Data center power requirements

How much power data centres use

At the global level, the International Energy Agency estimates data centres used around 415 terawatt hours (TWh), or about 1.5% of global electricity consumption in 2024. In its Base Case that roughly doubles to around 945 TWh by 2030.

In the United States, a Berkeley Lab report published by the Department of Energy found that data centres consumed about 4.4% of total U.S. electricity in 2023 and could reach 6.7% to 12% by 2028. In energy terms, use climbed from 58 TWh in 2014 to 176 TWh in 2023, and is estimated at 325 to 580 TWh by 2028.

For a single facility, the IEA gives a sense of scale: a typical AI-focused data centre consumes as much electricity as 100,000 households, but the largest ones under construction today will consume 20 times as much.

Where the power goes

The IEA's Energy and AI report breaks a data centre's electricity use into components:

• Servers: Around 60% on average
• Storage systems: Around 5%
• Networking: Up to 5%
• Cooling: About 7% in efficient hyperscale sites to over 30% in less-efficient enterprise sites
• Other infrastructure: UPS batteries, backup generators, lighting and offices

Two points follow. First, the IT load sets the baseline, and everything else scales with it. Second, cooling design is the largest controllable overhead.

How AI changes the load

AI workloads run on accelerated servers. The IEA says the rise of AI is accelerating the deployment of high-performance accelerated servers, leading to greater power density in data centres. In its Base Case, electricity use by accelerated servers grows by 30% a year.

For power planning, higher density means:

• More power per hall. The same building footprint needs a larger electrical supply.
• More heat per rack. Higher density pushes sites toward liquid cooling.
• Faster ramps. AI campuses are often built in large phases, so the power plan must grow in large steps.

Measuring efficiency: PUE

Operators track efficiency with power usage effectiveness, or PUE. It is total facility energy divided by the energy delivered to IT equipment. A PUE of 1.0 would mean every unit of electricity reaches the servers; anything above that is overhead for cooling, power conversion and building services. Lower is better.

Redundancy and backup

Data centres are designed so that a single fault does not take the facility down. That adds to the power requirement:

• N+1 and 2N designs. Spare capacity in power trains and cooling.
• UPS. Uninterruptible power supplies bridge the seconds between a grid fault and backup generation starting.
• Backup generation. On-site generators carry the load through longer grid outages.

The supply contract therefore has to cover peak demand plus headroom, not just the annual average. See data center backup power for more detail.

Securing the supply

For most new sites, the binding constraint is not equipment inside the building. It is getting enough firm power to the fence line. The IEA notes that a data centre can be operational in two to three years, while the broader energy system needs longer lead times. It estimates that around 20% of planned data centre projects could be at risk of delays.

The main options are:

• Grid supply. The lowest-effort option where capacity exists, but often limited by transmission and interconnection timelines.
• On-site or behind-the-meter power. Generation on the campus serving the load directly, often as a bridge until the grid catches up.
• Long-term firm low-carbon supply. Large buyers are signing agreements for new nuclear. See nuclear power for data centers and small modular reactors.

How BridgePower Nuclear meets the requirement

BridgePower Nuclear, a joint venture of USP&E and BAM Energy, delivers power to a site in two phases under one contract and one operator.

First, USP&E supplies thermal power on gas, diesel, HFO or dual-fuel, grid-tied or behind the meter. It is backed by 24+ years in power, 150+ power projects in 45+ countries, 2,000+ MW in stock and 400+ engineers and project managers.

A planning checklist

• What is the IT load at full build-out, and in what phases?
• What PUE does the cooling design target in this climate?
• What redundancy level (N+1, 2N) does the design require?
• When can the grid deliver the full load, and in writing?
• What powers the site until then?
• Is water available for cooling and power generation?
• What is the long-term low-carbon supply plan?

To talk through power for a specific site, contact our team.