Morgan Stanley's AI Debt Machine: How Wall Street's New Power Broker Is Financing the $10 Trillion Build-Out
There is a number that captures the scale of what Morgan Stanley has quietly built over the past eighteen months: 40. That is how many billion dollars the bank has sold in a new class of AI infrastructure bonds - a financial instrument that barely existed two years ago and is now reshaping how the world's most capital-intensive technology build-out gets funded.
On Monday, the Financial Times reported that Morgan Stanley has emerged as Wall Street's dominant architect of AI financing, edging past longtime rival Goldman Sachs in debt and equity capital market fees in the first half of 2026. The bank's capital markets revenue grew to $2.3 billion, up from $1.4 billion a year earlier, placing it second globally behind only JPMorgan Chase. The engine behind that surge is a financing model that did not exist in its current form before the AI boom - and that Morgan Stanley's bankers believe will become the defining credit market of the decade.
The Template Deal That Changed Everything
The story begins with a data centre developer called TeraWulf and a bond structure that William Graham, Morgan Stanley's leveraged finance co-head, designed to solve a specific problem: how do you get mainstream credit investors - insurers, pension funds, asset managers - to finance a data centre that will primarily serve Anthropic, an AI lab with no public credit rating and a history of losses?
The answer was to wrap the bond in Google's balance sheet. TeraWulf raised $3.2 billion at a 7.75 percent yield by issuing a security that combined the broad distribution of a corporate bond with the structural protections of a project finance loan, backstopped by a Google lease guarantee. The result was a new asset class: the AI infrastructure bond. Since that deal, Morgan Stanley has sold more than $40 billion of similar instruments and is now taking the structure to markets in Asia and Europe.
"Dollar amounts that used to be $1 billion, $2 billion, $5 billion are now $10 billion or $20 billion and higher," said Mo Assomull, Morgan Stanley's co-head of investment banking. The bank itself estimates the total AI build-out will consume $10 trillion of spending over coming years - a figure that implies the financing structures being invented today will need to scale by orders of magnitude.
From Buildings to Chips
Morgan Stanley has not stopped at financing the buildings. In May, the bank and MUFG arranged a $3.1 billion loan for neocloud CoreWeave to buy and install Nvidia graphics processing units - the first GPU financing done as a broadly syndicated term loan. The deal attracted nearly $20 billion in investor demand, a ratio that underscores just how hungry credit markets are for AI-linked paper.
The structure separates the data centre from the chips: the building is financed against a lease, while the GPUs are financed against long-term take-or-pay contracts. Graham described it with a memorable analogy: "The chip is the Ferrari... it needs a place to be parked." In March, Morgan Stanley helped CoreWeave price an $8.5 billion chip loan backed by a hyperscaler contract at just 2.25 percent above the benchmark rate. A subsequent loan backed by weaker AI lab credits priced at 4.5 percent above - a spread that illustrates exactly where the risk in this market lives.
The Risk Hiding in Plain Sight
The further the lending moves away from the pristine balance sheets of Google, Amazon, Meta, and Microsoft, the thinner the underlying credit becomes. Raj Joshi, a senior vice-president at Moody's Ratings, told the FT that the financial health of Anthropic and OpenAI is a risk he watches closely. "This is a huge capex investment cycle, you don't have parallels to it in history," Joshi said. "There is no playbook for this."
That warning deserves to be taken seriously. The AI debt market is being built on the assumption that demand for computing power will remain insatiable - that the hyperscalers will keep signing long-term contracts, that the AI labs consuming that compute will remain solvent, and that the revenue models justifying all of this capital will eventually materialise. Each of those assumptions is reasonable. None of them is guaranteed.
Vishal Khanduja, head of broad markets fixed income at Morgan Stanley, flagged earlier this month that credit risk is being undervalued in the market, pointing to Amazon's surprise $25 billion bond offering as evidence that supply is beginning to test investor appetite. Big Tech now accounts for more than 8 percent of the total US corporate bond market - a record high - and some analysts project the AI debt market could swell to $7 trillion by 2029.
What This Means for Markets
Morgan Stanley's rise to the top of AI dealmaking is not just a story about one bank winning market share. It is a signal that the AI build-out has entered a new phase - one where the financing infrastructure is becoming as important as the technology itself. The bank's prediction that AI infrastructure bonds will eventually represent the majority of new non-investment-grade debt supply is not a boast. It is a structural forecast about where capital markets are heading.
For investors, the implications cut both ways. The new financing structures have dramatically expanded the pool of capital available to fund AI infrastructure, which is good for the build-out. But they have also deepened the financial system's exposure to a single technology cycle in ways that have no historical precedent. When Graham says there is no parallel to this capex cycle in history, he is right - and that is precisely the reason the risk deserves more attention than the current spread levels suggest it is getting.
