalist David Sacks called the moment an “investment super-cycle.” Prominent investor Ben Horowitz said demand forecasts make the situation look nothing like a bubble. JPMorgan executive Mary Callahan Erdoes dismissed the idea that AI spending is a bubble and called the technology revolutionary for corporate operations.
Skeptics aren’t persuaded. Paul Kedrosky, a venture capitalist and MIT research fellow, warns capital is pouring into a largely speculative revolution. “The technology is very useful, but the pace at which it is improving has more or less ground to a halt,” he said, arguing it’s unrealistic to expect the same rapid cadence of breakthroughs to continue for years.
The huge infusion of cash
The flow of money into AI is enormous. OpenAI’s CEO Sam Altman has claimed the company makes about $20 billion a year and plans to spend $1.4 trillion on data centers over eight years—projections that depend on rapidly growing customer adoption. Yet studies and industry surveys suggest many firms aren’t seeing chatbots translate into profits, and only a small share of consumers currently pay for AI services.
Economist Daron Acemoglu of MIT, a 2024 Nobel laureate in Economic Sciences, says the industry’s promises are often exaggerated. Still, Amazon, Google, Meta and Microsoft are expected to spend roughly $400 billion on AI this year, mostly on data centers—some dedicating up to half their cash flow to that build-out. That scale of investment implies a revenue base far larger than seems likely, says Kedrosky: spreading the cost across every iPhone user globally would still require more than $250 per user.
To avoid draining balance sheets, big tech is tapping private equity and debt.
Paving the AI future with debt and risky financing
Goldman Sachs analysts found hyperscaler companies have taken on about $121 billion in debt over the past year—more than triple typical levels. Some deals use special purpose vehicles (SPVs) to keep debt off corporate balance sheets: outside investors fund data centers, the SPV borrows to buy chips, and the tech firm gets computing capacity without showing the loan on its books.
A recent Meta–Blue Owl joint venture illustrates the structure: Blue Owl borrowed $27 billion to finance a data center; Meta leases and effectively guarantees payments, owns a minority stake, and gets the computing power without the loan showing up on Meta’s balance sheet. If demand evaporates and the facility goes idle, Meta could still face massive payments. Analyst Gil Luria cautions that SPVs have a fraught history—recalling Enron—and warns they shouldn’t become the foundation of the industry’s future.
Enormous spending hinging on returns that could be a fantasy
Morgan Stanley projects Big Tech could spend about $3 trillion on AI infrastructure through 2028, with only half funded from operating cash flows. If demand slows or steadies below expectations, capacity could be overbuilt, leaving debt and lenders exposed. Luria points to the dot-com era, when debt-financed build-outs—like fiber networks—helped fuel a market collapse. The same dynamic could repeat if billions are sunk into data centers that aren’t ultimately needed.
Circular deals raise even more concern
Some investments appear circular. Nvidia announced a roughly $100 billion deal to finance OpenAI’s data centers; OpenAI would then use Nvidia chips in those facilities. Critics say such arrangements can inflate apparent demand—vendors effectively funding customers to buy their products—echoing practices seen at smaller scales and during past bubbles.
Smaller players are also entwined. CoreWeave, once a crypto-mining startup, pivoted to AI-focused data centers and now rents capacity to major AI firms. OpenAI struck deals that exchanged use of CoreWeave capacity for equity, which OpenAI could then use to pay for services. Nvidia owns part of CoreWeave and guaranteed to absorb unused capacity through 2032—another form of backstop that can mask real market demand. Acemoglu warns these structures could form a fragile “house of cards.”
Some high-profile investors see a pop coming
Signs of market nervousness are appearing. Peter Thiel reportedly sold an entire Nvidia stake; SoftBank trimmed a large holding. Michael Burry—famous for betting against the housing bubble in 2008—has publicly bet against Nvidia, accusing the industry of accounting tricks and circular financing that mask true end demand. Burry argues many customers are effectively funded by their suppliers.
Even some industry leaders admit excessive enthusiasm. OpenAI’s Sam Altman told reporters investors as a whole are “overexcited about AI,” while still calling AI “the most important thing to happen in a very long time.” Google CEO Sundar Pichai acknowledged “elements of irrationality” in the market and noted no company would be immune if a bubble burst.
Bottom line
The debate over whether an AI bubble exists pits those who stand to gain from continued, rapid spending against those who see speculative financing, rising debt, off-balance-sheet structures, circular deals and uncertain near-term demand. The scale of planned infrastructure investment and the financial engineering supporting it raise questions about what happens if growth doesn’t match expectations—questions that echo the warnings of prior tech bubbles.