Episode Details

Chief Fixed Income Strategist Vishy Tirupattur takes a look at how credit markets are adapting to fund the new phase of AI capex.

Read more insights from Morgan Stanley.

----- Transcript -----

Welcome to Thoughts on the Market. I am Vishy Tirupattur, Morgan Stanley’s Chief Fixed Income Strategist. 

Today – The critical question behind the AI-driven capex cycle that is front and center for markets year to date. How is credit market financing this ecosystem evolving? 

It’s Wednesday June 3rd at 2 pm in New York. 

When we first discussed the role of credit markets in financing the AI and data center build-out around the middle of last year, the direction of travel was clear. Realizing the transformative potential of AI requires unprecedented levels of capex. What has really surprised us since is the scale and speed of that spending, both of which have exceeded our expectations by a wide margin. 

The upward revision to capex expectations has been dramatic. A year ago, we projected the combined capex of the five large hyperscalers at roughly $450 billion in both 2026 and 2027. After the first quarter earnings reports, Morgan Stanley’s internet equity analysts, led by Brian Nowak, now expect hyperscaler capex of roughly $800 billion in 2026 and $1.2 trillion in 2027. One data point really captures the surge in the underlying demand for compute. According to OpenRouter, the global weekly token usage, which is a key proxy for compute, has risen by roughly 350 percent since early January, increasing from about 6 trillion tokens to 28 trillion tokens. 

Credit channels for financing this capex have not only been broader and deeper than we anticipated, spanning public and private markets, but have seen remarkable in the structural innovation that is blurring the lines between public and private markets. Over $200bn of public AI-related issuance across the different credit channels has happened just in the first five months of this year. We had previously assumed unsecured issuance would be limited by the scale of the largest non-financial issuers, confined to investment grade credit only, and largely USD denominated. Instead, some hyperscaler issuance has now far exceeded even the largest telecom names; funding has expanded well beyond USD into EUR, GBP, CHF, JPY and CAD markets. The issuer base has also broadened to include data center REITs and neoclouds, particularly in the high-yield market. 

The scope of financing has also widened beyond the data center shells themselves. GPU financing, which we assumed would be funded entirely through equity capital, has begun to migrate into credit markets. Funding is now coming through broadly syndicated loans and asset based financing, with ABS structures not far behind. 

Structural innovation illustrates how rapidly the credit ecosystem is adapting to the complexities of demands of AI-driven capex. Financings that combine elements of project finance, tranching, and residual value guarantees, along with high-yield issuance backed by hyperscaler guaranteed leases – these are innovations that we have never seen before. These structures have expanded the investor base, reduced the funding frictions, and further blurred traditional boundaries – between both corporate and project finance, and public and private credit markets. 

At the same time, physical, operational, and political constraints are beginning to shape the pace and the composition of the AI infrastructure build-out – and, by extension, the demand for financing. Grid access, power generation equipment, skilled labor, and permitting delays are emerging as significant constraints. These are compounded by political and regulatory frictions at the local, national, and international level. As power av