The Bank for International Settlements -- the institution often described as the central bank for central banks -- issued a warning that the current wave of AI infrastructure spending has taken on bubble characteristics severe enough that its collapse could "take the global economy with it," drawing explicit historical parallels to the dot-com crash and 19th-century railway mania.
The report singles out Oracle as the most exposed major player: the company's stock has fallen more than 40% in the past month even as it has committed $300 billion to the Stargate data-center project with OpenAI, a commitment that reportedly requires Oracle to borrow roughly $25 billion annually just to keep financing its side of the deal. The BIS frames this dynamic bluntly, with one cited industry voice describing the arrangement as "leasing property to release to somebody else who may or may not be able to pay it" -- a direct reference to Oracle building capacity for OpenAI, a customer that cannot yet self-fund its own operations and depends entirely on external capital.
Oracle isn't alone in scale of commitment: Amazon has planned more than $200 billion in AI buildouts this year, Microsoft roughly $190 billion, and Google around $180 billion -- meaning the capital at risk if demand assumptions prove wrong spans every major hyperscaler, not one company's balance sheet. That scale of aggregate commitment is what elevates this from a company-specific risk story to the systemic-risk framing the BIS is applying.
“That scale of aggregate commitment is what elevates this from a company-specific risk story to the systemic-risk framing the BIS is applying.”
The comparison to the dot-com era is instructive but imperfect: unlike 1999-2000, today's AI infrastructure spend is backed by companies with genuinely enormous existing cash flows (Microsoft, Google, Amazon) rather than pre-revenue dot-com startups -- but the BIS's concern is specifically about the financing structure underneath newer entrants like OpenAI, where debt and leased capacity, not equity, increasingly fund the buildout, echoing the leverage dynamics that made the railway mania and 2008 financial crisis so much more damaging than the dot-com bust.
The report also flags infrastructure-level constraints compounding the financial risk: power sourcing, permitting delays and growing local moratoriums against new AI data centers threaten to slow project completion even as the capital commitments to build them are already locked in -- a mismatch between spending pace and physical buildout pace that increases the odds of stranded or delayed assets.
For infrastructure and credit investors, the BIS warning is a genuine institutional-grade signal, distinct from the steady drumbeat of pundit-level "AI bubble" commentary that has run through 2025 and 2026 -- this is a body whose job is literally systemic financial-stability monitoring, and it chose to name Oracle specifically and draw a historical-mania comparison, which is a meaningfully stronger statement than typical market commentary.
The bear case for taking this at face value: the BIS has issued similar cautionary notes about other asset classes in past cycles without those warnings preceding an imminent crash, and Oracle's specific financing structure may not generalize to better-capitalized hyperscalers like Microsoft, Amazon and Google, whose AI capex is funded far more by existing free cash flow than by debt.
What to watch: whether Oracle's borrowing costs rise further as credit markets price in the BIS's concerns, whether OpenAI's revenue growth can plausibly catch up to the infrastructure commitments built around it, and whether other central banks or regulators echo the BIS's systemic-risk framing in the coming months.