This guide sets out the design sequence we follow as an EPC contractor for data center power, from the load profile to the finished power island.
Stage 1: Write a basis of design
Start with numbers everyone agrees on: IT load per phase, cooling load, power usage effectiveness assumptions, growth steps over the campus life, required availability and the site’s grid position. Add ambient conditions, altitude, water availability, noise limits and the air permit pathway. Every later decision traces back to this document, so it should be signed off by the developer, the operator and the engineer.
AI training halls deserve particular attention, because their load can swing quickly. That affects how the generating units, controls and storage are specified.
Stage 2: Decide the role of on-site generation
| Role | What it means | Design emphasis |
|---|---|---|
| Bridge power | On-site plant runs the campus until the grid connection is ready | Speed to power, relocatable or reusable equipment |
| Permanent prime power | On-site plant is the main source, grid optional | Efficiency, redundancy, fuel security, maintenance access |
| Grid parallel | Plant runs alongside the grid for capacity and resilience | Interconnection, protection, export rules |
| Backup only | Generators start when the grid fails | Start time, step load, fuel storage, testing |
Many campuses move between these roles over time. Designing for the later role from the start avoids tearing out switchgear and foundations when the grid arrives. Our page on behind-the-meter data center EPC covers how contracts handle that change.
Stage 3: Choose the generation technology
For prime and bridge roles at tens to hundreds of megawatts, gas turbines and gas engines are the usual candidates. Gas turbines offer high output from a compact footprint and work well in modular blocks; reciprocating gas engines offer good part-load efficiency and fast starts in smaller unit sizes. Diesel remains common for backup, and battery storage is increasingly used to handle fast load swings and to give the generators time to respond.
For hyperscale campuses in the USA, we offer the New USP&E FX-45 45MW Modular Gas Turbine for the Hyperscale Data Center sector in the USA, built as repeatable blocks so capacity can follow the campus build-out. Details are on the package page on our main site. Remanufactured heavy-duty turbines are another route; one of our references is seven GE Frame 6B units refurbished to zero hours for major US hyperscale data-centre developers.
Stage 4: Set the redundancy model
Decide how many units can be out of service while the campus still runs at full load: N+1 for one unit in maintenance, N+2 for maintenance plus an unplanned trip. Smaller, more numerous units usually make redundancy cheaper, because the spare is a smaller share of the total. The model also shapes the maintenance plan, since planned outages must fit inside the redundancy you have bought.
See N+1 with modular blocks for a worked discussion.
Stage 5: Secure the fuel
Gas-fired plant needs a firm supply contract, a pipeline lateral or a virtual pipeline, metering and a fuel gas conditioning skid matched to the turbine or engine maker’s specification. Many designs add liquid fuel backup for resilience, which brings storage, permits and dual-fuel equipment into scope. The fuel plan often has a longer lead time than the generating units, so it should start in Stage 1.
Stage 6: Integrate the electrical system
The generating units connect through step-up transformers and medium-voltage switchgear to the campus distribution. Protection coordination, earthing, harmonics from large UPS and power electronics, black start capability and the controls that share load between units all need engineering early. Voltage level choices affect cable sizes and costs across the whole site; our guide to industrial voltage levels explains the options.
For a broader view of how we approach this work, see our data center power station design and engineering article.
Common design mistakes
- Sizing the plant for day-one IT load and leaving no space, fuel capacity or switchgear positions for the later halls.
- Treating the power plant as a utility service and designing it after the building, which pushes it into a corner of the site with poor access.
- Choosing a technology before the fuel supply and air permit pathway are confirmed.
- Underestimating the step loads from large cooling plant and the fast swings from AI training clusters.
- Leaving operations and maintenance out of the design reviews, so access for cranes and engine or turbine changes is missing.
Each of these is easy to avoid at concept stage and expensive to fix once foundations are poured. That is the main argument for bringing the power EPC contractor into the design team early rather than tendering a finished drawing set.
Stage 7: Phase the build
Few campuses energise all halls on day one. A phased design installs the first generating blocks, the common fuel and electrical infrastructure sized for the final build, and space for later blocks. That keeps early capital lower and lets the plant grow as the tenant demand is confirmed.
USP&E for data center power
Our US teams in Colorado Springs and Knoxville lead data center power programmes, backed by engineering and manufacturing offices in London, South Africa and China. Across the business, USP&E counts 150+ power projects in 45+ countries and 25,000 MW contracted since 2002.
Offices: Colorado Springs (1755 Telstar Dr, Suite 300), Knoxville (800 S Gay St, Suite 700), London (162-168 Regent Street), Johannesburg (22 Kildoon Rd, Bryanston), Cape Town (31 Brickfield Road, Woodstock) and China (No. 1 Ningzhou Road, Wuxi). Regions: USA | AFRICA | UK | UAE | CHINA. ISO 9001 and ISO 45001 certified.
On-site generation design FAQs
What is on-site generation for a data center?
It is a power plant built on or beside the campus that supplies some or all of the electricity, either as bridge power, permanent prime power, grid-parallel capacity or backup.
Are gas turbines suitable for data center prime power?
Yes, when the design includes enough redundancy, suitable controls and a firm fuel supply. Modular gas turbine blocks are widely considered for campuses that need large capacity before the grid arrives.
How is redundancy designed for on-site data center power?
By choosing how many generating units can be unavailable while the campus still runs at full load, commonly N+1 or N+2, and fitting maintenance outages inside that margin.
Do data centers still need diesel generators with on-site gas plant?
Many designs keep some backup or dual-fuel capability for resilience, but the right mix depends on the fuel supply, permits and the availability target.
When should on-site generation design start?
At concept stage, alongside site selection, because fuel supply, permits and electrical layout have long lead times that can set the campus schedule.
Start your power design early
Share your load profile, site and timeline and our engineers will help you set the basis of design, compare technologies and plan the phasing.
Talk to an engineer through the data center power team, call +27 10 822 2324 or email info@uspeglobal.com. For near-term capacity, see our natural gas turbine inventory.
