Self-build, GPU-as-a-Service or a global operator? Choosing a data centre delivery model
There is no single right way to deliver data centre capacity. The best model depends on who holds the risk, how fast you need to move and what you want to own at the end.
Data centres, 4 minute read, 3 November 2026
Once a site is qualified, the next big decision is how the capacity will be delivered. We see three broad models in the market. Each one shares risk and reward differently, and the wrong choice can cost years.
Self-build
The owner or developer funds, designs and builds the facility, usually with specialist design and construction partners, and then leases it or operates it. This model offers the most control and keeps the long-term value with the owner.
- Suits landowners, developers and investors who want to hold infrastructure long term
- Requires significant capital and strong delivery partners
- The investment case usually depends on securing an off-taker early
GPU-as-a-Service
Compute capacity, typically for AI training and inference, is provided as a service. The customer pays for capacity rather than carrying the whole build. Providers range from specialist AI clouds to infrastructure investors who fund the hardware and housing.
- Suits AI workloads that need capacity quickly and flexibly
- Shifts hardware and refresh risk away from the end customer
- Depends on power-ready space that can host very dense racks
Partnering with a global operator
An established operator designs, builds and runs the facility, often with existing relationships with hyperscale customers. For a landowner or investor, this can be the fastest route to an operating asset with a proven track record behind it.
- Suits projects where speed and operational track record matter most
- Operators bring standard designs, supply chains and customers
- Commercial terms need careful structuring so each party's interests line up
How to choose
We start with four questions. What do you want to own when the project is complete? How much capital can you commit, and when? How quickly does the capacity need to be live? And who is the most likely end customer? The answers usually point clearly to one model, or to a blend of two.
We work with all three models around the world, so our advice isn't tied to any one of them. Our job is to match the model to the project's scale and timeline, then bring the right partners to the table.