The rapid expansion of artificial intelligence is emerging as a major challenge for electricity networks worldwide, with AI driven data centres expected to trigger a sharp rise in power demand while making it increasingly difficult for utilities to plan future capacity, according to a new report by Capgemini Research Institute.
The report, AI meets the grid: Shaping the data center power play, found that nearly 80% of electricity utilities expect demand patterns to become more volatile and unpredictable as AI workloads grow. More than three quarters of utility executives admitted they are already struggling to accurately forecast future electricity requirements because of the rapid pace of AI adoption.
One of the biggest concerns highlighted in the report is the emergence of “phantom” data centre power requests. Around 67% of electricity executives said they receive power capacity requests that never materialise, with nearly one fifth of these projects eventually abandoned. Such uncertainty raises the risk of utilities either over investing in grid infrastructure or failing to build sufficient capacity where it is genuinely needed.
The report also warns that electricity consumption from AI training and inference is expected to account for 60% of total data centre power demand within the next three to five years, up from 25% today. This shift is expected to replace conventional IT workloads as AI becomes the dominant computing application inside modern data centres.
Grid constraints are already forcing data centre operators to rethink their energy strategies. Nearly 29% of operators have deployed on site power generation, while another 39% plan to install behind the meter power systems over the next two years. More than 70% believe these systems will significantly reduce dependence on the public grid within five years.
Despite the mounting pressure, the report sees AI as part of the solution. About six in ten utility executives expect AI based analytics to improve grid reliability, reduce equipment failures and enhance operational productivity. However, only 16% of electricity companies have implemented advanced AI driven grid optimisation systems, suggesting significant room for adoption.
The study is based on surveys of 612 senior electricity executives and 175 data centre executives across 21 countries, highlighting the growing need for utilities, governments and data centre operators to jointly invest in smarter grids and diversified energy sources to support the next wave of AI infrastructure.
