What happened
Morgan Stanley's research team put a number on something the market has been circling for a year. In a note dated Friday, the bank forecasts $2.9 trillion in global data center capital expenditure through 2028, per CryptoBriefing. Hyperscaler internal cash flow plus investment-grade debt issuance covers roughly $1.4 trillion of that. The remaining $1.5 trillion has to come from somewhere else.
The analysts frame private credit as the base case. Asset-backed securitizations tied to GPU inventory and long-dated power purchase agreements, sale-leaseback deals on completed campuses, and direct lending from the Blackstones and Apollos of the world are the mechanisms. The bank isn't predicting a funding crisis. It's saying the capital stack has to broaden, fast.
Why it matters
The number reframes the AI buildout as a credit story, not just an equity one. For eighteen months the narrative has been about hyperscaler earnings and chip supply. Morgan Stanley is pointing at the plumbing underneath: who finances the concrete, the transformers, the Nvidia H200s, and the twenty-year power contracts.
That shifts attention toward players crypto investors already know. Core Scientific, Iris Energy, TeraWulf, and Hut 8 all sit on power infrastructure originally built for Bitcoin mining and are now pivoting hardware and real estate toward AI colocation customers. CoreWeave, which went public earlier this year, is the pure-play version of that trade. Riot Platforms disclosed in its last filing that it was in talks on an AI hosting agreement at its Rockdale site. The $1.5 trillion gap is the backdrop against which those pivots either print money or stall.