Behind AI's money-burning, Wall Street has invented a new business.

Morgan Stanley is turning AI infrastructure financing into a brand new asset class in capital markets, overtaking Goldman Sachs in investment banking through this approach.

According to the Financial Times on Tuesday, the Wall Street bank has become the main architect of AI data center financing structures, by packaging the balance sheets of tech giants with long-term compute power contracts into securities that can be sold to mainstream investors, thereby opening up new channels for financing this unprecedented wave of capital expenditure.

According to LSEG data, in the first half of this year, Morgan Stanley’s debt and equity capital market fees grew more than 60% year-on-year to $2.3 billion, jumping from fourth to second place in global capital market fee rankings, second only to JPMorgan Chase.

This model is reshaping the capital market landscape. As the scale of AI infrastructure financing jumps from the past level of $1-5 billion to $10-20 billion or even higher, more insurance companies, asset management institutions, and pension funds are being drawn into this track, deepening the ties between the financial system and AI compute demand.

New Financing Template: Leveraging Tech Giants’ Credit to Unlock Cheap Capital

The TeraWulf bond, designed under the leadership of William Graham, co-head of Leveraged Finance at Morgan Stanley, has become the industry template for AI infrastructure financing.

This structure combines bonds that can be sold to a wide range of investors with the protective terms of project loans, forming a hybrid instrument endorsed and guaranteed by Google. To further reassure investors, Morgan Stanley introduced a “lockbox” mechanism, transferring lease payments directly to bondholders and adding extra collateral. In the end, TeraWulf successfully raised $3.2 billion at a yield of 7.75%.

TeraWulf CFO Patrick Fleury stated that this innovative structure allowed the company to bypass the cumbersome step-by-step review process of traditional project financing loans, while borrowing at a sufficiently low cost to make the business model economically feasible. “In essence, we are financing by leveraging the credit strength of Google’s balance sheet,” he said.

The core logic is: when hyperscale cloud service providers such as Google, Amazon, Meta, or Microsoft guarantee a data center lease, financing costs are roughly cut in half. Morgan Stanley’s co-head of investment banking, Mo Assomull, described the difference as “the choice between the mid-to-high single digits and double that.”

From Data Centers to Chips: The Expanding Boundaries of Financing

Morgan Stanley has not stopped at data centers themselves, but is extending this financing logic to the chip level.

In May this year, Morgan Stanley and Mitsubishi UFJ Financial Group (MUFG) arranged a $3.1 billion loan for emerging cloud provider CoreWeave to purchase and deploy Nvidia GPUs. This was the first GPU financing completed in the form of a broadly syndicated term loan, bringing more capital into the field of chip financing. The loan attracted nearly $20 billion in investor subscription demand.

In this structure, data center buildings and chips are financed independently: the former backed by leases, the latter by long-term “use-or-pay” contracts. William Graham likened it to automobiles: “The chip is a Ferrari... it needs a place to park, so you need a data center to house the chip.”

Differences in credit quality are directly reflected in pricing. In March this year, Morgan Stanley helped CoreWeave complete an $8.5 billion chip loan backed by contracts from hyperscale cloud service providers, priced at 225 basis points above the benchmark rate; the May loan was backed by two AI labs, whose credit quality was relatively weaker, resulting in a price premium of 450 basis points.

Risk Accumulation: The Credit Chain Gets Longer

As financing structures extend beyond hyperscale cloud providers, potential risks are quietly accumulating.

Raj Joshi, Senior Vice President at Moody's Ratings, said he is closely monitoring the financial health of Anthropic and OpenAI. “This is a huge capital expenditure investment cycle, with no historical precedent,” he said. “There is no ready-made manual.”

Some rival bankers said that, given the controversies sparked by data centers in communities across the US, they do not wish to be the leading players in this field. JPMorgan CFO Jeremy Barnum also warned this week that, after reviewing the loan terms for some data center financing deals, “we will not participate.”

Morgan Stanley itself expects that AI infrastructure construction will consume $10 trillion of expenditure over the next few years. William Graham is even more optimistic about the market’s prospects, predicting that AI infrastructure bonds will eventually account for the majority of new annual non-investment-grade debt supply, characterizing it as “the first new market segment to appear in capital markets in the past 20 years.” However, whether this financing feast can continue ultimately depends on whether AI compute demand can deliver on its promises.

Risk Disclosure and DisclaimerThe market has risks, and investment must be prudent. This article does not constitute personal investment advice, nor does it take into account the special investment objectives, financial status, or needs of individual users. Users should consider whether any opinion, view, or conclusion in this article is suitable for their specific situation. Investment is at your own risk.