Energy demand for artificial intelligence in the US could increase tenfold by 2030.

According to the Electric Power Research Institute (EPRI), AI power demand in the United States could increase tenfold by 2030. According to the report , produced in collaboration with AI benchmarking research organization Epoch AI, total AI power demand is currently estimated at 5 GW but could exceed 50 GW by 2030. Rapid advances, particularly the training of large-scale "frontier models," are driving this growth in electricity demand. As the study highlights, AI is the primary driver of near-term data center growth.

The analysis focused on the technical drivers of AI energy consumption, modeling both demand trajectories for individual AI training sites and broader AI needs. By 2030, forecasts suggest that AI, including both training and inference, could consume over 5% of US generating capacity.

The growth in energy demand for AI can be estimated using various approaches, such as analyzing AI chip production projections, the investment plans of major AI companies, or assessments by data center and industry experts. EPRI's analysis examined several such estimates, including its own data center growth forecasts, forecasts from the International Energy Agency, and semiconductor shipment projections. The report takes into account the increased electricity demand for training the largest cutting-edge AI models. Its modeling estimates projected peak power demand for Meta Platforms' planned data center in Louisiana at 2 GW or more by 2030. A project known as Stargate, sponsored by a consortium that may include OpenAI and Microsoft, was expected to reach 5 GW by 2030, but EPRI stated that this project may have been canceled. The report also includes a 1.2 GW Stargate data center campus in Abilene, Texas, by 2026.

The Electric Power Research Institute (EPRI) warned that individual projections do not provide sufficiently robust evidence for future growth, especially at the end of the decade, because this would require a high rate of growth in AI investment. However, if current trends continue, "AI will represent a significant portion of the US energy sector by 2030." According to EPRI, it is not yet clear whether projected energy demand can be met. Although investments in hyperscalers suggest rapid growth, constraints in building transmission and next-generation networks could hinder the growth of data centers, the report states. The analysis suggests that energy demand for AI will likely reach gigawatt-scale training levels by 2028, but beyond that point, scalability is less certain. The rapid increase in energy demand will have societal implications, EPRI stated, with energy growth posing challenges for technology companies that have already committed to clean energy. Energy constraints due to AI demand could also trigger a greater push for energy growth, disrupting traditional planning processes and potentially resulting in environmental consequences.