Capgemini Research Institute reported that AI-driven data centers are making electricity demand harder to forecast. The study found rising demand spikes, phantom load requests and grid constraints. Utilities and data-center operators are increasing focus on AI analytics, on-site power and BESS. Storage is positioned as a key support technology for reliability and demand flexibility.
PROJECT SNAPSHOT
Report Name – AI meets the grid: shaping the data center power play
Published By – Capgemini Research Institute
Research Type – Global industry research report
Research Focus – Impact of AI-driven data center electricity demand on grid planning, BESS adoption, and power system reliability
Coverage – 21 countries across North America, Europe, APAC, and Latin America
Methodology – Survey of 612 senior electricity executives and 175 senior data center executives
Survey Period – January 2026
Key Technologies Covered – Artificial Intelligence (AI), BESS, behind-the-meter (BTM) power, AI-based grid analytics, Small Modular Reactors (SMRs)
Key Findings
Demand forecasting – 77% struggle to predict future electricity demand.
Phantom load requests – 67% report data-center load requests that may not materialize.
Supply pressure – 68% expect shortages as AI demand grows faster than supply.
AI power use – AI workloads may rise from 25% to 60% of data-center electricity demand in 3–5 years.
BESS role – 73% are investing in BESS to support grid flexibility.
On-site power – Over 70% expect on-site power to reduce grid reliance within five years.
Purpose - To assess how AI is changing electricity demand and grid planning.
Industry Implication - BESS, AI grid tools, and diversified energy sources are becoming critical for managing volatile AI data-center demand.
Organizations Surveyed - Electricity utilities and data-center operators with revenues above USD 500 million and USD 250 million.
Capgemini Research Institute reported that AI data centers are raising electricity use and making demand forecasts less reliable. The report said changing AI workloads are creating planning challenges for utilities and data-center operators.
The Institute derived its conclusions from surveys conducted among 612 senior electricity executives in organizations earning more than USD 500 million annually. In addition, another 175 responses were collected from senior executives of data-center-owning and operating organizations earning more than USD 250 million per year. The research was conducted in January 2026.
Electricity executives said uncertain demand is now a key planning problem. Around 67% reported “phantom” data-center load requests, meaning proposed electricity demand from data centers that does not always become real consumption. About 19% of requested capacity never moved ahead, making grid planning harder and raising the risk of building too much or too little infrastructure.
Utilities must determine where and when new grid capacity should be developed while avoiding stranded assets. Data-center operators also face planning risks because infrastructure decisions depend on demand forecasts, grid availability and connection schedules.
About 77% of surveyed electricity executives said forecasting future demand has become harder. The report linked the difficulty to changing AI power-use patterns, which can vary more than traditional electricity demand. About 68% of respondents also expect electricity shortages because data-center demand is growing faster than available supply expansion.
Geographic concentration of data centers is creating additional pressure on electricity networks. More than half of electricity executives identified concentrated electricity demand as a challenge for reliable service because clusters of high-density facilities create localized grid bottlenecks that complicate infrastructure planning and system stability.
Capgemini said AI is increasing electricity demand while also supporting grid planning and reliability. Electricity consumption from AI training and inferencing is expected to increase from 25% to 60% of total data-center electricity demand during the next three to five years, replacing other information technology workloads. Around six in ten electricity executives expect advanced AI analytics to improve failure reduction, operational productivity, outage prevention and service restoration by more than 10%.
Current adoption of AI for electricity network management remains limited. Only 45% of surveyed organizations currently use AI for grid optimization, while 16% have implemented advanced AI-driven approaches for optimizing power flows, improving resilience and supporting real-time system performance. Capgemini said faster grid modernization supported by AI and climate technology will be necessary because electricity network construction timelines remain a constraint for meeting future demand.
Data-center operators are also changing electricity supply strategies in response to grid constraints. Nearly three in ten respondents already deploy on-site power generation, while 39% plan to add on-site or behind-the-meter power systems during the next one to two years. More than seven in ten respondents expect on-site solutions to reduce dependence on electricity grids within five years. About 86% consider independent operation from electricity networks a competitive advantage.
Research also found broad support for diversified electricity sources. Around 78% of electricity executives and 73% of data-center executives said renewable energy alone cannot currently provide continuous electricity supply for large AI workloads. Both groups reported active investment in BESS to help address reliability requirements. More than 68% of electricity and data-center executives also viewed natural gas as a transitional option until renewable generation and storage technologies become available at larger scale, while small modular nuclear reactors were identified as a longer-term option requiring additional deployment time.
Capgemini said future electricity planning will require closer alignment between infrastructure investment, energy sourcing and AI-enabled operations as electricity demand becomes more uncertain and system complexity continues to increase.