Most GPU data center investment means buying a slice of someone else’s balance sheet. This is the other kind. You finance and own the physical edge facility outright. SovereignOne specifies it, builds it, commissions it, and then operates it for you under contract. The asset sits on your books. The pager sits with us.
Those are CBRE’s figures for the first half of 2026. Read them as an investor rather than as an operator: new supply is not competing for tenants, it is being spoken for while it is still a drawing.
CBRE’s own conclusion is that power availability and infrastructure delivery timelines remain the most decisive factors in where capacity lands, and that local approval has become as critical to site selection as power and fiber.
So the scarce asset is not the GPU. It is an energised site, with local approval, finished inside a window that still matters to the tenant. That is what a GPU data center investment is really buying.
A SovereignOne facility is a finished modular data center delivered to your site — not a construction project you supervise. It arrives built, wired, cooled and tested. Five lengths, one density:
| Length | Racks | Capacity |
|---|---|---|
| 7 ft | 1 | 125 kW |
| 12 ft | 2 | 250 kW |
| 20 ft | 5 | 625 kW |
| 40 ft | 10 | 1.25 MW |
| 52 ft | 14 | 1.75 MW |
Capacities are nominal per unit at 125 kW per rack.
Units combine, so a site grows by adding modules rather than by rebuilding. A pad that starts with one 40-foot unit can carry several without a redesign.
Cooling is specified per deployment rather than fixed by the shell: direct-to-chip liquid, rear-door heat exchangers, immersion, or JouleForce molecular beam cooling from Forced Physics. What the tenant’s hardware needs decides it.
You bring the site, or we assess candidates against the three things that actually decide it: available power, a ground path, and local approval. Most failed sites fail here, before anything is ordered.
Configuration follows the tenant. Rack count, density and cooling method are set by what will run in the unit, not by a catalogue tier you pick from.
The unit is built and commissioned off site while your site works — pad, utility tie-in, fibre — proceed in parallel. That parallelism is where the schedule comes from.
Total time is four to six months, against two to five years for a conventional build.
Ownership without an operator is a liability, not an asset. The operations agreement is what turns a GPU data center investment into a turnkey one rather than a box on a pad.
Per-megawatt benchmarks are built for hyperscale and they mislead badly below a megawatt, where a large share of the cost stops scaling with capacity. Before committing to a GPU data center investment, ask instead for:
We work through why each one matters, and where the money hides in a quote, in the Resources library.
You do. A GPU data center investment here is an outright purchase: you finance and own the facility, and SovereignOne’s position is the operations agreement, not an interest in the asset.
The facility is yours, so the operating agreement is a contract you can let run out. The unit is not keyed to us, and the monitoring and access handover is written into the agreement rather than negotiated at the end of it.
Yes. It is a container, not a building. Relocation means disconnection, transport and re-energisation at the new site — slower than a move-in, far faster than a rebuild. That optionality is part of what you are buying.
Land, utility upgrades beyond the tie-in, and the tenant’s IT hardware. The rack counts and capacities above describe the facility, not the compute installed in it.
Any country, provided no US Government export restriction applies to that deployment.
Send the site, the power you have available, and the offtake you are underwriting. We will tell you which configuration fits, what the site works look like, and what the schedule actually is.
Nothing on this page is an offer of securities, an investment recommendation, or a projection of returns. Capacities shown are nominal equipment specifications.