SCALE / 01
≈15 kW
One server
Private inference, evaluation, or an early production workload
Equipment nameplate only—the room also needs supporting power and cooling.
Sovereign AI, in plain English
How much infrastructure does sovereign AI actually need?
Sovereignty starts with a deployable model and hardware control—not megawatts. One server can support useful private inference. UNICANN's primary audience is AI and enterprise buyers evaluating a fully converted U.S. cannabis cultivation facility with a 1.5–5 MW post-conversion facility input at the service point, for sustained multi-rack demand. Owners of powered facilities can explore the owner path.
Vendor-hosted model: the provider supplies the compute and power.
Deployable model: you choose where the controlled hardware runs.
Find the right scale
A model can start in kilowatts. A business can grow into megawatts.
UNICANN's primary audience is AI and enterprise buyers assessing a 1.5–5 MW post-conversion facility-input target at the service point. Megawatts are not required to run your first model; they matter when sustained demand becomes an infrastructure operation. Owners of powered cultivation facilities can explore the owner path.
SCALE / 02
60–230 kW
One production rack
A dense dedicated cluster with rack networking and storage
Equipment range depends on the selected vendor architecture.
SCALE / 03
1.5–5 MW
Dedicated node
Many racks, support systems, operating margin, and resilience
Post-conversion whole-facility input at the service point—not the power draw of one model.
Decision point
Do you already have a sustained multi-rack workload?
If not, start with cloud or a ready colocation rack. If yes—and model rights and control requirements are documented—use the advanced planner to test a specific site.
This link does not send any planner values.
Planning scale only. Equipment nameplate, utility-deliverable power, and whole-facility draw are different measurements.