A wall of debt, spread across every market, is needed over the next five years to finance data-centre expansion.
JPMorgan, in research cited by Bloomberg, estimates hyperscalers will require about US$1.5 trillion of investment-grade bonds to help build the computing power for AI.
Total capital needs will be at least US$5 trillion, and potentially could be as high as US$7 trillion.
Even after layering in roughly US$150 billion from leveraged finance and as much as US$40 billion a year via data-centre securitisations, the bank sees a sizeable gap to be filled by private credit and, in part, governments.
High-grade issuance tied to AI sites could reach US$300 billion in 2026, close to one-fifth of a US$1.6 trillion market. Demand has surged: recent jumbo deals from Meta and Oracle drew record orders, underscoring investor appetite.