When an AI operator needs capacity, the usual options sit at two extremes. At one end is retail colocation: you rent racks in someone else's facility, on their design, with their cooling. At the other is building your own data center: full control, but also land, utility negotiations, permits and construction before the first server arrives.
A powered shell sits between the two. The provider supplies a dedicated building and the electrical capacity the deployment needs. The tenant designs and installs everything inside: compute, distribution within the hall, cooling and operations.
Who provides what
In a typical powered shell arrangement, the provider is responsible for the building, the utility service and electrical capacity, distribution up to an agreed handover point, site access and coordination of connectivity. The tenant is responsible for its GPUs and servers, cooling systems, in-hall power distribution, redundancy design and day-to-day operations.
The handover point is where those responsibilities meet. Agreeing it early, for power, cooling services and fiber, is one of the most useful things both sides can do.
Why AI operators consider it
- Control. Dense GPU deployments often need specific cooling, such as direct-to-chip liquid or immersion, that standard colocation halls aren't designed around.
- Capital focus. Money goes into compute rather than industrial real estate.
- A head start. Starting from an existing building can remove steps compared with developing a site from scratch.
Questions to ask any provider
- How much capacity is live today, and how quickly can it scale?
- How is usable IT capacity calculated for your design?
- Where exactly are the handover points?
- Which fiber carriers are confirmed, and are there diverse entry paths?
- What documentation can you share, and when can we visit the site?
We're happy to answer every one of them about our sites. Start with our available locations.