But a growing chorus of skeptics says the fever is bigger than the boosters admit. Venture capitalist Paul Kedrosky argues that while the technology is useful, the pace of improvement has slowed and it is unrealistic to expect the rapid cadence of breakthroughs to continue indefinitely. MIT economist and Nobel laureate Daron Acemoglu warns that industry promises are often exaggerated and that policymakers and investors should be cautious.
The sums flowing into AI are enormous. OpenAI’s CEO Sam Altman has said the company now earns roughly $20 billion a year and plans massive data-center investment over the next several years — projections that depend on fast, sustained customer adoption. Independent studies and surveys, however, suggest many firms are still struggling to turn chatbots into consistent profit, and most consumers do not yet pay for advanced AI services. Meanwhile, analysts estimate Amazon, Google, Meta and Microsoft together could spend about $400 billion on AI this year alone, largely on data-center capacity. Morgan Stanley has even projected cumulative spending on AI infrastructure could top $3 trillion through 2028. Those figures imply a revenue base far larger than current adoption supports; Kedrosky notes that, distributed across the global smartphone base, the numbers demand an implausibly high per-user return.
To fund this build-out without immediately weakening balance sheets, big tech is turning to private capital and heavy borrowing. Goldman Sachs analysts found hyperscalers have taken on roughly $121 billion in new debt over the past year, more than triple typical levels. Financing structures have grown more complex: special-purpose vehicles (SPVs) and third-party deals let companies obtain capacity without showing the loans on their own books. That can mask risk. One recent high-profile example involved a Blue Owl financing of a Meta data center where the asset was funded and leveraged off Meta’s balance sheet while Meta leased and effectively guaranteed payments, keeping the debt at the SPV level. If demand falls and facilities sit idle, those payment obligations can still fall back on the tech companies.
Critics point to the dot-com era for a cautionary parallel: during that cycle, debt-financed build-outs such as fiber networks contributed to a collapse when expected demand didn’t materialize. If billions of dollars of data centers are built and not needed, lenders and corporate sponsors could be exposed.
Worry deepens around circular and backstopped deals. Reports of a roughly $100 billion financing arrangement tied to Nvidia and OpenAI raised concerns that vendors are effectively funding customers who then spend on those vendors’ chips, inflating apparent demand. Smaller providers like CoreWeave, which pivoted from crypto mining to AI renting, have become deeply intertwined: OpenAI took equity in exchange for capacity, Nvidia holds stakes, and commitments to absorb unused capacity through long horizons can obscure true market appetite. Acemoglu likens these layered guarantees to a fragile house of cards.
Market signals are starting to show unease. High-profile investors have trimmed or exited large AI-related stakes: Peter Thiel reportedly sold his Nvidia holding, SoftBank pared back exposure, and Michael Burry has publicly bet against Nvidia, accusing parts of the industry of accounting tricks and circular financing that hide weak end demand. Even some industry leaders acknowledge excess: Sam Altman has said investors are broadly overexcited about AI, and Google’s Sundar Pichai has warned of elements of irrationality that could leave companies vulnerable if sentiment turns.
Bottom line: the debate separates those with incentives for continued rapid spending from those who see rising debt, off‑balance-sheet financing, circular deals and uncertain near-term demand. The scale of planned infrastructure investment and the financial engineering behind it raise serious questions about what happens if growth stalls — questions that echo the warning signs of past tech bubbles.”}