Hyperscalers are opening the door to significant risk with AI-related short term lease commitments, which threaten to reduce their financial and operating flexibility and leave investors out of pocket, a report from Moody’s Ratings has found.

While these short-term commitments reduce their upfront capital investments, hyperscalers may be understating their economic risk, leaving investors in jeopardy – especially in the event that the expected profitability of the AI market does not materialize.

US data center leases have historically spanned 10-15 years, but the fast-paced growth of AI is leading to shorter lease terms that align with the lifespan of compute infrastructure.

These leases are often backstopped by a large off-balance-sheet guarantee from the lessee. Moody’s said that assessing this credit now requires a “deep dive into the borrower’s full leasing commitments, renewals periods, and Residual Value Guarantees (RVGs).

Typically, RVGs make up any shortfall in value of assets compared to the pre-agreed amount if the lessee cancels the lease early or does not renew the lease.

However, due to the rapid hardware cycles that come hand-in-hand with AI workloads, hyperscalers are able to argue they aren’t "reasonably certain” of renewal, allowing extension periods and RVGs to be excluded from lease liabilities.

Moody’s said this creates a “reporting deferral” where official disclosures “fail to show the full scale of long-term economic risk.”

According to data from the ratings agency, hyperscaler data center commitments grew markedly in 2025, with $969 billion in committed future leases. More than two-thirds ($662bn) is for leases yet to be commenced.

“To put that into perspective, the amounts related to leases not yet on the balance sheet is 113 percent of the hyperscalers’ most recent adjusted debt,” Moody’s report said, “we foresee a material increase in adjusted debt and lease-related cash outflows for these companies in the coming years.”

“While we expect this will be mitigated by an increase in earnings, we recognize there is significant uncertainty with respect to the growth and profitability of the AI market. While leasing the assets reduces the companies’ upfront capital investments, having a significant amount of leases will also reduce the companies’ financial and operating flexibility, particularly if there is a rapid change in industry conditions. Because the data centers are not owned, they cannot be sold or pledged to support additional borrowing.”

These short-term leases cover assets under construction or close to commencing construction, which hyperscalers will not take control of for at least two to three years. Moody’s said that the decision to extend such leases will depend on the hyperscalers willingness to make additional investments in hardware, as key technological components for use in these data centers have a useful lifespan of four to six years.

This means that the decision to renew the lease associated with such a data center is a “much more material” capital investment decision, made at a time when the company’s AI strategy may be markedly different.

“All this uncertainty around lease term renewals should justifiably cause investors to wonder how the lessor (and its creditors) are comfortable with the construction of such a large, specialized asset, when the related lease term is uncertain,” the report said.