The expansion of data centers for Artificial Intelligence is rewriting a rule that once seemed unchangeable in the United States: the 10-15 year leases that have sustained the colocation and hyperscale business for decades. According to a Moody’s Ratings report covered by industry press, hyperscalers are increasingly signing shorter-term leasing agreements, aligned with the lifespan of compute hardware. The agency warns that the implication is a rise in economic risk that isn’t always clearly reflected in financial statements—particularly as AI capacity commitments surge.
The industrial logic behind shorter leases is understandable: leasing reduces initial capital expenditure and offers greater flexibility if technology shifts. However, Moody’s points out that this flexibility can be partly illusory: many contracts include that act as a “parachute” for data center owners and, in certain scenarios, can become a significant potential cost for tenants. In practice, a strategy designed to avoid CAPEX can start to resemble debt—albeit with less visibility for investors.
The concerning mechanism: Residual Value Guarantees
Moody’s highlights a growing element in AI-related contracts: Residual Value Guarantees (RVGs). Simply put, an RVG is a commitment to cover the difference if the asset (for example, a specialized data center) is worth less than an agreed-upon amount when the tenant cancels the lease, terminates early, or chooses not to renew. Economically, it functions to give developers and their financiers confidence in building infrastructure worth billions, even with shorter lease terms.
The tension arises when accounting does not “capture” all this exposure. Under U.S. standards, the recognition of certain liabilities depends on judgments such as whether a renewal is “reasonably certain.” In AI data centers, where hardware cycles are accelerating and strategies can change within a few years, tenants might argue that it’s not “reasonably certain” they will renew. This allows, in some cases, extension periods and associated guarantees to remain off official obligations on the balance sheet, despite having real economic exposure.
Moody’s describes this effect as a “timing delay” in reporting, where official disclosures do not fully reveal “the scale” of long-term economic risk.
$969 billion committed, with $662 billion yet to start
The figure turning this technical debate into a market issue is volume. According to Moody’s, hyperscalers’ commitments in data center leasing grew strongly in 2025, reaching $969 billion in future lease obligations. Over two-thirds—$662 billion—relate to leases that have not yet begun.
Moody’s adds a straightforward but unsettling benchmark: the portion of commitments still off-balance sheets amounts to 113% of the most recent adjusted debt of these groups. Put another way, if a significant part of these contracts ends up as recognized liabilities, leverage ratios and cash flow projections could be materially impacted.
The agency anticipates an increase in adjusted debt and lease-related cash outflows in the coming years. While growth in earnings might mitigate that impact, it issues a caution: uncertainty remains about AI market growth and, most critically, profitability. If the scenario deteriorates, a large leasing portfolio reduces financial and operational flexibility: a leased data center cannot be sold or pledged for additional financing.
Lease now, occupy in two or three years
An element complicating risk assessment is time. Many short-term leases cover assets under construction or about to commence construction. That is, hyperscalers sign today but don’t take control of the asset until two or three years later. By then, demand, technology, GPU pricing, or even corporate strategy could have changed.
Moody’s links this uncertainty to the physical cycle of AI: key components installed in these data centers often have a 4 to 6-year lifecycle. This makes the decision to renew a lease more akin to a major investment choice, as renewal usually involves reinvesting in hardware to stay competitive. When making the decision, the company could be pursuing a very different AI plan from what justified the original contract.
This time gap prompts Moody’s to pose an uncomfortable question—one that also challenges owners: if a lease is short and renewal uncertain, why are lenders and owners comfortable financing such a specialized asset?
An illustrative example: SPVs, short leases, and large guarantees
The debate deepens when financing is structured through Special Purpose Vehicles (SPVs). In these setups, external investors finance and own the data center, while hyperscalers lease it. The agency notes that, from an economic perspective, such long-term leases resemble debt—even if they don’t appear as debt on financial statements.
In this context, the Financial Times cited a case involving Meta and a project in Louisiana, where the residual value guarantee exposure could reach as high as $28 billion. This example illustrates how such risks can remain “hidden” in notes and probability thresholds, rather than appearing directly as liabilities on the balance sheet. Moody’s, based on that info, indicates it will perform its own probability assessments when analyzing creditworthiness of these companies.
What investors and analysts should watch out for
Moody’s warning isn’t just about accounting—it’s an invitation to view the AI cycle through a risk management lens. Even if leases are shortened to match hardware lifespan, but are offset by guarantees and off-balance sheet commitments, the question shifts from “how much CAPEX is saved” to “what future obligations are being assumed, under what conditions, and with what degree of real optionality.”
For investors, transparency is the challenge. For the data center market, the structural challenge is building increasingly specialized AI assets, which depends on confidence in contracts and credit. If that confidence hinges on guarantees not readily visible in primary financial statements, the risk of misunderstandings—and sudden adjustments—grows when conditions change.
Frequently Asked Questions
What financial risks do short-term leases for AI data centers pose?
They reduce upfront investment but can entail significant future commitments (renewals, guarantees, and payments) that limit financial flexibility if AI profitability doesn’t develop as expected.
What is a Residual Value Guarantee (RVG) in a data center contract?
It’s a guarantee where the tenant covers a potential decrease in asset value if they cancel or don’t renew the lease, protecting the landlord and lenders from uncertainty over the lease term.
Why can some of these commitments remain off-balance sheet?
Because under certain accounting standards, if a renewal isn’t “reasonably certain,” extension periods and some guarantees may not be recorded as lease liabilities, creating a mismatch between economic risk and reported liability.
What should companies consider when signing leases for data centers under construction?
The actual timeline (occupancy in 2-3 years), hardware lifespan (4-6 years), renewal scenarios, and associated guarantees (like RVGs), since future decisions might involve reinvestment and could influence the AI strategy.