Most of the attention in data centre power goes to hyperscale campuses measured in hundreds of megawatts. They are also the projects most exposed to grid constraints, because only transmission-level connections can serve them.
A large and growing share of new capacity is much smaller. Edge facilities serving local latency-sensitive workloads, AI inference clusters, enterprise and sovereign data centres, and colocation sites in secondary cities typically need 1 to 20 MW. At that scale, the supply options look very different.
The 1–20 MW segment
Smaller facilities share a few characteristics:
- They are modular. Prefabricated halls and power/cooling skids, deployed in phases.
- They are location-flexible. They need fibre and proximity to users, but not a specific transmission node.
- They are schedule-driven. Customer contracts often start within 12–24 months of signing.
- They are sensitive to connection cost. A network upgrade that is marginal for a 300 MW campus can sink a 5 MW project.
For these projects, waiting in a connection queue behind much larger loads is the worst of both worlds.
Co-location: bring the data centre to the power
The alternative is to plan the facility and its power supply as one site. Rather than finding land with spare grid capacity, the developer finds land that works for fibre, customers and planning — and builds a dedicated power plant next to it.
CNP supplies this as modular generation of 1–20 MW per site. The plant sits on an adjacent pad, connects directly into the facility's switchgear and grows in blocks as the data halls do.
Designed to run off-grid
Co-location only works if the power supply can stand on its own. CNP plants for data centres are designed to run fully off-grid, with:
- Full backup and redundancy in generating capacity, so maintenance or a fault in one block does not reduce supply.
- Battery and UPS integration for seamless transfer between sources.
- Operations and maintenance by CNP, with continuous remote monitoring.
On this basis CNP offers contractual availability of up to 99.999% at the point of supply. A grid connection can be added later for export or as an additional backup path, but the facility does not depend on it to go live.
What changes in site selection
When power is co-located, the site selection checklist changes:
| Grid-led siting | Co-located siting |
|---|---|
| Spare capacity at a nearby substation | Space for a generation pad next to the building |
| Connection offer and timeline | Equipment delivery and site works |
| Network upgrade contribution | Energy agreement with the power supplier |
| Limited to grid-rich locations | Can follow fibre, customers and cheaper land |
This widens the map. Industrial estates, business parks and regional towns with good fibre but a constrained local network become viable.
What a co-located site looks like
A typical co-located arrangement has three zones on adjacent land:
- The data halls — prefabricated or conventionally built, with their own cooling and UPS.
- The power campus — modular generation blocks, switchgear, controls and storage, sited alongside.
- Shared services — access roads, security, fibre routes, and a single operations and monitoring centre.
Because the power campus is dedicated to the data centre, it can be designed around the facility's electrical architecture from the start, rather than adapted to a utility connection point.
Phasing capacity with tenants
Colocation and enterprise facilities rarely fill on day one. Co-location lets power follow the tenants:
- Phase 1: first hall energised with its own generation plus redundancy.
- Phase 2 onward: generation blocks added as each hall is fitted out and contracted.
- Throughout: the switchgear and site layout are designed for the final capacity, so expansion does not interrupt running halls.
Capital — or, under Energy-as-a-Service, contracted capacity — scales with revenue rather than running years ahead of it.
Heat as a resource
Onsite generation produces recoverable heat. For a data centre this opens options that a grid connection does not, such as absorption cooling that uses recovered heat to offset part of the cooling load, or supplying heat to neighbouring users. Whether these make sense depends on the site, but they are worth considering at the design stage.
Owning the power, or buying it
Developers can choose how to hold the power plant:
- Energy-as-a-Service. CNP finances, builds, owns and operates the plant and sells the power under a long-term agreement. No capital from the data centre.
- Customer-owned. The developer owns the plant; CNP delivers it and provides long-term operations support.
- Joint development. For portfolios or larger strategic sites.
Most developers choose Energy-as-a-Service so their capital stays in the IT and building infrastructure that earns their return.
Next step
If you are planning a 1–20 MW facility — or have a site that cannot get a connection on time — talk to us about co-located power. See also Data centre power.
