Arab Finance: The rapid growth of AI-powered data centers is making electricity demand increasingly volatile and difficult to forecast, complicating grid planning and raising the risk of power shortages, according to a new report from the Capgemini Research Institute.
The report, based on a survey of more than 600 senior electricity executives at organizations with annual revenue exceeding $500 million, found that utilities are entering a new phase as AI workloads drive more unpredictable electricity consumption.
About 77% of electricity executives said they struggle to accurately forecast future demand as AI-related consumption patterns become less stable, while 68% expect power shortages because demand from data centers is growing faster than electricity supply can be expanded.
The study also highlighted a growing disconnect between projected and actual demand, with 67% of electricity executives reporting "phantom" data center load requests. On average, around 19% of those requests never materialize, distorting demand forecasts and increasing the risk of over- or under-investment in grid infrastructure.
The uncertainty is forcing utilities to make difficult decisions about where and when to invest in grid modernization, while hyperscale data center operators face challenges planning infrastructure amid uncertain demand forecasts, grid availability, and connection timelines.
More than half of electricity executives also cited the geographic concentration of data centers as a major obstacle to reliable service, saying clusters of high-density facilities are creating localized bottlenecks that strain grid stability.
"AI is transforming electricity systems far beyond demand growth. It is exposing structural constraints in grid capacity, planning, and power availability, while making demand more dynamic and harder to predict," Claire Gauthier, Global Head of Energy & Utilities at Capgemini, said in the report.
She added that utilities will play a key role in balancing grid resources with customer-owned assets using AI-enabled insights to support future data center expansion.
The report projects that electricity used for AI training and inference will increase from about 25% of total data center electricity demand to 60% over the next three to five years, largely replacing other IT workloads.
At the same time, respondents said AI could improve grid operations. Around six in 10 electricity executives expect advanced AI analytics to deliver more than 10% gains in areas including failure reduction, operational productivity, and outage prevention and restoration.
Despite those expectations, AI adoption remains relatively limited. Only 45% of utilities surveyed said they currently use AI for grid optimization, while just 16% have implemented more advanced AI-driven systems to optimize power flows, improve resilience, and enhance real-time grid performance.
The report said lengthy grid infrastructure construction timelines remain a major constraint on accommodating AI-driven demand, underscoring the need to accelerate grid modernization through AI and climate technologies.
Grid constraints are also prompting data centers to invest in on-site power generation. Nearly 30% of respondents said they already use on-site power solutions, while 39% plan to deploy behind-the-meter or near-site generation within the next one to two years. More than 70% expect such systems to significantly reduce reliance on the grid within five years, and 86% view the ability to operate independently from the grid as a competitive advantage.
The report also found broad agreement that a diversified energy mix will be needed to support reliable data center growth. Around 78% of electricity executives and 73% of data center executives said renewable energy alone cannot yet provide continuous power at the scale required for large AI workloads, with many organizations investing in battery energy storage systems.
More than two-thirds of respondents also identified natural gas as a near-term transitional energy source until renewable generation and energy storage can be deployed at sufficient scale, while acknowledging the tension with long-term decarbonization goals.
