Utilities Fail to Predict AI Power Needs, Finds Capgemini

More than three quarters (77%) of utilities are struggling to forecast the energy demand required by the expansion of AI-driven data centres, according to Capgemini’s new report.

The electricity consumption alone from AI training and inferencing is expected to rise from 25% to 60% of total data centre electricity demand in the next three to five years, largely displacing other IT workloads. 

“AI is reshaping the energy landscape, both accelerating electricity demand and creating new opportunities to improve how energy systems are planned, managed and optimised,” says Claire Gauthier, Global Head of Energy and Utilities at Capgemini, which published the report titled AI Meets the Grid: Shaping the Data Center Power Play yesterday. 

“The challenge facing the industry extends beyond adopting AI. It is embedding intelligence into the way organisations operate, enabling them to continuously adapt, make better capital allocation decisions and respond more effectively to an increasingly volatile environment.”

Forecasting energy for AI workloads

Forecasting has become hard because AI workloads are unpredictable, with consumption patterns are unstable and difficult to model. 

For example, a company might see a massive, multi-week spike in compute demand while training a new foundation model, followed by a sudden drop to a baseline level that fluctuates wildly based on unpredictable real-time user queries. 

As a result, utilities expect demand variability to emerge as a major system challenge, requiring new approaches to planning and operations. 

The majority of the 600 electricity executives surveyed by Capgemini see AI as a force multiplier for grid planning and reliability, with six in ten expecting advanced AI analytics to deliver over 10% improvements in failure reduction, operational productivity, and preventing and restoring outages. 

However, less than half (45%) say they are currently using AI for grid optimisation and only 16% of electricity organisations have implemented advanced AI-driven approaches to optimise power flows. 

“For the first time, organisations have the opportunity to optimise the energy value chain end to end by bringing together engineering expertise, operational technology, digital capabilities and AI within a single operating model,” Claire says. 

“Success will depend not only on technology adoption, but on the ability to orchestrate complex systems, modernise infrastructure including the grid, leverage ecosystem partnerships and make data-driven investment decisions.”

The push for power flexibility

To support AI workloads without sacrificing reliability, data centres are moving away from a reliance on renewables alone and are investing heavily in a diversified energy mix, with 86% of operators viewing the ability to operate off-grid as a major competitive advantage. 

Leaders across both the tech and utility sectors agree this varied approach is essential for long-term growth, with both sides actively funding battery storage systems to bridge the power gap. 

Data centres need to have flexibility, asserts Gerhard Salge, CTO of Hitachi Energy.

“First of all, you need to have complementary power delivery from, for example, solar, wind or hydro” he says.

“Then, you can play with the profiles of which complement each other and exchange power when one is high and the other might be low. The more complementary your sources of energy are, the better you can balance those changes out.

“Storage can also help to fill in gaps when you have an oversupply.”

By filling in the gaps of clean energy supply, electricity executives will have more flexibility to deal with unpredictable power shortages. 

“Those organisations that build operational agility into the core of their business will be best placed to create long-term value while strengthening resilience, competitiveness and sustainability,” Claire concludes.