The Trillion-Dollar AI Buildout Runs on Debt. Central Banks Are Starting to Notice.

The world's biggest technology companies are on pace to spend more than $1 trillion combined on AI infrastructure this decade: chips, data centers, power grids, networking. Most coverage has framed this as a technology race. The Bank for International Settlements, the Basel-based forum that convenes the world's major central banks, is framing it as a credit cycle.

In its Annual Economic Report 2026, the BIS argues the financing behind the AI buildout is concentrating leverage in the same places that preceded past crises. Hyperscalers and chip leaders are still borrowing aggressively even though some generate tens of billions in annual free cash flow, because AI capital expenditure is outrunning internal cash generation at this scale. That borrowing is increasingly landing on the balance sheets of large commercial banks, and increasingly on nonbank private-credit funds that have taken share from traditional lenders. BIS Bulletin 120 reaches the same conclusion from a different angle: the migration of AI-adjacent lending into private credit may not disperse systemic risk so much as hide it, because those lenders operate with lighter disclosure and capital buffers than regulated banks.

The Basel III Endgame "take two" published last year by US Federal banking agencies would force large US banks to hold more loss-absorbing capital, equity and high-quality reserves, against riskier loan books: data-center lending, structured financings tied to AI infrastructure, and exposures to private-credit vehicles that themselves fund AI projects. Freshfields' explainer walks through the eight key changes; EY and PwC converge on the same mechanism. In practice, a higher capital floor on a data-center loan reduces the return a bank can earn on a given dollar of equity allocated to that loan, which means the loan either gets priced higher, gets rejected, or gets pushed to a private-credit fund willing to charge more but able to absorb less loss.

The European Central Bank's Financial Stability Review, May 2026 special article flags rising financial-stability risks from nonbank credit intermediation feeding the AI capex super-cycle, complementing the BIS signal on both sides of the Atlantic. The convergence matters because it suggests the repricing is not a US-specific quirk of one bank's loan book; it is a coordinated regulatory response to a financing pattern central banks now consider systemically risky.

The BIS report doesn't argue regulators should slow AI development, nor does it predict an imminent crisis. Its recommendation is narrower and more technical: the broader nonbank credit ecosystem funding AI investment needs tighter oversight, and banks should hold more capital against the AI-linked portion of their books. AI financing is becoming more expensive and harder to obtain not because policymakers object to AI, but because they've concluded the current financing pattern concentrates risk in ways past experience has repeatedly punished.

AI loans are increasingly bundled, syndicated, and sold across borders, with the same underlying data-center asset sitting on multiple balance sheets in different forms. That multi-balance-sheet stacking is what central banks now treat as systemic. The Financial Stability Board has separately tracked the same pattern in nonbank leverage, and American Banker has reported on the warnings directly. Fortune's coverage frames the moment as a $1 trillion gamble.

The next round of AI capex headlines will keep arriving with trillion-dollar totals attached. Inside those releases, the financing structure will increasingly carry more signal than the capex number alone: who is bank- or fund-financed, how much is syndicated rather than self-funded, and what portion is being pushed into private credit. A bank- and fund-financed capex line now carries a different systemic cost than one paid out of operating cash flow, because the financed version is the version Basel III Endgame will reprice first. Yahoo Finance's syndication is where most readers will encounter the framing; the underlying analysis is in the BIS report itself.

The question for the next earnings call is no longer how much is being spent on AI. It is who is holding the paper.