Morgan Stanley Turns AI Data Centers Into Wall Street's Newest Bond Machine

Morgan Stanley's capital-markets fees hit $2.3 billion in the first half of the year, up from $1.4 billion twelve months earlier. That jump was enough to push the bank past Goldman Sachs into second place globally in capital-markets fees, trailing only JPMorgan, according to LSEG data. The driver behind the surge is not a wave of tech IPOs or M&A activity. It is debt, specifically the bonds and loans Morgan Stanley has engineered to fund the physical infrastructure behind the AI boom: data centers, chips, and the power contracts that keep them running.

The bank has moved more than $40 billion in what amounts to a new category of structured debt, selling most of the paper to investors rather than holding it on its own balance sheet. William Graham, Morgan Stanley's co-head of leveraged finance, has called AI infrastructure bonds the fastest-growing segment in the market and the first genuinely new bond category to emerge in two decades.

How the deals are built

The mechanics rely on a familiar trick: attaching a AAA-rated tech giant's credibility to a project that would otherwise carry far more credit risk. TeraWulf, a data-center developer, raised $3.2 billion at a 7.75% yield on a bond effectively backed by Google, even though most of the compute capacity the deal funds is already contracted out to Anthropic. The structure lets bondholders price the debt closer to Google's risk profile than to a leveraged infrastructure builder's, while Anthropic gets the compute it needs to keep scaling.

Morgan Stanley has run the same playbook elsewhere. It arranged $27 billion in financing for Meta's Hyperion data-center site and structured a $35 billion chip-related deal for Broadcom. On CoreWeave, the AI cloud provider that has become a bellwether for this financing wave, the bank priced one loan backed by a hyperscaler at 225 basis points over the benchmark rate, and a near-identical loan backed by two AI labs at nearly double that spread, 450 basis points. The gap illustrates how much a single-name hyperscaler guarantee is worth to investors compared with backing from AI labs that, however well-funded, don't carry the same balance-sheet weight or credit history.

Not everyone is buying in

The strategy has also drawn skepticism from rivals. JPMorgan's chief financial officer said the bank examined some of these AI infrastructure bond structures and chose to sit the business out, a rare public signal of caution about how these deals are priced and secured. The divergence points to a real debate on Wall Street over whether corporate guarantees from hyperscalers like Google and Meta adequately offset the underlying risk that data-center demand, largely driven by a handful of AI labs' compute needs, could shift or consolidate faster than the debt gets repaid.

For now, investor appetite has been strong enough that Morgan Stanley has been able to sell most of what it originates rather than warehouse the risk. That distribution capability, more than any single deal, is what has turned AI infrastructure financing into a durable, fee-generating business rather than a one-off trade.