As companies rush to bring bond issues to market to fund their AI-related spending, this new debt is competing with government bonds for investor demand.
I recently discussed the impact of the flood of AI-related bond supply on the economy and the broader fixed income markets, as well as the details and risks of these new issues, with Steve Boothe, our head of global investment-grade fixed income, and Mark Stodden, a credit analyst.
Arif: How do you see AI capex, and the bond issuance funding it, feeding into long-term economic growth?
Steve: AI capex is an important incremental growth engine for the U.S. economy. In the near term, spending on data centres, chips, power, and related infrastructure is directly supporting investment and growth. The longer-term question is whether the productivity gains and future cash flows justify the amount of capital being deployed. If they do, the cycle can become self-reinforcing, supporting higher equity valuations, tighter credit spreads, and strong cash flows that support additional investment, which supports growth and valuations and higher real yields. But if returns disappoint or the cost of capital rises, capex could slow quickly.
Arif: Is the flood of AI-related issuance affecting corporate credit curves?
Steve: Absolutely. A disproportionate amount of incremental investment-grade supply is being placed into the long end of the market, contributing to steeper credit curves. Approximately one-third of year-to-date investment grade corporate supply has been in the long end of the curve,1 with an outsize contribution from AI-related issuers. I currently see this as a supply and relative valuation issue, not a broad deterioration in credit quality. Long duration investors are being asked to absorb heavy Treasury supply at the same time that hyperscalers and other AI-related borrowers are issuing large amounts of long-dated debt. The clearing price for duration must adjust.
Arif: What is the best way to think about portfolio construction when adding exposure to AI-related bonds?
Mark: A portfolio could accumulate sizable exposure to AI-related risk simply through positions in individually attractive deals— without making a thematic allocation decision. The question is not how much AI to own in absolute terms, but how much active AI risk a portfolio should take relative to its benchmark. The pace of AI credit formation is even more important for portfolio construction. AI could account for roughly half of the growth in the broader U.S. dollar-denominated credit opportunity set through 2029. This reinforces the need to plan AI exposure deliberately and treat new transactions as competing uses of a portfolio-specific risk budget.
Arif: Do you anticipate that hyperscaler credit quality will deteriorate?
Steve: No. Most hyperscalers still have some of the strongest balance sheets in the corporate market. I think supply and the repricing of duration risk will pressure their bonds well before conventional credit deterioration becomes an issue. The more interesting question is what these balance sheets will be like three to five years from now if AI remains this capital intensive. Investors need to look beyond reported debt and begin to consider capacity commitments, guarantees, and other contingent liabilities.
Arif: Where do you see risks in the AI credit cycle?
Mark: The market looks set to stay in the current mode of booming AI demand with scarce compute supply for the near term, but the key risk in the transition to the next part of the AI cycle is that compute will become more abundant. If compute supply eventually outstrips demand while AI consumption remains strong, the values of bonds issued by infrastructure providers and technology suppliers could suffer.
Arif: Where do you find value in the new AI-related supply?
Steve: We are finding more interesting opportunities away from straightforward long-dated unsecured hyperscaler bonds, such as in investment-grade data center and infrastructure financings where investors are being paid for complexity. But these deals are very idiosyncratic and require a differentiated underwriting process relative to vanilla unsecured risk. We care about the quality of the tenant, contract terms, leverage, amortization, construction risk, power availability, geography, and refinancing risk.
