The industry that worries most about AI demand is now one of the industries using AI to cope with it. Utility Dive reported on 22 October that utilities are bringing AI tools into grid planning and maintenance as data center requests pile up, though mostly as narrow helpers for their engineers rather than systems that control anything physical.
Where It Is Already in Use
The clearest example in the report comes from Portland General Electric. Working with GridCARE, which models demand hour by hour to find spare capacity, the utility found more than 80 MW that could be offered to data center connections years earlier than expected, by GridCARE's account. Other applications named are load forecasting, vegetation management, document search, anomaly detection and workflow automation.
A survey by National Grid Partners, the venture arm of National Grid, covering utilities in the US, the UK and Europe, puts numbers on the trend as reported by Utility Dive:
- about 80 % of utility leaders said their company had fully deployed at least one AI application for planning large-load customers;
- 87 % said outdated regulation and rate case rules hold back the return on AI and innovation spending;
- 83 % said the cost of building infrastructure for AI data centers reaches residential customers through their bills.
The roles hardest to fill, according to the same survey, are domain experts who also understand AI, data engineers and machine learning engineers.
Why Nobody Lets the Model Touch the Switches
The people quoted were careful. Pradeep Tagare, interim president of National Grid Partners, said the successful projects are tools built for one function, and that agentic AI acting on physical assets has not been widely deployed. Others warned that many demonstrations work only under controlled conditions, that most utilities could not reconstruct what an AI system read and why it decided as it did, and that utility records often differ from what is actually installed, so anything automated on top of them repeats the error at speed.
Regulators are starting to ask questions too. New York has told its utilities to file inventories of their AI use, Connecticut's attorney general and consumer counsel asked on 13 October for a similar review there, and a FERC staff report on this year's cybersecurity audits noted growing use of AI tools and the access-control risks that come with them.
Faster Answers, Same Paperwork
For a data center developer, better forecasting on the utility side could mean a quicker and more honest answer to the first question of any project: how much power, and when. It does not remove the rest. Where the grid can offer only part of the load at first, the gap is often covered by temporary or on-site generation, and the standby plant still needs its air permit, its fuel plan and its test schedule, however clever the utility's software has become.
A forecast is also only as good as the load figure it is fed. Developers who send utilities an inflated request to hold a place in the queue make the planning tools less useful for everyone, their own project included.
Sources: Utility Dive, National Grid Partners