The US electricity grid is likely to be highly constrained and less stable by 2030, and datacenters aren’t helping.

Some regions are forecast to have a safety margin that leaves little reserve capacity to cover emergencies.

These gloomy predictions come in a report from energy infrastructure biz Schneider Electric. It calculates there will be just enough grid capacity to power all expected electricity demand through 2030, but the safety margins are set to become “increasingly thin” as the end of the decade approaches, with some regions likely to come dangerously close to a shortfall.

Schneider used forecast data and metrics supplied annually to the North American Electric Reliability Corporation (NERC) by US Regional Transmission Operators (RTOs) and Independent System Operators (ISOs) to gauge the level of grid capacity and safety margins. It then analysed how well it might accommodate all projected demand, including that from datacenters, between 2025 and 2030.

The report, “Assessing the US Power System’s Ability to Support Data Center Growth” [PDF],  says the grid is now experiencing a surge in energy requirements, following a lengthy period of little growth or even a decline due to systems becoming more efficient.

This echoes an earlier report from management consultancy Bain & Company, which claimed energy companies had been caught on the hop by the unexpected rapid increase power requirements, which could outstrip supply in just a couple of years, unless utilities are able to boost their energy generation by up to 26 percent by 2028.

Schneier warns that seven of the 13 NERC regions are likely to be at a much higher risk of issues such as outages, rolling blackouts, voltage reductions, and price increases during peak demand periods, while being severely limited in the ability to add new loads to the grid.

These seven regions include some of the largest, meaning that they cover a huge swathe of the continental US. Those in less danger of outages include California, the NPCC New England and New York regions, the SERC southeast and Florida regions, plus the area covered by the Midwest Reliability Organization.

To be clear, Schneider says that in the “high risk” regions, there is expected to be sufficient grid supply to meet peak demand. But a lack of safety margin means that if the grid should experience an unexpected problem, such as a large generator site suddenly going offline, there may not be enough supply to meet peak requirement at that time.

It isn’t just datacenters to blame, despite numerous reports forecasting that booming demand due to AI is likely to see global bit barn energy use double by 2030, from sources such as Goldman Sachs and the International Energy Agency. Also in the firing line are EV adoption, manufacturing onshoring, and the ongoing electrification of devices and processes.

The problem with rapidly rising energy requirements is that the electric industry is heavily regulated at local, state, and national level, and the planning and permitting processes for adding new generation and transmission capacity are “notoriously long and complex involving various technical studies with multiple layers of federal, state, and local approvals,” the report states, meaning utilities are struggling to keep pace.

The implications of all this are that datacenter operators are forecast to pay more for energy and face new rate structures, according to Schneider, particularly in hotspots such as Northern Virginia, Dallas/Ft Worth, Chicago, Phoenix, and Silicon Valley.

It also says energy availability is set to become the dominant factor in site selection, and many developers may choose to build in secondary markets where grid capacity is less constrained, such as Louisiana, Mississippi, and Wisconsin.

• AI datacenters want to go nuclear. Too bad they needed it yesterday
• Nvidia GPU roadmap confirms it: Moore’s Law is dead and buried
• Schneider Electric plugs into AI’s power hunger with Nvidia digital twin tech
• Microsoft tempted to hit the gas as renewables can’t keep up with AI

The ability to get access to energy when needed is also driving interest in connecting bit barns directly to grid-scale generators in a “behind the meter” fashion, and prefabricated datacenter modules may become more appealing.

One example of the latter outside the US is a Brit startup which is putting AI datacenter infrastructure inside enclosures similar to shipping containers, which are then deployed on the same site or adjacent to a biogas generator facility, directly drawing power from it for a lower rate than connecting via the grid.

By extension, on-site energy storage and/or power generation will become more common, with campuses evolving to accommodate natural gas turbines, generators fueled by hydrotreated vegetable oil (HVO), co-located wind/solar, fuel cells, and battery energy storage systems.

The report also warns the rush to build new datacenters for AI over the next several years is driving up carbon emissions, with large operators expressing concern about meeting their net zero commitments.

Some coal and natural gas plants that had been earmarked for retirement are also being kept up and running to provide sufficient baseload power for the grid, with the result that efforts in the US to reduce greenhouse gas emissions will be negatively impacted, at least in the short term.

Schneider’s advice for datacenter developers is to consider using energy procurement companies to obtain a faster grid connection, and to get involved with the utility firms early on for their input on site selection, design, and connection requirements.

It also recommends looking at energy-as-a-service, touted as a new kind of product where a customer can outsource their energy provisioning completely to a provider. This covers procurement, installation of any necessary equipment, and even possibly building on-site power generation capacity. ®

https://www.theregister.com/AMP/2025/06/03/schneider_electric_says_us_grid/