BIS Warns: Debt-Driven AI Investment May Replay Historical Bubbles
Taylor Wilson
The Bank for International Settlements warned that the current AI infrastructure investment race has outpaced every prior tech boom in both scale and speed — if productivity gains fail to materialize, the debt structures financing this build-out will amplify the crash, a systemic risk signal investors in AI cannot afford to ignore.
What exactly is the BIS worried about?
BIS economist Phurichai Rungcharoenkitkul notes that AI infrastructure spending has surpassed every historical precedent in just about three years.
This means → the more capacity the industry builds, the higher the productivity bar it must clear to justify the spending — and the harder the fall if it doesn't.
In plain terms = the bigger the bet, the more real-world value AI must create to break even — scale itself becomes the risk.
Which historical bubbles are being invoked?
The BIS draws direct parallels to three episodes: the U.S. canal mania of the 1830s, the British railway mania of the 1840s, and the dot-com bubble of the late 1990s.
All three ended the same way: sharp corrections followed by broader economic damage.
This means → the BIS is not saying AI is useless — it is saying even when the technology is real, an investment pace that outruns adoption triggers a bust. Railways ultimately transformed the world, but railway investors went bankrupt in droves.
Where is the money coming from — and why is the financing structure dangerous?
Tech companies are funding AI expansion through bond issuances and increasingly complex financing arrangements.
Some structures have become circular: cloud-infrastructure firms take equity stakes in AI developers in exchange for compute-purchase commitments — binding their financial fates together.
In plain terms = Company A borrows to build a data center; Company B promises to buy the compute; A uses that promise to raise more capital. If B's demand falls short, the entire chain snaps at once.
A $6 trillion wager — who is placing it?
JPMorgan and Goldman Sachs estimate AI-related spending will approach $6 trillion by 2030, most of it debt-financed.
The BIS sees AI as a winner-take-all market: industry-wide investment drives collective progress, but returns concentrate in a handful of winners.
This reflects a structural contradiction: every company is incentivized to invest more (fear of falling behind), yet aggregate industry spending almost certainly overshoots actual demand.
What does the BIS concede — and what remains unresolved?
The report acknowledges that latent demand for AI services is "clearly enormous" — large enough to support massive compute expansion.
The core question: whether the current build-out pace has already exceeded what demand can sustain, and whether complex debt structures can survive a stress test if expectations disappoint.
In plain terms = the BIS is not bearish on AI — it is saying a leg running too fast may be more dangerous than one standing still, especially when that leg is propped up by borrowed money.
Content is for reference only, not financial advice.