EPRI, Epoch AI Joint Report Finds Surging Power Demand from AI Model Training

A new report from the Electric Power Research Institute (EPRI) and research firm Epoch AI is sounding the alarm on the soaring electricity demand from artificial intelligence (AI), predicting a significant challenge for power grids in the coming years. The joint analysis, released today, forecasts that the power required to train a single advanced AI model could exceed 4 gigawatts (GW) by 2030, enough to power millions of homes. This surge is driven by the AI industry’s push for larger, more complex models despite efficiency gains.

The report notes that total U.S. power demand for AI, which includes both training and deployment, is currently estimated at around 5 GW but could balloon to over 50 GW by 2030. This projected demand is equivalent to the total global electricity consumption of all data centers today and represents a rapidly growing slice of overall data center power use. According to Jaime Sevilla, director of Epoch AI, the energy demands for training advanced AI models are doubling annually, a rate that could soon rival the output of some of the largest nuclear power plants.

EPRI and Epoch AI stress that this isn’t just an energy problem; it’s a grid infrastructure problem. To accommodate this rapid growth, power providers and data center developers must collaborate on innovative solutions. The report highlights the need for a “build-to-balance” approach, where new infrastructure is built in tandem with flexible data center designs.

This flexibility is a key focus for EPRI’s DCFlex collaborative, which brings together over 45 companies, including tech giants like Google, Meta, and NVIDIA, to make data centers more dynamic. Instead of being passive consumers, data centers could become assets that improve grid reliability and lower costs by, for example, geographically distributing training workloads or shifting power usage during peak hours. This proactive approach will be critical for accelerating grid connections and ensuring a stable and reliable energy future.

While AI’s power appetite is a major concern, EPRI President and CEO Arshad Mansoor also pointed to the technology’s potential to improve the energy system itself. AI applications are expected to play a crucial role in enhancing grid operations, optimizing energy distribution, and accelerating the integration of clean energy sources. The report underscores a need for continued research and development to manage the energy paradox of AI, a technology that both consumes vast amounts of power and holds the key to a more efficient and sustainable energy future.