THE QUESTION
Picture two companies buying GPU capacity from the same cloud provider. One signs up for a reserved cluster, paying a fixed monthly fee on a one-year term. The other pays as it goes, drawing on-demand capacity with no reservation at all. On paper the two look like variations on the same purchase, yet one of these contracts may need to land on the balance sheet as a lease while the other almost certainly won’t.
Most people start by looking at the word on the invoice, “reserved” or “on-demand”, and assume that settles things, but it doesn’t, at least not on its own. The real test comes down to two things about the asset itself, whether something specific has actually been identified, and who genuinely controls how it gets used once the contract is running. Billing structure barely factors into either one.
The General Rule
A contract contains a lease when it conveys the right to control the use of an identified asset for a period of time in exchange for consideration, and getting there usually means working through three questions in order.
First, does the contract point to a specific asset, a particular server, rack, or cluster, rather than just a category of hardware.
Second, could the supplier actually swap that asset for a different one without much cost or hassle, since a real substitution right like that means nothing has truly been pinned down in the first place.
Third, assuming an asset really has been identified, does the customer get substantially all the economic value from using it and get to decide how it operates, rather than simply how much of it to draw on.
All three need to land in the customer’s court before anything qualifies as a lease, and any one of them falling short turns the arrangement into an ordinary service, regardless of how the invoice describes it.
HOW IT USUALLY GOES
Six contract patterns show up again and again in compute and data center deals. The same provider often runs several of them side by side.
Dedicated cluster, no substitution.
The customer reserves specific racks, and the contract bars the provider from swapping them out.
Dedicated hardware, maintenance-only substitution.
The provider can pull a failed unit or push a generation upgrade, but has no right to move the workload just to rebalance its own fleet.
On-demand, shared pool.
The customer draws capacity from a fungible pool the provider allocates at will.
Reserved capacity, substantive substitution right.
The customer books a guaranteed quantity, but the provider can move the workload to different hardware whenever it chooses.
Committed capacity in a shared facility.
The customer commits to a fixed amount of power or rack space in a data hall it doesn’t otherwise control, without naming specific equipment.
Managed colocation.
The customer’s own equipment sits in the provider’s space, and the provider handles power, cooling, and security.
| Dedicated cluster | Repair swap | Shared pool | Live substitution | Capacity commitment | Colocation | |
|---|---|---|---|---|---|---|
| Identified asset | Yes | Yes | No | No | No | Yes, the space |
| Customer directs use | Yes | Yes | N/A – no identified asset | N/A – no identified asset | N/A – no identified asset | Yes, the space |
| Conclusion | Embedded lease | Embedded lease | Service | Service | Service | Lease of space, service for management |
Figure 1. Six compute and data center patterns, whether each clears the identified asset and control tests, and the conclusion it reaches.
Reserved and on-demand turn out to be commercial labels more than accounting ones, and they don’t map cleanly onto “lease” or “service.” A contract that says reserved can still be a service if the provider keeps a real substitution right, and a dedicated cluster can land on the balance sheet even when the provider keeps an ordinary right to swap out a broken unit. Committed capacity can steer clear of leases for an entirely different reason, since with no substitution right anywhere in sight, a power or rack commitment that isn’t tied to specific, physically distinct equipment usually never clears the identified asset hurdle in the first place.
THE DECIDING FACTOR
Whether a specific asset is identified, and who directs its use.
The answer usually sits buried in the service schedule attached to the master agreement, in the appendix that spells out whether hardware is dedicated or pooled, whether the provider keeps a substitution right, and who actually makes the operational calls once the contract is live. Procurement and infrastructure teams write those terms to manage uptime, capacity, and service levels. But the same terms can decide whether the arrangement is a lease or a service. The technical appendix shapes both the operating model and the accounting fact pattern.
Technical references. ASC 842-10-15-3 on the definition of a lease, 15-9 through 15-14 on identifying an asset, substantive substitution rights, and the maintenance and repair carve-out, 15-16 on a capacity portion of a larger asset that is not physically distinct, and 15-17 through 15-26 on the right to control use, including economic benefits and the right to direct use. IFRS 16 applies substantively identical guidance.
WHERE WE COME IN
Whether a specific asset is identified, and who directs its use.
While the patterns above cover the common structures in cloud and data center contracts, in practice the answer sits in the substitution rights, service schedules, and operational control terms specific to each arrangement. We perform that detailed analysis, and we advise on how compute and infrastructure contracts are structured so the lease conclusion lands where the business intends, documented well enough to survive an audit, a diligence process, and the scrutiny of a registration statement. As a firm with no audit practice, we can give you a view rather than a question.
Start a conversation at https://uniqus.com/contact-us/
ABOUT BY DESIGN
By Design is a Uniqus series on the accounting decisions companies make before they know they are making them. Each issue takes one topic, sets out how the analysis usually runs, and names the single upstream choice that decides the outcome.
