The Datacenter Trap
On August 20, 2025, the Louisiana Public Service Commission approved Entergy's request for 3 new gas plants specifically to handle a single datacenter for Meta. This datacenter hasn't been built yet, nor does it have a set completion date. Overall, Entergy forecasts a 12-13% annual growth in demand from industrial customers, 60% of which comes from new datacenters. A few states north, Ameren's 2025 update to its resource plan forecasts an increase of 1.5 GW of datacenter load by 2032. Out west, Pacificorp forecasts 1.8% annual load growth, and on the east coast Dominion Energy expects a 5% year-over-year increase in load[5], both due to datacenter growth.
This load forecasting isn't confined to utilities either. MISO projects load to grow by 1.1-2% each year due to "industrial and technological changes". PJM expects 9.5% growth by 2030, again due to datacenters. ERCOT expects 22GW of datacenter load by 2030. And on and on. Across the contiguous states, everyone seems to be predicting a huge increase in electric load due to datacenters that have not yet been built.
However, large multi-year expansion of datacenter capacity is far from a sure thing, despite what the energy sector may assume. In reality, it is not certain that any datacenter will be completed or used. Even at the best of times, large datacenters are multi-year, multi-billion dollar infrastructure projects. Like all such projects they run a number of risks which can terminate the project, including construction cost overruns and equipment shortages. And this isn’t the best of times either. The current datacenter build out is being driven mainly by the AI craze. AI datacenters are specially designed for LLM training models and are purpose built to house millions of GPUs and their associated cooling – they are not readily convertible to other uses. Any drop in AI-related spending will most likely result in stranded datacenters that do not operate, and do not require power from the grid.
The uncertainty of datacenter construction should also make one wonder about the accuracy of forward load forecasting by utilities in general. In fact it is not unusual for load forecasts to overestimate expected load. To take just one example, Ameren predicted in 2011 that electricity demand in its Missouri territory would grow to almost 11GW by 2025; by 2023, that forecast had shrunk to only around 7 GW. In fact, from 2008 to 2020, Ameren has consistently overestimated their actual demand. Forward load projections in general are not certainties, but are instead a speculation about the future.
You may be tempted to think that this isn't a big deal: So what if the datacenters don't get built--some investors will lose money and the demand won't materialize, but who cares? Except that these forecasts do not exist in a vacuum, and they are not presented as an idle curiosity. Utilities use these forecasts to justify their desires during the regulatory process. When PG&E goes to the California Public Utility Commission and says that their datacenter pipeline has swelled to 10GW, it isn't for fun: they also asked the CPUC to approve an interim rule related to large loads. When Ameren asked the Missouri Public Service Commission to approve a 700 MW gas-and-storage plant, they claimed it was needed to support new large load customers. Entergy very explicitly tied a single datacenter to 3 gas projects.
When a regulated utility builds new generation, that generation becomes part of the rate base, and is used as further justification for a rate-increase--the utility is able to pass the cost of their capital investment on to the people who live in their territory. Utilities are gambling on a specific outcome, and they are doing so with customer money. If the datacenters are built and the demand arises, then those datacenters will have ready power and the utility will have huge customers to make happy. And if the datacenters aren't built, the new power plants will be paid for by their existing customers in the form of higher rates.
And what will the utility do if the power plants are built, but the demand never materializes? Ameren answers that for us: the new plants will "[Support] the ability to attract additional new large load customers in the intermediate- to longer-term even if…material large load customer load did not materialize". In other words, Utilities are just going to keep trying to find big customers, even if these _specific_ ones don't show up.
And why wouldn't they? Utilities can always afford to keep gambling with our money.