Translate the workload into an IT envelope
Site readiness starts with the equipment that will operate there. List the server configuration, rack layout, electrical input, cooling requirements and network needs. State the intended IT load separately from utility service capacity. Utility power also supports cooling and other facility systems; a service rating is not the same as usable capacity at the racks.
Agree the measurement boundary before using an efficiency assumption. Facility energy and IT energy are different quantities, and performance can vary with load and conditions. Do not convert a megawatt figure into a GPU count without an explicit equipment, cooling and redundancy design reviewed by qualified professionals.
Ask for utility evidence, not a land claim
Request written utility information covering available service, voltage, connection scope and energization milestones. Identify who pays for upgrades, transformers, switchgear, deposits and ongoing demand charges. Clarify interruption or curtailment conditions and the metering point. A landowner's estimate or a nearby substation does not establish an enforceable supply commitment.
The IEA's Energy and AI report identifies grid constraints as a material issue for data-center development. That is context, not proof that a particular site is blocked or ready. Keep the utility's actual project evidence, its date and unresolved conditions visible in the decision brief.
Follow the heat beyond the rack
Verify the server manufacturer's cooling requirements against the proposed facility design. Direct-to-chip cooling still requires heat rejection, compatible fluids and materials, maintenance access and responsibility for each loop. Check residual air cooling, peak ambient conditions, water availability where relevant, and what happens when a component is unavailable.
Uptime Institute's public survey summary cautions against assuming that very high rack densities are ordinary. It is not a certification of any facility. A brochure saying high-density or AI-ready cannot substitute for a reviewed rack-level power and thermal envelope for your exact hardware.
Make external works and acceptance explicit
A modular enclosure does not remove foundations, drainage, access, upstream power or local planning obligations. The Open Compute Project's Swedish Modules specification illustrates those categories, but it is an older design with European references, not a current local approval or a high-density GPU design standard. Your jurisdiction and equipment need their own qualified assessment.
Name an owner for civil works, electrical design, permits, fiber, safety, commissioning and ongoing operation. Ask for carrier feasibility and route diversity rather than relying on carrier logos. Include factory inspection, site acceptance and integrated load and failure testing in the responsibility matrix. Treat unresolved utility or permitting milestones as dependencies, not delivery guarantees. A rental bridge or an existing facility may be the more practical near-term choice while those dependencies are resolved.
Primary sources & further reading
IEA: Energy and AI executive summaryUptime Institute: public 2025 survey summaryOpen Compute Project: Swedish Modules specificationIndependent source information is not a ComputeBrokers partnership, live allocation or approval. Confirm current terms before acting.