Morgan Stanley forecasts $2.9 trillion in global data center spending driven by AI from 2025 to 2028—but projects that the biggest technology companies will cover only about half through their own cash flows. That leaves an estimated $1.5 trillion financing gap, with private credit expected to supply much of the external funding.
Key takeaways
- Morgan Stanley expects hardware to absorb $1.6 trillion of the projected spending.
- Private credit could supply roughly $800 billion through asset-specific loans.
- The US data center power deficit is forecast to reach 32 to 33 gigawatts by 2028.
AI-driven data center spending splits between hardware and construction
Morgan Stanley allocates about $1.6 trillion to IT hardware, including chips and servers, and $1.3 trillion to physical infrastructure and construction. The forecast covers both the computing equipment needed for artificial intelligence and the facilities that house it.
Amazon, Microsoft, Alphabet, Meta and Oracle are projected to contribute about $1.4 trillion from their own cash flows. Those hyperscalers would therefore self-fund roughly half of the anticipated investment, leaving the rest to external financing.
Private credit takes a larger role in financing the buildout
Morgan Stanley’s research suggests private credit markets could supply approximately $800 billion of the $1.5 trillion shortfall. That funding would use asset-specific structures: loans secured against particular facilities or equipment rather than a borrower’s overall balance sheet.
Such arrangements already feature in financing for independent computing providers, with lenders taking graphics processing units and customer software contracts as collateral. Banks also finance physical construction, arranging and syndicating project loans.
The research points to potentially higher yields from asset-specific private credit financing than from conventional corporate debt. In addition to this borrowing, commitments from hyperscalers under lease agreements total more than $1.2 trillion in undiscounted future payments.
Those lease payments support much of the construction lending behind the buildout. The spending forecast thus sits alongside substantial contractual commitments to the developers providing the facilities.
Nvidia has also participated in discussions about backstop guarantees of up to $250 billion to support data center debt financing. The discussions put supplier guarantees alongside private loans and leases in the funding picture for AI-driven data center spending.
US power supply presents a separate constraint
The US data center power shortfall is forecast to increase to 32 to 33 gigawatts by 2028. That projected deficit represents a physical constraint on the buildout beyond the financing gap.
Morgan Stanley’s investment forecast runs from 2025 through 2028, the same endpoint as the revised US power-shortfall projection.
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.