Nvidia’s meteoric rise — its market value roughly tripling in two years — has made CEO Jensen Huang a symbol of the AI boom. Huang has downplayed talk of an AI bubble, but many of the loudest defenders of continued heavy investment have the most to gain. Meanwhile, a growing number of analysts and economists say the risks beneath the hype are real and widening.
Some investors and advisers describe the current surge as a “super-cycle” rather than a bubble. Figures like David Sacks, Ben Horowitz and Mary Callahan Erdoes argue that demand and growth justify aggressive spending. Skeptics counter that much of the new capital is flowing into speculative projects whose returns are uncertain. Venture investor Paul Kedrosky warns that enormous sums are chasing a revolution whose rate of improvement may be slowing. Economist Daron Acemoglu cautions that useful AI tools will arrive over the next decade, but current expectations are inflated.
The scale of planned spending is striking. OpenAI’s CEO has said the company generates about $20 billion a year and could spend up to $1.4 trillion on data centers over eight years — a buildout that presumes rapidly expanding customer adoption. Industry estimates, however, show most businesses are not yet seeing chatbots or large-language models meaningfully change profits, and relatively few consumers pay for AI services.
Big Tech’s commitments are enormous: Amazon, Google, Meta and Microsoft are expected to spend roughly $400 billion on AI this year, mainly on computing infrastructure. Morgan Stanley has projected roughly $3 trillion in AI infrastructure spending by tech firms through 2028, with about half of that needing outside funding if internal cash flows aren’t sufficient. Analysts warn that if demand weakens or capacity outstrips need, much of those investments could become stranded.
Financing practices add to the unease. Companies are leaning on debt and complex vehicles that can hide liabilities. Goldman Sachs analysts estimate hyperscalers took on about $121 billion in debt over the past year — a more than 300% jump from normal levels. Firms are using special-purpose vehicles, joint ventures and lease-backed structures so large loans don’t appear on their balance sheets. One example involved a $27 billion loan to build a data center, where a tech buyer held a minority stake, received full capacity, but the loan didn’t show up as the buyer’s debt — a setup that could leave the buyer exposed to large payments if the project fails.
Critics see echoes of past excesses. Off-balance-sheet financing and circular deals — where suppliers fund customers who then buy the supplier’s hardware — can inflate apparent demand. Nvidia’s strategic arrangements with AI operators, and deals among chipmakers, cloud providers and specialized data-center firms can create interlocking obligations that obscure true end-user uptake. Startups that pivoted from crypto mining to AI compute, like CoreWeave, have struck multi-billion-dollar usage and equity deals with large AI firms; those cross-ownership and capacity-guarantee arrangements worry economists who fear they mask weak organic demand.
Some market participants have reacted. Investors including Peter Thiel and SoftBank trimmed Nvidia holdings, and Michael Burry has publicly bet against Nvidia, arguing parts of the AI economy rely on circular financing and optimistic accounting. Even industry leaders acknowledge froth: OpenAI’s CEO said investors may be “overexcited,” and Google’s CEO has warned of irrationality and cautioned that no company would be immune in a downturn.
Observers point to the dot-com era, when debt-financed infrastructure builds far outpaced real demand, producing big losses. The central risk today is a mismatch of assumptions: vast bets that AI revenue will scale quickly and monetize easily. If performance gains slow, enterprise adoption lags, or monetization proves harder than promised, the web of debt, off-balance financing and interdependent deals could expose fragility — stranded assets, lender losses and broader market pain.