What to Consider When Analyzing AI Issuance

Having previously discussed how an experienced and multi-skilled credit research team can help investors navigate the size and diversity of AI-related issuance, let’s examine PPM’s clear and consistent approach to differentiating these credits.

In our view, a useful model to consider for AI-related debt deals in the corporate bond market is the high yield data center construction project. Given the uncertainty in data center supply and demand in the outyears of many of these projects, PPM’s credit research team examines the strength of the cash flows associated with the data center as our primary form of credit support.

First, we look at construction risk. High yield data center issuances reviewed by PPM to-date have generally had a pre-leased tenant in place, with rent commencement contingent on the data center’s completion. Many deals to-date enable the tenant to cancel the lease entirely if construction is delayed past a certain point. To understand this risk, analysts consider various factors such as the construction timeline itself, general contractors/power providers involved, and equipment/labor procurement progress. In our view, the ideal data center is typically already built and has a tenant; otherwise, we prefer construction near completion with all labor and equipment sourced. A reliable contractor providing a guaranteed maximum price can also be a credit strength we consider.

Next, we assess tenant risk. As these deals typically have a single tenant, assessing if the tenant is creditworthy is critical to evaluating whether lease payments (and by extension principal and interest payments) are likely to be made. In our view, the ideal tenants generally include hyperscalers, which are currently some of the highest-rated credits in our coverage universe. Other tenants have included so-called “neoclouds” that are newer, AI-focused cloud providers. PPM analysts are cautious on neocloud tenanted data centers, as risks related to the creditworthiness of a neocloud are likely highly correlated to those related to the residual value of the data center if the tenant were to default.   

Related to tenant risks are lease cancellation risks. Lease cancellation due to construction delays has been mentioned, but some leases we have seen contain cancellation clauses related to outages or convenience. Lease cancellation related to outages are often derived from Service Level Agreements, or “SLAs.” An SLA outlines what requirements a data center provider must meet, such as specific power uptime requirements or temperature and humidity thresholds. Many leases are typically cancellable by the tenant if the landlord violates an SLA a specified number of times. Some leases may also be cancelled for convenience but may feature a cancellation fee paid by the tenant. PPM considers whether this cancellation fee appears sizable enough to cover the quantum of debt for the life of the lease.

Lastly, we assess whether the structure of the bond itself provides creditor protections. The credit research team generally has a strong preference for amortization within the lease term, as that helps insulate creditors from both re-leasing risk and residual value risk. Although data center demand is strong today, overbuilding in the next 10 to 15 years could make re-leasing less likely and impair the underlying value of the data center, potentially to the detriment of creditors.

Up next… in our third blog post on AI issuance, we share insights from our private and structured team on how to assess ABF deals.