Morgan Stanley becomes Wall Street's top bank for AI debt trading
(Morgan Stanley) has emerged as Wall Street’s leading force in building the financing structure behind the artificial intelligence (AI) boom, designing new debt and equity financing models to channel tens of billions of dollars in large-scale construction of data centers. Since last year, the bank has become the lead advisor, leading the largest and most innovative AI infrastructure financing deals, according to industry executives. That includes a $100 million deal for data center developer TeraWulf. (Google) backed a $3.2 billion debt financing package for Meta's Hyperion data center project with Blue Owl, and recently advised Broadcom on a $35 billion chip financing deal. The bank's rapid growth in AI dealmaking helped it surpass its long-term rival in debt and equity capital markets fee income in the first half of this year, according to London Stock Exchange Group (LSEG) data (Goldman Sachs), revenue grew to $2.3 billion from $1.4 billion in the same period last year. This moved Morgan Stanley to second place in the global capital markets fee income rankings (after ), higher than the fourth place in the same period last year. These deals show that AI is reshaping not only technology but also capital markets. Rather than relying solely on traditional project financing or corporate lending, bankers are increasingly designing structures that package long-term computing contracts and the balance sheets of big tech companies into securities that can be marketed to mainstream investors. The result is a vastly expanded pool of capital available for investment in AI infrastructure, while more closely linking the financial system to the ongoing demand for AI computing. "What used to be $1 billion, $2 billion, $5 billion is now $10 billion or $20 billion or more," said Mo Asomour, co-head of investment banking at Morgan Stanley. Silicon Valley's tech giants say they can't fill customer orders and continue to increase spending. Morgan Stanley itself predicts that AI infrastructure construction will consume US$10 trillion in expenditures in the next few years. The key to getting low interest rates and billions of dollars of capital is bringing in hyperscale players—Google, (Amazon), Meta and (Microsoft) - They entered the AI boom with healthy balance sheets. When one of the companies guarantees a data center lease, financing costs are roughly halved. “Would you rather finance it at a rate in the mid-to-high single digits or twice that,” Asomour said. William Graham, co-head of leveraged finance at Morgan Stanley, pushed for the deal, which has become a template for AI infrastructure financing. His team designed a broadly marketable bond for data center developer TeraWulf. But they added some protections to the project loans, creating a hybrid vehicle with Google “backing up” (i.e. providing guarantees). According to people familiar with the matter, most of the data center's capacity will be reserved for Anthropic. The structure attracted new credit investors such as insurance companies, asset managers and pension funds into data center construction financing, allowing TeraWulf to raise $3.2 billion at a 7.75% yield. TeraWulf Chief Financial Officer Patrick Fleury said the new construction bond allows them to skip the slow, phased lender review required to take out traditional project finance loans from banks - their main alternative - while still borrowing at a cost low enough for the business economics to stand. "Effectively, we borrowed against the strength of Google's balance sheet," Fleury said. To further reassure investors, Morgan Stanley adopted features of project finance, such as a "lock box" mechanism that segregates lease payments and pays them directly to bondholders, and added other additional collateral. Since the TeraWulf deal, Morgan Stanley has sold more than $40 billion of this construction bond product and rolled out the structure into new markets in Asia and Europe.
Graham predicted: “We will reach a stage where AI infrastructure bonds will make up the majority of new non-investment grade debt issuance each year. This is the fastest growing segment of the market and the first new market segment to emerge in the past 20 years.” Some rival bankers say they are reluctant to be viewed as the top player in the space, given the unpopularity of data centers in communities across the United States. JPMorgan Chief Financial Officer Jeremy Barnum also warned this week that the bank looked at the loan terms of some data center financing deals and "we were like, 'Yeah, we're not going to do that.'" Morgan Stanley also helps expand the model beyond the data center itself. In May this year, the bank and Group (MUFG) arranged a $3.1 billion loan for neocloud company CoreWeave to purchase and install Nvidia ( ) Graphics processing unit (GPU)—the chip needed to drive advanced AI. This is the first GPU financing completed in the form of a syndicated term loan, bringing in a broader pool of funds - this time to finance the chips themselves. Graham said the loan attracted nearly $20 billion in investor demand, underscoring interest in a financing model that relies less on the value of the chips and more on the use of chip contracts. Under this structure, GPUs and data centers are financed separately: the buildings are backed by leases, while the chips are financed under long-term take-or-pay agreements. "You might have a car payment, you might have a garage bill," Graham said. "The chip is …It needs a place to park. So you need a data center to house that chip. " While the chips serve as additional collateral, Graham said credit investors are more focused on who signed the contract to lease the chips. Morgan Stanley helped CoreWeave price an $8.5 billion chip loan in March, backed by a contract with a hyperscaler, at an interest rate of 2.25 percentage points above the benchmark rate. The chip loan in May was backed by two AI labs with weak credit and was priced at the base rate plus 4.5 percentage points. The wider spreads also highlight a key risk underpinning the AI funding boom. The farther away a loan is from the balance sheet of a very large company, and the closer it is to the AI lab that consumes a lot of computing power, the weaker the underlying credit quality will be. Raj Joshi, senior vice president at Moody’s Ratings, said the financial health of Anthropic and OpenAI is a risk he watches closely. Joshi said: "This is a huge capital expenditure investment cycle, there is no precedent in history. There is no template to follow."
Graham predicted: “We will reach a stage where AI infrastructure bonds will make up the majority of new non-investment grade debt issuance each year. This is the fastest growing segment of the market and the first new market segment to emerge in the past 20 years.” Some rival bankers say they are reluctant to be viewed as the top player in the space, given the unpopularity of data centers in communities across the United States. JPMorgan Chief Financial Officer Jeremy Barnum also warned this week that the bank looked at the loan terms of some data center financing deals and "we were like, 'Yeah, we're not going to do that.'" Morgan Stanley also helps expand the model beyond the data center itself. In May this year, the bank and Group (MUFG) arranged a $3.1 billion loan for neocloud company CoreWeave to purchase and install Nvidia ( ) Graphics processing unit (GPU)—the chip needed to drive advanced AI. This is the first GPU financing completed in the form of a syndicated term loan, bringing in a broader pool of funds - this time to finance the chips themselves. Graham said the loan attracted nearly $20 billion in investor demand, underscoring interest in a financing model that relies less on the value of the chips and more on the use of chip contracts. Under this structure, GPUs and data centers are financed separately: the buildings are backed by leases, while the chips are financed under long-term take-or-pay agreements. "You might have a car payment, you might have a garage bill," Graham said. "The chip is …It needs a place to park. So you need a data center to house that chip. " While the chips serve as additional collateral, Graham said credit investors are more focused on who signed the contract to lease the chips. Morgan Stanley helped CoreWeave price an $8.5 billion chip loan in March, backed by a contract with a hyperscaler, at an interest rate of 2.25 percentage points above the benchmark rate. The chip loan in May was backed by two AI labs with weak credit and was priced at the base rate plus 4.5 percentage points. The wider spreads also highlight a key risk underpinning the AI funding boom. The farther away a loan is from the balance sheet of a very large company, and the closer it is to the AI lab that consumes a lot of computing power, the weaker the underlying credit quality will be. Raj Joshi, senior vice president at Moody’s Ratings, said the financial health of Anthropic and OpenAI is a risk he watches closely. Joshi said: "This is a huge capital expenditure investment cycle, there is no precedent in history. There is no template to follow."