There are two ways to get a GPU cluster running without building a data center. You can own the hardware and rent the space and power for it. That's colocation. Or you can contract a cluster that an operator owns and runs. That's reserved GPU capacity.
The two options, one line each
- Colocation: you own the servers. A data center gives you the space, power, cooling and security. You run the hardware, or a partner runs it for you.
- Reserved GPU capacity: an operator owns the servers and runs them. You contract a set number of GPUs for a set term and use them over the network.
When colocation fits
- You already own GPUs, or you have them on order.
- You want control of the hardware: what it is, how it's set up and who can touch it.
- You plan to run it for years.
- You have people who can operate it, or a partner who will.
What you take on: buying the hardware, running it, and matching the site to it. The site has to supply enough power per rack and the right kind of cooling, air or liquid, for the servers you own.
When reserved GPU capacity fits
- You need GPUs working without buying or installing anything.
- You don't want to operate hardware.
- You want a particular GPU generation and don't want to manage its delivery and setup.
- What you need is compute, not equipment.
What you take on: a contract for a fixed number of GPUs over a fixed term. Expect to pay for what you reserve, whether you use all of it or not.
What the two have in common
- They are multi-year commitments. On the capacity we broker, minimum terms run 36 to 60 months for both.
- The building matters either way. The GPUs sit in a data center in both cases. Its power, cooling and reliability affect you whether you own the servers or not.
- Both take lead time. Space, power and hardware all have to be ready on the same date. Start the conversation before you need the capacity.
A simple way to decide
Ask yourself three questions.
- Do you own the GPUs, or will you? If yes, you need colocation.
- Do you want to operate hardware? If not, reserved capacity is the simpler path.
- How long will you need it? Either way, plan in years.
Many teams end up doing both: reserved capacity for what they need now, and colocation for hardware they own later.
What to have ready before you ask for quotes
For colocation:
- The market or region
- Power, in kW or MW
- Air or liquid cooling
- When you need it, and for how long
For reserved GPU capacity:
- The GPU type
- The cluster size
- The region
- When you need it, and for how long
The more of this you can state up front, the faster an operator can answer.
Where XIRR fits
XIRR Advisors brokers both: Tier III colocation space and power for hardware you own, and reserved GPU compute capacity (H100/H200/B200/B300/GB200/GB300/Vera Rubin) from neocloud operators. We do not sell GPUs or hardware.
If you're not sure which one you need, tell us what you're trying to run and when. We'll point you to the right side of our network.
To see how the hardware affects the building, read Liquid vs Air Cooling for AI GPUs: Rack Density Explained.
Share your requirements. We'll canvas the market.
Tell us what you need (for colocation space and power: market, kW or MW, cooling and timing; for reserved GPU capacity: region, GPU type, cluster size and timing) and we'll canvas colocation and neocloud operators in parallel. Shortlist in 48 hours. We don't sell hardware. We broker colocation space and power, and reserved GPU compute capacity.
Earlier conversations get better terms. When you engage early, we have time to negotiate with vendors before you need to commit. We're paid only when a deal closes, and the fee structure is agreed in writing before any introduction.