J-BoxPlain answer. In a powered shell, the owner delivers land, power, cooling and core infrastructure and the tenant owns and refreshes the compute; in a turnkey or GPU as a service model, the owner or a partner owns the whole operating platform including the GPUs. The choice sets who carries hardware refresh risk and who captures the upside.
The question underneath the structure
Every AI capacity deal answers one question first: who owns the GPUs. Everything else follows from the answer. Whoever owns the compute carries the refresh cycle, the utilization risk and the upside. Whoever owns the infrastructure carries the long duration asset and the power position.
The three structures are three answers to that question.
From the tenant's seat and the owner's seat
| Powered shell | Turnkey | GPU as a service | |
|---|---|---|---|
| Tenant gets | A powered, cooled hall on a lease; full control of the hardware | A commissioned facility with compute installed, on a lease | Capacity by the rack, the cluster or the hour |
| Tenant's capital | High; it buys the compute | Low to moderate | Lowest; operating expense |
| Tenant's control of refresh | Full | Shared or none | None |
| Owner gets | A durable infrastructure asset on a long lease | A larger asset and a larger lease | Operating revenue |
| Owner's exposure | Power, cooling, hall; no hardware risk | Hardware refresh and residual value | Hardware, utilization, pricing, operations |
| Typical fit | Neoclouds, large AI operators, anyone whose business is the GPUs | Enterprises that want capacity without a capital program | Enterprises and resellers that want elasticity |
Refresh risk is the whole game
GPUs are refreshed on a cycle measured in a few years, and each generation changes power, density and cooling requirements. Whoever owns the hardware owns that cycle: the capital to buy the next generation, the residual value of the last one, and the facility changes required to host it.
Neoclouds generally want that risk because it is their business and their margin. Enterprises generally do not, because compute is a means to an end. Capital partners price it very differently depending on who holds it. A structure that puts refresh risk on the party least equipped to carry it will not survive the first refresh.
What neoclouds typically ask for
A shell. Powered, cooled, commissioned at their density, with a lease long enough to amortize the fit out and flexible enough to expand. They want to own the GPUs, run the cluster and sell the hours. The owner's job is to deliver the hall and the power and to stay out of the compute.
What enterprises typically ask for
Turnkey or GPUaaS. They want the capability without building a program around it. The capital partner or the owner holds the compute; the enterprise pays for capacity. The question for the owner is whether it has an operating partner that can run compute as a business, because that is a different competency from delivering infrastructure.
How JBox structures around the requirement
The tenant's requirement comes first: megawatts, density, cooling, timeline, geography and how it wants to buy. JBox then matches that to a site under documented control and structures the deal with the capital partners at the table. Where the tenant wants a shell, we deliver a shell. Where it wants turnkey or GPUaaS, we bring the compute through our IT ecosystem and the operating partners to run it.
Questions & answers.
No. Colocation is a shared facility built to a standard the operator chose. A powered shell is built or configured to the tenant's density and cooling, on a lease the tenant controls, and the tenant owns the compute.