Credit Notes: Three Themes in AI-Related Credit

Research evidence excerpt

Credit Notes: Three Themes in AI-Related Credit

Goldman Sachs Credit Notes

related to the quick pacing of supply and the potential for lower returns on training

capex (following news of open-weight model performance) has a clear passthrough to

corporate risk levered to AI. As we discussed in our initial note on datacenter structures,

these bonds are less exposed to marginal changes in capex ROI and should be

underwritten on the project-specific terms. There is even a potential upside scenario, in

our view, where investors in existing construction deals could benefit from less resource

and construction demand associated with softer sentiment—assuming interest and

principal does not require extending current lease terms.

Exhibit 2: Datacenter construction bond spreads have

tightened, while offtake corporate bonds have slightly

widened

Spreads for datacenter financing JV bonds vs. duration-matched

corporate bonds from the ultimate lease guarantor

bp Datacenter construction vs. corporate guarantor bond spread

300 Duration-matched corporate bond JV spread JV - corporate bond

-50

-100

Initial Current

Median

Source: Bloomberg, Goldman Sachs Global Investment Research

While we find value in the excess yield associated with these structures, we think there

are a few considerations investors should keep in mind. First, supply technicals for these

structures have thus far been benign (Exhibit 3) relative to corporate AI supply—

especially given their novelty, excess yield, and secured structuring. We anticipate

project finance issuance only rises from here, as we estimate current issuance covers

only a fraction of expected compute capacity. We expect dispersion to rise as deals

continue to be announced, especially if construction timelines grow more aggressively or