---
title: AI Demand Drives Surge in Data Center Energy Consumption
url: https://www.elseif.net/ai-demand-drives-surge-in-data-center-energy-consumption
published: 2026-10-12T13:06:26+00:00
language: en
section: Models
source: https://www.zerounoweb.it/data-center/consumi-data-center-impatto-ai-su-energia/
organizations: Gartner, International Energy Agency
publisher: elseif
---

# AI Demand Drives Surge in Data Center Energy Consumption

Data center energy consumption is expected to rise significantly due to the race for artificial intelligence, according to Gartner. Consumption is projected to reach 565 TWh in 2026, which represents a 26 percent increase compared to 2025.

Energy has become a primary factor in the success of AI initiatives, influenced by graphics processing units, cooling requirements, and new infrastructure. While computing capacity was traditionally treated as an elastic resource, the widespread use of generative AI and GPU based workloads is shifting the focus toward electrical power. Energy consumption is now a strategic budget item and a planning constraint that determines the location, timing, and conditions for building infrastructure.

For chief information officers and chief information security officers, these energy concerns involve service continuity, operating costs, vendor selection, and the ability to scale AI projects without hitting physical limits. Historically, efficiency improvements balanced the growth of digital services, but this equilibrium has broken.

According to the International Energy Agency, data centers consumed approximately 415 TWh in 2024, accounting for about 1.5 percent of global electricity demand. This demand is heavily concentrated in a few geographic hubs, specifically in China, Europe, and the United States, where it is already impacting local grids.

Future projections indicate continued growth. The International Energy Agency estimates consumption will exceed 900 TWh by 2030, more than double current levels. Higher projections from Gartner, which include the full AI capacity ecosystem, suggest consumption could exceed 1,200 TWh.

AI changes the nature of energy use by introducing computational workload peaks that require dedicated infrastructure and clusters capable of full power for long periods. While AI chips show increasing performance per watt, the absolute consumption of individual components rises with each generation. This increases power density per rack, putting pressure on electrical panels and heat disposal systems.

Gartner estimates that by 2026, AI optimized servers will account for 175 TWh of consumption, compared to 195 TWh for conventional servers. AI servers are expected to surpass conventional servers in consumption by 2027.

Energy use is split between training and inference. Training large language models is intensive but time limited, whereas inference grows as AI is adopted across enterprises. This makes the energy cost of AI services structural throughout their entire life cycle.

Most energy is consumed by IT components, including servers, accelerators, and storage, with GPUs dominating AI profiles. Cooling is the second largest energy expense. Because air cooling struggles with AI workloads, there is a shift toward liquid cooling and integrated thermal design.