$820 Billion AI Buildout Now Runs on Private Credit
Hyperscalers, the largest cloud and AI compute operators, are projected to spend that much this year, and BlackRock says nonbank lenders are filling the gap that public markets cannot supply.
Hyperscalers, the largest cloud and AI compute operators, are projected to spend that much this year, and BlackRock says nonbank lenders are filling the gap that public markets cannot supply.
Hyperscalers, the largest cloud and AI compute operators, are projected to spend nearly $820 billion on data centers and AI infrastructure in 2026, up roughly 80% from a year earlier. BlackRock says they cannot fund that bill from public markets alone. The Bank for International Settlements, a forum for central banks that publishes economic research, said in March that the alternative carries bank balance-sheet risk.
Jean Boivin, head of the BlackRock Investment Institute, told Bloomberg Television on July 7 that AI infrastructure spending will force the world to 'leverage up' and that private capital is positioned to play a bigger role financing hyperscalers. Public bond markets and bank balance sheets cannot supply the scale at the speed hyperscalers need to keep training runs and inference capacity online. Loans and credit facilities from nonbank lenders, including private credit funds, insurers and structured vehicles, are filling the gap.
Private credit is not new. The scale is. Bloomberg Intelligence projects that the six largest U.S. hyperscalers will spend nearly $820 billion on capex in 2026, roughly 80% above the prior year's record and an outlier against the past decade of corporate spending. The anchor under that forecast is what hyperscalers already guided for fiscal 2025: Amazon committing about $100 billion, Microsoft earmarking $80 billion for AI data centers, and Alphabet directing roughly $75 billion into capex.
On each new project, the marginal dollar now starts with private credit funds and structured vehicles. When a hyperscaler needs another $20 billion to build a campus coming online in 18 months, the funding does not begin in a corporate bond roadshow. It begins with private credit funds and structured vehicles that can size up faster and tolerate project risk. Some of that debt eventually refinances into public bond markets when conditions allow. The split has shifted: private credit absorbs a bigger share of the new money.
The same BIS review noted that AI hyperscalers are increasingly drawing on private credit firms to help finance AI infrastructure, with potential bank balance-sheet spillovers. The risk channels are concrete. Off-balance-sheet vehicles hold the loans. Bank funding lines back those vehicles. Guarantee structures activate if a project stumbles. If private credit appetite turns procyclical in a downturn, refinancing pressure shows up first inside the vehicles, then at the guarantee level, then on bank books.
Boivin's thesis and the BIS flag landed in the same week. Industry coverage has tracked the $700 billion-plus run rate on hyperscaler capex through 2026. The question the BIS review raises is what happens to that run rate when private credit conditions tighten and the off-balance-sheet layer has to roll over.
What to watch next: how much of the 2026 capex shows up as private credit drawdowns versus public bond issuance, and whether any hyperscaler has to restructure an off-balance-sheet financing vehicle in a higher-rate environment. BIS economists have already flagged the channel. The next test is whether the channel transmits.