GPU as a service (GPUaaS) vs leasing: which fits enterprise AI?

Simon Harrsen - EVP North America Published October 14, 2026  ·  8 min read

Most enterprises running AI are not AI companies. They are manufacturers, banks, insurers, and health systems. They need GPU capacity for a defined set of workloads, and they often need it on premises. For them, the hard question is how to pay for that capacity, and who carries the risk as the hardware ages.

This article compares GPU as a service with leasing. It also covers the OEM consumption programs that sit between the two. It is part of our guide to AI infrastructure financing. One note on where we stand. CHG-MERIDIAN leases hardware. We do not offer GPU as a service. We write this comparison from the lessor's side, and we say where a service model is the better fit.

What is GPU as a service (GPUaaS)?

GPU as a service, or GPUaaS, is a way to use GPU servers without holding them. A provider owns the servers and runs them, usually in its own data center. You pay for access.

There are two ways to pay. On demand, you pay for each GPU hour you use. With reserved capacity, you commit to a fixed period and get a lower rate.

Three things define the model. You never hold the hardware. The provider runs the facility, including power, cooling, and repairs. And when your term ends, access stops. There is nothing to return and nothing to keep.

Providers range from the large public clouds to specialized GPU clouds, often called neoclouds. It is a fast-growing market. Gartner forecasts that spending on AI-optimized infrastructure as a service will reach $42 billion in 2026.

One warning about the name. Some hardware vendors use “GPU as a service” for something different: metered hardware installed in the customer's own data center. Lenovo TruScale GPUaaS is one example. In this article, we treat that on-premises version as an OEM consumption program, because it works like one.

GPU as a service vs leasing: what is the difference?

The difference between GPU as a service and leasing is what you pay for. With GPUaaS, you pay for access to compute. With a lease, you pay to use specific physical hardware.

The other differences follow from that one. Start with location. GPUaaS runs in the provider's data center. Leased hardware runs in yours, or in your colocation space.

Next, operations. The provider operates GPUaaS. Your team operates leased hardware.

Last, billing. GPUaaS is billed by the hour or as reserved capacity. A lease is a fixed payment over a set term. For GPU servers, that term is typically 24 to 36 months.

The word “leasing” can confuse things. Some cloud providers say you can “lease a GPU” when they mean hourly access. That is not an equipment lease. An equipment lease covers complete server systems that sit in your own racks.

GPUaaS, OEM consumption, lease, or purchase: how do the four models compare?

Enterprises have four ways to get GPU capacity: GPU as a service, an OEM consumption program, a lease, or a purchase. The table compares them on six points that matter to IT and finance.

When does GPU as a service fit?

GPU as a service fits when demand is variable, short-lived, or not yet understood. Maybe you are testing a use case. Maybe you run training jobs only now and then, or you need to cover a temporary peak. In those cases, paying by the hour matches your cost to your actual use. You commit little, and you can stop.

It also fits when you lack the facility. Current GPU servers draw far more power per rack than traditional servers. Many need liquid cooling. Uptime Institute's 2026 Global Data Center Survey reports that power demand per rack is climbing, driven in part by AI. If your data center cannot support that, a provider's data center can.

The model gets weaker as your use gets steadier. An hourly rate has to cover the provider's hardware, its idle capacity, its facility, and its profit. Run the same GPUs around the clock for years, and those hourly charges can add up to more than the hardware itself costs.

Reserved capacity lowers the rate. But it is still a commitment of one to three years, and you have nothing left at the end. Deloitte's Tech Trends 2026 describes the same split. Leading organizations use public cloud for experiments and elastic training. They use private infrastructure for predictable, high-volume inference.

Data location is the other limit. Regulation, contracts, or internal policy may require certain data to stay on premises. For those workloads, GPUaaS in a provider's data center is not an option.

Where do OEM consumption programs such as HPE GreenLake and Dell APEX fit?

OEM consumption programs put vendor-owned hardware in your data center and bill it as a service. HPE GreenLake, Dell APEX, and Lenovo TruScale are the main examples. They solve the data location problem that GPUaaS cannot. That makes them the closest alternative to a lease.

The programs work in a similar way. The vendor installs the equipment in your data center or colocation space. The vendor keeps ownership and meters your usage. You usually commit to a baseline over a multi-year term, and you pay more when your use rises above it. Extra capacity often sits on site, ready to switch on. Support is included.

These programs fit when you want one vendor to answer for the whole stack. They also fit when you want room to grow into capacity that is already installed.

There are three trade-offs to read closely. First, the program is built around one vendor's hardware. Second, the committed baseline is a fixed cost, even though the pricing sounds variable. Third, hardware, financing, and service are rolled into one rate. That makes the price hard to compare with other options.

One point for finance teams. Calling a contract “as a service” does not settle the accounting. If the contract gives you control of specific hardware, it can contain a lease under ASC 842, whatever the program is called. Confirm the treatment with your auditors.

When does leasing on-premises GPU hardware fit?

Leasing fits when three things are true. The workload is steady. The data must stay on premises. And you have the facility and the people to run the hardware. That describes most production AI in data-sensitive enterprises. Think of inference that runs all day, fine-tuning on proprietary data, and models tied to core operations.

A lease works like this. You choose the servers. A lessor buys them. You make a fixed payment over the term, and that payment does not change with usage. That is the trade. You get a known cost and full control of the hardware. In return, keeping it busy is your job.

An independent lessor is not tied to one manufacturer. One contract can cover GPU servers from one vendor, storage from a second, and networking from a third. That matters, because AI clusters rarely come from a single vendor. CHG-MERIDIAN works this way when leasing servers and data center equipment.

Leasing does not fit every case. If utilization is low or hard to predict, you pay fixed amounts for idle hardware. And if your data center cannot power or cool the systems, the hardware has nowhere to go.

Who holds the residual value risk, and what happens at end of term?

Residual value is what hardware is still worth when you are done with it. Residual value risk is the chance that it ends up worth less than planned. With GPUs that chance is high, because each new generation lowers the value of the one before it.

Every model hands that risk to someone. In GPU as a service, the provider carries it. In an OEM consumption program, the vendor carries it. Neither does so for free. They build the risk into their rates. In a purchase, you carry it. In a fair market value lease, the lessor carries it.

Here is how a lessor handles it. Before the lease starts, the lessor estimates what the servers will be worth when the term ends. Take a server that costs $400,000. Suppose the lessor expects it to be worth $100,000 after three years. Your payments are based on the $300,000 of value you use, plus the cost of financing. You do not pay for the full $400,000.

Now suppose the estimate is wrong, and the server is worth only $60,000. That $40,000 shortfall belongs to the lessor, not to you. It is also why a lessor's skill at reselling used equipment matters. The better the lessor is at finding buyers, the better the price it can offer you.

The end of the term is where the four models differ most. With GPU as a service, access ends. If you still need capacity, you renew at that day's market rate. With a consumption program, you renew, or the vendor takes the hardware back. With a purchase, the hardware is yours to reuse, sell, or dispose of.

With a lease, you have three choices. You can return the hardware and move to newer equipment. You can extend the lease if the servers still do the job. Or you can buy them.

How to choose: five questions that settle the decision

The choice between GPU as a service and leasing comes down to five questions. They cover your workload, your data, and your exit. Answer them before you compare prices.

1. How steady is the workload? Variable or unproven demand points to GPUaaS. Steady, high utilization points to hardware on premises.

2. Where must the data live? If it cannot leave your environment, GPUaaS is out. That leaves a consumption program, a lease, or a purchase.

3. Can your facility and team run it? If not, GPUaaS or a vendor-supported consumption program fills the gap.

4. One vendor or several? A consumption program follows one vendor's stack. A lease from an independent lessor can cover all of them.

5. What should happen at end of term? If you want a planned refresh, with the choice to return, extend, or buy, that is a lease.

Many enterprises will use more than one model. GPUaaS can cover experiments and peaks. Leased hardware can cover steady production. What matters is that you answer the last question on purpose.

GPU hardware ages faster than the servers you are used to. So before you sign, decide three things: who carries that risk, how long you are committed, and what happens to the equipment at the end. That is a financing and lifecycle decision. Make it deliberately.

About the author Simon Harrsen EVP North America, CHG-MERIDIAN Simon Harrsen is EVP North America at CHG-MERIDIAN, with nearly 15 years of experience in the equipment leasing and technology lifecycle industry. He leads the company's growth and technology lifecycle strategy across the North American market, and has been quoted in and contributed bylines to publications including Monitor Daily, Channel Insider, Channel Dive, and Procurement Magazine, among many others. Connect on LinkedIn › | Email Simon ›