BIS: AI boom is pivoting from cash flows to debt — and it all rides on the hype

A BIS Bulletin from central-bank economists Aldasoro, Doerr and Rees traces a structural change in how the AI buildout is paid for. AI-related capital spending — data centres, servers, networking, cooling, power and the chip fabs behind them — has climbed to roughly 1% of US GDP, pushing total IT-related investment to about 5% of GDP and past its dot-com peak. The difference from 2000 is telling: this time the spending comes from the IT producers themselves, not the firms buying IT. That investment has done real macroeconomic work, contributing around 0.4 percentage points to GDP growth annually and accounting for nearly half of recent quarterly growth, cushioning the drag from tariffs. Data centre outlays alone are projected to grow by $100–225 billion over five years.

The financing story is where the risk sits. The hyperscalers driving the boom (Alphabet, Amazon, Meta, Microsoft, Oracle) have historically carried little debt and self-funded from cash flow, but capex has now outrun free cash flow, and equity issuance is unattractive given volatile, concentrated valuations. So they are turning to debt — bonds, leases, loans and, increasingly, private credit. Private credit lending to AI-related firms has jumped from near zero to over $200 billion, close to 8% of all outstanding private-credit loans, and could hit $300–600 billion by 2030. Notably, the terms on those loans (spreads around 6.2%, maturities near five years, similar collateral rates) look no different from loans to any other sector.

That pricing is the report’s central warning. If lenders treat AI loans as ordinary-risk while equity markets price in extraordinary future returns, one side is wrong: either credit investors are underpricing the risk as their exposure balloons, or equity is overpricing the payoff. Add circular financing across the AI ecosystem (the OpenAI–Nvidia web of deals) and leverage moved off balance sheets, and a disappointment in AI earnings could trigger sharp, correlated corrections in both debt and equity. The authors judge overall systemic risk as moderate — the boom is smaller relative to GDP than the shale or dot-com booms — but note that busts have historically shaved more than a point off GDP growth, and that hidden leverage plus heavy investor concentration in US AI equities could make the spillovers worse this time.