Morgan Stanley has become the dominant adviser for the largest and most innovative AI infrastructure financing projects since last year, according to industry insiders.
These include a $3.2 billion bond for Google-backed data center developer TeraWulf, a $27 billion debt package to tie up Meta’s Hyperion data center with Blue Owl and, most recently, advising Broadcom on a $35 billion chip financing deal.
The bank’s growth in AI-related businesses helped it surpass longtime rival Goldman Sachs in revenue from debt and equity markets fees in the first half of the year. Revenues rose to $2.3 billion, up from $1.4 billion a year earlier, LSEG data showed. This put Morgan Stanley in second place in the world for capital markets fees, just behind JPMorgan Chase, after being fourth a year earlier.
These deals show how artificial intelligence is not only changing the technology sector but also the capital markets. Instead of relying solely on traditional project finance or corporate borrowing, bankers are increasingly creating structures that turn long-term contracts for computing capacity and the balance sheets of large technology companies into securities that can be sold to institutional investors.
Such an approach has greatly expanded the amount of capital available to fund AI infrastructure, but has also tied much of the financial system to continued strong demand for AI.
“Amounts that used to be a billion, two or five billion dollars are now $10 billion, $20 billion or more,” said Mo Assomull, one of the investment banking executives at Morgan Stanley.
The fastest growing market segment
The technology giants from Silicon Valley claim that they cannot satisfy all the demands of their clients and continue to increase their investments. Morgan Stanley estimates that the construction of artificial intelligence infrastructure will attract as much as $10 trillion in investment in the coming years.
Key to securing favorable financing and billions of dollars in capital was the inclusion of the so-called hyperscalers – Google, Amazon, Meta and Microsoft – who entered the AI expansion with exceptionally strong balance sheets. When one of these companies guarantees a data center lease, financing costs are almost cut in half.
“Where do you want to raise money — at interest rates in the mid to high single digits or at double that rate?” Assomull asked.
William Graham, one of the heads of the debt financing sector at Morgan Stanley, was behind the arrangement that has become a kind of model for financing AI infrastructure, reports SEEbiz.
His team designed a bond for data center developer TeraWulf, creating a security aimed at a wide range of investors. At the same time, certain elements of project financing were incorporated, which created a hybrid instrument supported by Google. Most of the data center’s capacity will be reserved for Anthropic, according to sources familiar with the project.
That structure attracted a whole new group of credit investors, including insurance companies, asset managers and pension funds, allowing TeraWulf to raise $3.2 billion at a yield of 7.75 percent.
Patrick Fleury, CFO of TeraWulf, said the new construction bond allowed them to avoid the slower procedures that are characteristic of traditional bank loans for project financing, while still obtaining capital on favorable terms.
“We were practically able to borrow by relying on the strength of Google’s balance sheet,” Fleury said.
To further reassure investors, Morgan Stanley has implemented certain project financing mechanisms, including special rental collection accounts that ensure revenues go directly to bondholders, along with additional collateral instruments.
Since the deal with TeraWulf, Morgan Stanley has sold more than $40 billion in construction bonds and is expanding the same model to markets in Asia and Europe.
“We will get to a point where AI infrastructure bonds will make up the majority of new annual supply of high-yield debt,” Graham predicts. “It’s the fastest growing segment of the market and the first new segment we’ve won in 20 years.”
Some rival bankers admit they don’t want too much exposure to this segment, especially because of growing opposition from local communities to the construction of large data centers across the US.
JPMorgan CFO Jeremy Barnum warned this week that the bank has analyzed some data center financing deals and concluded that it simply does not want to finance some of them.
Key risk
Morgan Stanley helped expand the model beyond financing data centers themselves. In May, the bank and MUFG secured a $3.1 billion loan to CoreWeave for the purchase and installation of Nvidia graphics processors, key to the development of advanced AI systems.
It was the first financing of GPU infrastructure realized through a widely syndicated term loan, thus opening a new source of capital for financing the chips themselves.
According to Graham, the loan attracted nearly $20 billion in investor interest, showing strong demand for financing models that rely more on chip usage contracts than their market value.
Under that structure, the GPU chips and the data center are financed separately. The building is financed through leases, while the chips are financed through long-term take-or-pay contracts.
“You can think of it as a car and a garage. The chip is a Ferrari, but it needs a place to park. That’s why you need a data center,” Graham explained.
Although the chips themselves represent additional insurance, investors have paid much more attention to who is behind the contract for their use.
In March, Morgan Stanley helped CoreWeave secure an $8.5 billion chip loan at an interest rate of 2.25 percent above the benchmark rate. The loan from May was supported by weaker credit profiles – two AI laboratories – so its price was 4.5 percent above the reference rate.
It is precisely this difference that shows the greatest risk of the AI financial boom. The more funding moves away from the strong balance sheets of tech giants like Google, Microsoft or Meta and relies on AI companies that consume massive amounts of computing power, the higher the credit risk.
