Data center construction
How a data center gets built
A data center is built in overlapping phases (site and power, shell, mechanical and electrical, commissioning, handover), and the long-lead electrical and cooling equipment, not the shell, usually sets the schedule.
Updated 19 November 2026
How long does it take to build a data center?
It depends on the size, the power and the equipment. Securing power and permits comes first and can take a long time; the building itself goes up faster than the specialized equipment inside it can be made and delivered. Published ranges vary widely by project, so plan from the equipment lead times and the power date, not the shell.
What are the phases?
Site selection and power, design, the foundation and shell, the interior (power, cooling, network and the white space), commissioning, and handover. On large projects these overlap: equipment is bought during design, and commissioning starts while the last rooms are still being finished.
Why does the equipment set the schedule?
Data centers use specialized electrical and cooling equipment with long lead times, so it has to be ordered early, often before the design is final. A late change to that equipment, or to the pipes and cables that connect it, moves the whole schedule.
How big are they?
Power ranges from several megawatts for a smaller facility to over 100 MW for a hyperscale campus.
Who are data centers built for?
Data centers are built for different kinds of clients, and each one means something different when it asks for a building fast. Three come up most:
- Hyperscalers. The large cloud service providers that run hyperscale data centers, sites with at least 5,000 servers built for very large workloads. What they ask: when does the next power block or cluster become usable, and who owns its acceptance?
- Colocation providers. They own and run the building and rent space, power and cooling to tenants, who bring and keep their own servers. What they ask: when is the space, power and cooling each tenant contracted for ready to hand over?
- Neoclouds. Cloud companies built for GPU compute and AI work, renting out GPU capacity instead of a broad menu of cloud services. What they ask: when can the installed cluster pass acceptance and serve the workload a customer contracted for?
The types overlap: a hyperscaler or a neocloud may lease space in a colocation building instead of building its own.
What does fast mean to each of them?
Each one is waiting on a different date. A contractor's fabricated date, an owner's energized capacity date and an AI operator's revenue-capable compute date are three different milestones. Speed only matters on the critical path: if utility power or commissioning sets the finish, fabricating racks faster gains nothing. So start by asking which date the client is measured on, then find what actually sets it.
What changes with liquid cooling?
More of the cooling arrives as pipework at the racks: more piping to size, route, coordinate and test. That is where model checks and clash coordination earn their keep.
What is commissioning?
The stepped testing that proves the building works, from factory tests of each machine to a final test of every system running together. Data center commissioning levels, explained.
Where does the model fit?
The model carries every late change into the shop and the field. In our worked example, Lone Star 48 Hall 2, thirty-six cooling branches went from DN 150 to DN 200 late in design; the model check caught every one, and the clash it caused was resolved in the model, before the first weld.
Learn it on one real project
The Data Center Delivery Lab follows one data center from the building to the campus, using the same worked project.
Sources: Autodesk, A Guide to Data Center Construction; IBM, What is a hyperscale data center?; Equinix, What is colocation?; Micron, What is a neocloud?.
See it on your own model.