The rapid growth of AI-powered data centers is not only increasing demand for electricity, but is also making it significantly more difficult to predict, challenging the way electrical systems are planned and managed. The vast majority of energy sector managers expect more extreme and unpredictable peaks in electricity demand, while more than three-quarters admit to having difficulty accurately forecasting future needs, according to the latest report from the Capgemini Research Institute, AI meets the grid: shaping the data center power play.
The study, based on a survey of more than 600 senior managers of electricity companies with an annual turnover of more than $500 million, shows how electricity systems are entering a new stage marked by the increasing unpredictability of AI workloads. According to the report, forecasting electricity demand has become much more complex, although artificial intelligence itself is also part of the solution, with most managers expecting it to significantly improve efficiency and operational performance.
More volatile and uncertain electricity demand
Beyond the growth in demand, the main challenge is uncertainty. Electricity companies are planning more and more infrastructure to meet a demand for electricity that, in some cases, never occurs.
The report reveals a growing disconnect between forecast demand and actual demand: 67% of industry leaders refer to “phantom” capacity requests from data centers, of which approximately 19% never materialize. This distorts electricity forecasts and increases the risk of both over- and under-sizing investments.
This uncertainty poses a significant challenge for capital allocation. Utilities must decide not only how much to invest, but also where and when to prioritize the modernization of the electricity network to respond to future demand without generating underutilized assets. For large data center operators (hyperscalers), the challenge is equally complex, since they must make important infrastructure investment decisions in a context marked by uncertain forecasts, the availability of electricity in the network and connection times.
«Artificial intelligence is transforming electrical systems far beyond the increase in demand. “It is revealing structural limitations in grid capacity, planning and electricity availability, while making demand more dynamic and difficult to forecast,” said Claire Gauthier, Global Head of Energy & Utilities at Capgemini. “The challenge is no longer just about how much energy is needed, but whether electricity can be delivered reliably, in the right place and at the right time. “Utilities will play a critical role as system orchestrators, using AI to balance grid and customer resources, accelerate available capacity, and enable the next phase of data center growth.”
The dual role of AI: driver of demand and ally
According to the report, electricity consumption associated with training and inference of artificial intelligence models will increase from 25% to 60% of total data center electricity demand over the next three to five years, progressively displacing other traditional computing loads.
At the same time, industry leaders consider that AI can become a multiplier factor to improve the planning and reliability of electricity networks. About six in ten expect advanced AI-based analytics capabilities to improve failure reduction, operational productivity, and power outage prevention and recovery by more than 10%.
AI Adoption Still Limited
Despite their advantages, the implementation of these technologies remains limited. Less than half (45%) of power companies are currently using AI to optimize electricity grid management and only 16% have deployed advanced AI-based solutions to optimize electricity flows, strengthen resilience and improve system performance in real time.
The report also points out that the long lead times required to build new electricity infrastructure constitute another major obstacle to absorbing the rapid growth in demand derived from data centers. Therefore, it highlights the urgent need to accelerate the modernization of electricity networks through artificial intelligence and other climate technologies, in order to guarantee a reliable, affordable and sustainable supply.
On-site electricity generation gains prominence
Faced with network limitations and connection delays, data centers are evolving from traditional backup systems toward self-generated solutions and behind-the-meter (BTM) systems. Almost three out of ten organizations already have this type of solutions and 39% plan to incorporate them in the next one or two years. Furthermore, more than seven in ten believe that these technologies will significantly reduce their dependence on the electricity grid within five years.
Artificial intelligence is transforming electrical systems far beyond the increase in demand
86% of respondents consider the ability to operate partially independently of the electricity grid to be a competitive advantage. This evolution is redefining the traditional relationship between electricity companies and large electricity consumers, generating new opportunities, but also important coordination challenges.
A diversified energy mix, key to sustainable growth
The report concludes that having a diversified energy mix will be essential to guarantee reliable and resilient growth of data centers. 78% of power company executives and 73% of data center managers believe that renewable energy alone cannot yet provide enough continuous, large-scale electricity to meet the needs of AI-powered data centers.
Both groups claim to be actively investing in battery storage systems (BESS) to fill this electricity supply gap.
They also agree that long-term solutions, such as small modular nuclear reactors (SMR), will still take years to be widely deployed. Meanwhile, more than two-thirds (68%) believe that natural gas will be a necessary transition solution to ensure the availability of electricity until renewable energy and storage technologies reach sufficient maturity, although they recognize that this creates tensions with decarbonization objectives.
“For both energy providers and data center operators, the main challenge no longer lies solely in increasing capacity, but in doing so in a context of uncertainty, speed limitations and increasing system complexity,” concludes Claire Gauthier. “Success will depend on the ability to align infrastructure investment, electricity supply and AI-based operations to manage both the magnitude and volatility of demand, while balancing reliability, cost and sustainability.”
