AI’s Appetite for Power Generates Challenges for Data Centers

As organizations of all shapes and sizes rush to implement AI use cases, data centers are finding it difficult to secure enough power to support high-density AI workloads. Due to AI, U.S. data center power demand is projected to exceed 80 GW in 2030, up from 25 GW in 2024, according to McKinsey research.

The expected surge in power demand poses a significant challenge for data centers, as the grid currently lacks sufficient capacity to handle it. Increasing capacity takes time. Utilities and power distributors typically move much more slowly than the expected demand growth. Connecting a new data center to the grid can take years – as much as seven years in Northern Virginia’s Data Center Alley.

This means that data center operators, especially hyperscalers expected to handle most AI workloads, need to be creative in securing the energy required for AI expansion. As a result, some unexpected moves are occurring, with Meta, Google, Amazon and Microsoft looking to source nuclear power for their AI data centers. Microsoft has made a $16 billion deal to source power from Three Mile Island in Pennsylvania starting in 2027.

To maintain grid stability in Data Center Alley and surrounding areas, West Virginia coal-fired plants that were previously scheduled for decommissioning are now expected to operate beyond their planned retirement dates. In Abilene, Texas, developers of the first Stargate data center want to build a 350MW natural gas power plant on-site.

Solutions to the AI Power Challenge 

AI workloads use significantly more energy than traditional computing, requiring high-density chip clusters, GPUs, and specialized hardware to generate text, images, graphics, and video. AI-enabled racks consume up to six times more power than traditional racks and require hybrid cooling approaches to keep things running smoothly. All of this translates to higher power consumption, which necessitates careful infrastructure planning that goes beyond just selecting compute resources.

But with grid constraints, how can data centers secure the power they need? There is no single solution. Sourcing nuclear power and putting off the decommissioning of coal plants provide some relief, but other strategies are necessary.

Some approaches remain aspirational—such as small modular reactors (SMRs), which could be deployed onsite to supply nuclear power to data centers. Other solutions that are already available include:

• Utilizing power from sustainable energy sources such as solar, wind, geothermal and hydroelectric
• Generating power on site with gas turbines, solar and wind systems, or modular fuel cells that use hydrogen to produce power
• Building Battery Energy Storage Systems (BESS) on-site that store renewable power to ease the burden on the grid

Grid modernization is also needed, but that isn’t something data centers can accomplish on their own. However, they can play a role by participating in DCFlex, an initiative by the Electrical Power Research Institute (EPRI) involving utilities, data centers, and technology companies. The goal is to accelerate grid connections by leveraging technology that enhances grid stability and improves asset utilization. 

At the rack level, the first step in designing an AI rack infrastructure is to understand the workload: what level of CPU/GPU performance is required and how many servers are needed. Then, assess whether your existing power infrastructure—busway, RPP, etc.—can handle those demands or if upgrades are necessary. From there, choose an intelligent Server Technology Rack PDU that matches your power density and outlet type requirements, ensuring proper branch protection and failover support. Then evaluate phase-balancing, load monitoring, and circuit-protection needs. Invest in rack PDUs that include environmental sensor support and power-quality analytics at the outlet and at the PDU inlet for better visibility and reliability.

Innovation to Solve AI’s Power Challenge 

AI has the potential to profoundly transform how we do business – and how we live. But that transformation cannot happen without solving the power-sourcing crisis. Addressing this challenge requires innovation, collaboration, and a continued commitment to sustainability from data center operators. It also starts with implementing technology that supports high-density AI workloads and helps drive energy efficiency through effective power management. Learn how Server Technology's intelligent rack PDUs help data centers achieve the right balance with reliable performance and best-in-class features.