Skip to main content
  • News Story

Beyond Power: The New Engineering Reality of AI Data Centers

As AI data centers draw more power, prioritizing safety depends on designing power distribution, protection and cooling as one connected system.

Two workers in a well-lit data center

November 18, 2026

Ahead of the SPS Expo in Nuremberg, Germany, on Nov. 24-26, presenter Dirk Mueller of UL Solutions explains how rising power demand is changing the way AI data centers are built and kept safe. Based in Frankfurt, Mueller is director of principal engineering for UL Solutions’ Energy and Industrial Automation group. He will present on the topic of Data Center Power Infrastructure on Tuesday, Nov. 24, at 2 p.m. in Stage Hall 6.

Key Points

  • By 2027, a single AI rack could draw as much power as 65 homes, which makes delivering that power safely a greater engineering challenge.
  • Emerging 800-volt direct current (DC) architectures change how equipment must be protected and how IT personnel can safely work on it.
  • Liquid cooling is extending beyond servers into power cables, connectors and distribution equipment.

A single rack of AI servers, roughly the size of a large refrigerator, could draw as much power at peak as 65 homes by 2027, according to the International Energy Agency (IEA). A single data center can hold hundreds of these racks.

Worldwide, data centers used around 485 terawatt-hours of electricity in 2025, up 17% in a year. That is about one and a half times the United Kingdom’s total electricity demand in 2025, and the IEA expects global data center electricity consumption to roughly double by 2030.

The immense growth in power demands required by AI changes the job for anyone designing, engineering or running a data center. The proliferation of AI servers within these facilities requires managing the complexities of power distribution, protection and cooling together as a connected system. A change in voltage, for example, changes both how equipment is protected and how staff work.

While data centers have existed for decades, how to manage these new and extraordinary power demands is an emerging consideration for the industry.

Looking at the Whole System

The whole system has to be taken into account when designing a data center and assessing its safety, both in normal operation and when a fault occurs. A comprehensive view is essential because each component may satisfy its applicable requirements but behave differently after it is connected to the rest of the system.

The move to 800-volt direct current (DC) is a clear example. Data centers have traditionally taken in 400- or 480-volt alternating current (AC) and converted it to 48-volt DC at the rack. With a single AI server rack set to draw more power than traditional equipment, 48 volts would require enormous currents and thick copper cabling. 

A higher voltage delivers the same power with less current, like moving water at higher pressure through a thinner pipe. Solid-state transformers help make that possible, as covered by UL 2877, the Outline of Investigation for medium-voltage power supplies. But a higher voltage also changes what happens when something goes wrong and who can safely work on the equipment.

Faults in an 800-volt system behave differently than faults in previous systems. An 800-volt system relies on capacitors, which keep voltage steady as computing demand rises and falls. In a short circuit, they release their stored energy all at once, like a reservoir emptying, creating a very fast, very large surge of current. How big that surge gets depends on how the system is built, so the devices that cut power during a fault must be designed around the actual installation.

The move to higher-voltage DC also affects the people who install and maintain this equipment. In earlier data center designs, much routine work around IT equipment could be performed while systems remained energized. At 800 volts DC, exposure to energized conductors creates significant electric-shock and arc-flash hazards. Maintenance may therefore require de-energization, additional safeguards, specialized training and appropriate protective equipment.

How Standards and Equipment Are Adapting

To enable innovation while keeping people and facilities safe, safety standards must be updated. Many standards were written specifically for lower-voltage designs, so UL Solutions has reviewed more than 60 power-related standards to see where they may need to evolve for 800-volt DC. The work covers product updates, common DC requirements and industry collaboration on open questions.

The heat associated with higher power densities is also pushing liquid cooling beyond the servers where it was first adopted and into power-delivery equipment. Liquid-cooled cables, connectors and busways, the enclosed channels that carry power along a row of racks, are starting to emerge. 

Data center developers are increasingly using prefabricated equipment to integrate power and other infrastructure before it reaches the project site. Open power skids, covered by UL 2140, and prefabricated modular data centers, covered by UL 2755, are evaluated as integrated assemblies rather than solely as collections of individual components.

Where UL Solutions Comes In

As an independent third party, UL Solutions tests and certifies power equipment, and assesses how it performs as part of a complete system, from 800-volt DC components to prefabricated skids and modular data centers.

More information about how UL Solutions supports data centers with safety, security, performance and market access offerings is available on the Data Centers page at UL.com. 

About the Author
Dirk Mueller is director of principal engineering for the Energy and Industrial Automation group at UL Solutions, where he leads the engineering team and provides governance for the company’s simulation offering. He holds a Dipl.-Ing. in electrical engineering (equivalent to an M.S.) from the Technical University of Darmstadt and a business degree from SGD in Darmstadt.