Morgan Stanley maps $1.5 trillion AI data-center financing gap
Morgan Stanley says AI data-center spending needs about $1.5 trillion of outside capital. The underwriting lesson is lease cash flow, power access and refinancing evidence, not AI demand alone.
Morgan Stanley estimates that global data-center spending will reach $2.9 trillion through 2028. According to the bank's July research and an August 6 podcast, roughly $1.5 trillion of that total must come from outside capital. Private credit and asset-based finance could provide $800 billion of this gap. Another $150 billion could come via data-center asset-backed securities (ABS) and commercial mortgage-backed securities (CMBS). The data-center financing surge involves a complex credit structure. Morgan Stanley projects that $1.5 trillion in outside financing will be required by 2028. This shifts the focus from general growth to specific collateral, lease terms, power contracts, refinancing windows, and underwritable cash flows. Porter Stansberry noted these financing questions in an August 11 thread. Additional details appear in Morgan Stanley Research , Morgan Stanley's published podcast transcript , Moody's data-center credit-risk page , and L&G's securitized-credit analysis . Morgan Stanley breaks down the capital stack as follows: $1.4 trillion of hyperscaler capital expenditure will come from cash flow, $200 billion from investment-grade tech debt, $800 billion from private credit or asset-based financing, $150 billion from securitized credit, and $350 billion from other sources. In the podcast, Vishwas Patkar stated the $1.5 trillion gap exceeds the size of both the high-yield and leveraged-loan markets. Financing is moving into several different underwriting channels. Corporate bonds rely on a single large issuer. Private credit and asset-based finance rely on project cash flows, leases, collateral, construction progress, and specific lender protections. Securitization typically occurs after an asset is built, leased, and generating income. L&G noted this sequence in a February report on data-center securitization. The firm stated that most securitized issuance refinances existing income-generating assets instead of funding new construction. Data-center ABS and CMBS issuance exceeded $25 billion in 2025. L&G also identified a refinancing risk: many deals have a five-year repayment target. This makes the period between 2028 and 2030 a critical test if refinancing markets decline. Moody's analyzes this market through credit risk. Its data-center page states that developers and landlords will use equity, bank loans, corporate bonds, securitized bonds, and project-finance vehicles. Leverage will likely increase for developers building hyperscale facilities between 2026 and 2028. Moody's also notes that power, water, and development approvals act as constraints. For operators in merchant cash advance (MCA), revenue-based financing (RBF), or equipment finance, the parallel is found in repayment logic. Rapid demand can mask poor repayment structures. A lender might see a merchant with high revenue but miss that the cash flow relies on a single processor, a seasonal buyer, an expiring contract, a second-position lender, or a landlord dispute. Similarly, a data-center lender might see AI demand but overlook risks regarding power delivery, tenant renewals, lease increases, obsolescence, construction timing, or the exit market. The underwriting process requires identifying the revenue source, the constraints, hidden debt-like obligations, and the exit strategy. If a deal relies on refinancing, lenders must identify the specific future buyer and the evidence required to secure them. Crowding is also a factor. If private credit absorbs hundreds of billions in data-center finance, lenders will remain selective elsewhere. The insurance, pension, and private-credit buyers funding infrastructure also fund warehouse lines and asset-backed facilities. If these buyers concentrate on one theme, specialty-finance originators should expect more scrutiny regarding concentration, liquidity, documentation, and performance history. AI infrastructure is not inherently bad credit. If leases are durable, power is secured, the tenant mix is stable, and a refinance path exists, the assets may be attractive. However, AI demand is not collateral. Collateral consists of the contract, the operating asset, power access, tenant payments, and the market capable of refinancing the structure. Opinion The collateral test is more than just the building AI infrastructure credit depends on leases, power access, tenant concentration, and takeout markets, not just rising compute demand. Smaller finance desks face the same reality: high demand in a sector does not fix a weak credit file. It only increases the number of lenders asking for proof of cash flow. Plutus tracks how capital-market pressure reaches the underwriting file at plutusai.net .