Tracking credit-funded AI datacenter builds

Research evidence excerpt

Tracking credit-funded AI datacenter builds

Goldman Sachs The Credit Line

Even though these agreements are more transparent, it is only a fraction of public debt

market funding related to AI (again, Exhibit 1). Indeed, the gap between anticipated

datacenter compute funded by these agreements and all US datacenter power demand

capacity by 2027 remains very large (Exhibit 2). Taken together, we think the sample

represented by these JVs are, on their own, a small fraction of the broader buildout, and

not necessarily representative. That said, the line of sight from funding costs to project

outcomes remains an important thread as investors grapple with the sheer scale of the

current, and projected, buildout.

Exhibit 2: Joint-venture deals are only a fraction of current

compute, but we believe they contain some important

signals

GW

Anticipated datacenter power demand capacity by 2027

JV debt-affiliated All US datacenters

Source: Aterio, Goldman Sachs Global Investment Research

JV documentation varies, but market pricing has largely coalesced

As we have discussed previously, the deal terms across these structures vary

meaningfully. Yet, despite these idiosyncrasies, investors have largely traded them as a

homogeneous cohort. Yields have generally clustered in a range (Exhibit 3) and spread

moves across structures have been more correlated than the broader USD IG and HY

markets (Exhibit 4). In our view, that pattern likely reflects the fact that investors have

been underwriting a common thematic exposure first, and specific structural differences

second. These are not credits levered to the operating performance of an ultimate

guarantor in the conventional sense, but they are levered to the same underlying theme:

sustained demand for more compute and continued AI investment.