A GPU lease starts with the equipment you control

IFSCA's August 18 consultation proposes treating operating leases, including specified hybrid leases, of GPUs and connected data-centre equipment as a financial product. It is a proposal inviting public comment, not a completed regulatory change.

The useful boundary appears in paragraph nine of the consultation paper. It distinguishes leasing identified equipment under a customer's control from purchasing compute capacity from a shared pool. The proposal addresses the former.

My procurement rule follows that distinction: before discussing a monthly GPU price, identify what the customer controls and what happens when the equipment's economic usefulness changes.

Similar invoices can describe different obligations

As an illustration, I would bring two hypothetical offers to a review. One gives the customer control over an identified set of equipment for a defined term. The other promises a quantity of compute capacity, with the provider managing the pool used to deliver it.

Both may arrive as monthly invoices. That resemblance does not make their operational or financial consequences identical. In the first, the team needs to understand the equipment, its location, availability and contractual return conditions. In the second, the service commitment and the provider's freedom to deliver capacity become central.

I would not decide accounting classification from these short descriptions. The actual arrangement needs assessment by the people responsible for the applicable accounting and legal treatment. A product being called a lease, or a regulatory category being proposed, does not settle every customer's balance-sheet question.

What the engineering team can contribute is a precise description of the intended use. Which workloads require the equipment? What performance assumptions make the commitment sensible? Which changes could make the capacity unsuitable before the term ends?

Those answers are more useful to the contract review than a generic statement that AI demand will grow. Demand for compute does not guarantee demand for this equipment, in this configuration, for this customer.

Price the exit as carefully as the entry

The paper presents leasing as a way to address high upfront costs and obsolescence risk. How those risks are divided in a particular deal still depends on its terms. I would ask who bears the consequences of an early refresh, reduced demand or a lower resale value.

A hypothetical model-serving workload might become cheaper to operate on a different accelerator, or shrink after a product change. The lease can remain payable even when the original business case weakens. The relevant question is what the contract permits the customer to change, return, transfer or replace, and at what cost.

This is where the infrastructure plan and financing proposal need to meet. A refresh assumption that exists only in an engineering slide cannot protect a customer whose contract makes the change expensive. Equally, a favourable return clause is of little use if moving the workload requires months of integration work.

Before comparing offers this week, write a short scenario in which the workload changes halfway through the term. Ask each provider to explain the contractual and operational response. Have finance and legal evaluate the answer alongside the technical team.

The consultation is worth following because it makes an emerging financing route more explicit. Its immediate lesson for practitioners is a question they can already ask without waiting for the proposal to become final.

For GPU financing, name the controlled asset and the exit obligation before celebrating the monthly price.