Operating vs finance lease

Both are leases, both give possession, both are documented similarly. The substance is in who ends up carrying each element of ownership — and every downstream difference falls out of that allocation.

The classification is not a label chosen at signing. It is a conclusion drawn from what the documents actually do, and the tests all point at the same question: at the end of this arrangement, who was always going to end up with the equipment?

If the answer is the lessor, the arrangement is a rental of part of the asset's life. Payments cover that part plus a return, the lessor keeps a residual position, and the user's exposure ends when the term does. If the answer is the user, the payments must have covered essentially the whole value of the asset, the lessor is not exposed to what it is worth afterwards, and calling the arrangement a lease describes its form rather than its substance.

Everything else in the table follows mechanically. A structure that recovers full value must run for most of the asset's life, so the term row falls out. A lessor with no residual position has nothing to lose from obsolescence, so the technology-risk row falls out. A user certain to acquire the asset needs no option priced at value, so the option row falls out — which is why the end-of-term mechanic is the single clearest tell, and why it gets its own page in FMV vs dollar-out.

The last row is the one worth sitting with. On accelerators the willingness of a third party to write an operating lease is information. It means a party whose business is remarketing this equipment is prepared to take a position on what it will be worth. Where only full-payout terms are available, the market has declined to take that position — and a sponsor who proceeds anyway is taking it instead, usually without pricing it.

The mechanics of both structures, including what to read closely in each, are set out on GPU lease structures. This page assumes them.

An operating lease wins under conditions that are all versions of the same statement: the user's need for this equipment has an end date, and the user would rather not own what is left.

Demand is contracted and finite. A compute contract with a defined term and no committed follow-on is the textbook case. Matching the lease term to it leaves the user with no equipment to house, insure or remarket the day the customer leaves.

The user cannot price obsolescence. Handing the residual to a party that can is not a cost, it is a purchase — and it is usually the cheaper of the two mistakes available. A user who keeps the residual because the payments looked lower has bought a position it has no method of valuing.

The compute layer has to stay separable. Where the site, shell and generation equipment sit in long-dated structures, an operating lease is the cleanest way to keep the shortest-lived layer out of them. This is the structural argument, and it is why the tenor mismatch page treats leasing as one answer to a structuring problem rather than a financing preference.

Flexibility has real value. Where the workload mix is genuinely uncertain, the ability to return equipment and re-specify at the end of a term is worth paying for. That optionality is what the higher nominal cost buys.

It is the wrong answer where the user knows it will keep the equipment. Paying a lessor to hold a residual the user then buys back at value is paying for an option it always intended to exercise. It is also the wrong answer where the return conditions are onerous enough to swallow the benefit — return standards, de-installation, freight and excess-wear definitions are where an attractive rate is recovered, and they should be read before the payment schedule, not after.

A finance lease wins where the user was always going to end up with the equipment, and the question is only how the purchase is funded.

The user intends to run the asset for its productive life. An operator with durable utilisation, a redeployment path and a view on the product cycle is describing ownership. Structuring it as a finance lease is a funding decision, and a reasonable one — it spreads the payment and can be arranged quickly through vendor and equipment channels.

No third party will underwrite a residual. Common on older generations and less liquid configurations. Where the market will only write full-payout terms, the choice between the two structures is not really available, and the honest response is to treat the arrangement as debt-funded ownership and evaluate it that way.

Certainty is worth more than flexibility. A fixed schedule to a known end state removes negotiation risk at term end. For a user with no intention of returning anything, that is a genuine benefit rather than a constraint.

The user wants the upside. If values hold better than the market expects, the finance-lease user keeps the difference. That is a position, and a user taking it deliberately is on solid ground. A user taking it because the structure was presented as cheaper is not.

It is the wrong answer wherever the demand behind the equipment is shorter than the asset's life. A finance lease over accelerators supporting a contract that ends well before the schedule does leaves the user with a continuing obligation and no customer — the exposure that makes term matching the first thing to check. It is also the wrong answer where the sponsor is leasing specifically to keep the asset off its balance sheet, because a finance lease does not do that and signing one for that reason buys the cost without the effect.

The prior question — whether to hold the equipment at all — is on lease vs own compute.

Structures are not always presented under the name their substance supports, and the distinction is worth establishing from the documents rather than the term sheet heading.

Start at the end of the term. What are the user's options, and what do they cost? An option to buy at whatever the equipment is then worth leaves the residual with the lessor. An option to buy for a token amount does not, and never did.

Compare total payments against the asset's value. A schedule that recovers substantially the whole cost is a full-payout structure whatever it is called. This is the most reliable single test and it needs no legal opinion.

Check the term against economic life, not against nameplate life. On accelerators these differ sharply. Economic life ends when newer parts deliver enough more per unit of power and space that older ones stop clearing their operating cost, and that point arrives well before the hardware stops working — the argument is made on GPU residual value and depreciation. A three-year term can be substantially all of an asset's economic life even when the equipment will physically run for six.

Read the return conditions as economics. In a genuine operating lease they are the lessor's protection of its residual and they will be specific. Where they are vague or unusually easy, ask why the lessor is indifferent — it may be because it does not expect the equipment back.

Ask who is insuring and maintaining what. Obligations that look like ownership usually are.

The accounting treatment is a consequence of these answers, and the auditors will reach it independently. Establishing the substance first means the treatment is a confirmation rather than a surprise — and it means the sponsor negotiated the risk allocation it wanted, rather than the one implied by a structure it accepted on its label.

Where the perimeter of the wider financing is also open, the two decisions interact: a portfolio structure can absorb residual exposure that a single-asset one cannot. That interaction is examined in single-asset vs portfolio financing.