Capacity and Service Framework

A 24 MW mixed-workload example

This example allocates 24 MW of IT capacity across four workload groups. Rack counts are capacity arithmetic for coordination; they are not a verified floor plan, an equipment order or a server count.

Source basis: September 9, 2026 planning package · Not independently re-underwritten.

24 MW

Total example IT capacity

640

Illustrative rack positions

150 kW

AI-rack planning envelope

Planning data & boundaries

Workload	IT allocation	Planning density	Rack positions
AI / GPU computing	12 MW	150 kW / rack	80
Enterprise CPU computing	8 MW	20 kW / rack	400
Storage / backup	2 MW	25 kW / rack	80
Networking / control	2 MW	25 kW / rack	80
TOTAL	24 MW	Mixed density	640

How the rack arithmetic works

Rack positions equal allocated IT kilowatts divided by the planning kilowatts per rack. For example, 12,000 kW divided by 150 kW produces 80 AI rack positions. These are density envelopes, not guarantees about usable floor space or installed equipment power.

Racks, servers and GPUs are different measures

The number and type of servers inside each rack depend on the selected equipment. Rev B contains a specific hardware example, not a permanent equipment choice for a facility opening years later. Do not call a GPU count a server count. No hardware procurement is authorized or priced here.

Facility design interfaces

Validate point and rolling loads, full coolant-filled mass, rigging paths, clearances, rack connections, busway, coolant distribution and residual air cooling. Reconcile in-rack and shared network/control loads so the same power is not counted twice.

Ownership and recovery

The working boundary places building, electrical/UPS, heat rejection, physical security and facility controls with the facility entity. Tenant compute, storage, active fabric, software and customer backups remain separate unless agreed otherwise. Local storage capacity does not establish off-site recovery.

EPIC CONSTRUCTION MANAGEMENT LLC