US AI data centers could be forced to reduce power use during grid emergencies under a new PJM proposal aimed at protecting reliability across 13 states and Washington, D.C. The plan would mainly affect new large loads that have not secured their own independent power supply by June 1, 2027.
What PJM Is Proposing
PJM Interconnection wants to build a Large Load Registry to track major sites, their service areas, and whether they bring their own generation or contracted supply. Under the proposal, new large loads that do not secure enough backup supply could be curtailed before other emergency load-management tools are used. PJM defines a “large load” as a customer with at least 50 MW of peak demand at a single site within a one-mile radius.networkworld+1
Why This Matters
The proposal reflects a widening gap between AI-driven electricity demand and the pace of new generation coming online. PJM says new large loads could rise by about 70 GW by 2038, while roughly 15 GW of generation has retired since 2022. That makes power access a bigger constraint for data center growth, especially in regions where the grid is already stretched.insidelines.pjm+2
Who Is Most Exposed
The facilities most at risk are new builds that have not lined up dedicated power sources yet, especially smaller operators without the resources to self-supply. Large hyperscalers such as Amazon, Google, and Microsoft are generally better positioned to build their own backup generation or contract for supply. Areas with dense data center growth, including Northern Virginia and parts of Ohio, Pennsylvania, and Maryland, are seen as especially vulnerable.networkworld+1
Practical Constraints
Building on-site generation is not a quick fix. New power plants can take years to permit, order, and connect, and equipment lead times can stretch to 3-5 years. In the near term, smaller measures such as battery backup or agreeing in advance to curtailment during emergencies are more realistic. Even today, most data centers have emergency generators, but relatively few can run independently from the grid for long periods.networkworld
Broader Industry Impact
This is not just a data center issue; cloud customers could also feel the effects indirectly if their providers are in constrained regions. If several facilities in the same area are curtailed at once, enterprises could face higher costs or temporary service limits during emergency conditions. Analysts say companies should ask providers about their backup power plans and spread critical workloads across multiple regions.networkworld
What Happens Next
PJM’s move is likely a sign of things to come as other grid operators confront similar pressure from AI-related demand growth. If supply remains tight, more regions may require large loads to bring their own power or accept curtailment risk. That could make electricity availability a central factor in future data center site selection and investment decisions.
AI Data Centers Face New Grid Risk
US AI data centers could be forced to reduce power use during grid emergencies under a new PJM proposal aimed at protecting reliability across 13 states and Washington, D.C. The plan would mainly affect new large loads that have not secured their own independent power supply by June 1, 2027.
What PJM Is Proposing
PJM Interconnection wants to build a Large Load Registry to track major sites, their service areas, and whether they bring their own generation or contracted supply. Under the proposal, new large loads that do not secure enough backup supply could be curtailed before other emergency load-management tools are used. PJM defines a “large load” as a customer with at least 50 MW of peak demand at a single site within a one-mile radius.networkworld+1
Why This Matters
The proposal reflects a widening gap between AI-driven electricity demand and the pace of new generation coming online. PJM says new large loads could rise by about 70 GW by 2038, while roughly 15 GW of generation has retired since 2022. That makes power access a bigger constraint for data center growth, especially in regions where the grid is already stretched.insidelines.pjm+2
Who Is Most Exposed
The facilities most at risk are new builds that have not lined up dedicated power sources yet, especially smaller operators without the resources to self-supply. Large hyperscalers such as Amazon, Google, and Microsoft are generally better positioned to build their own backup generation or contract for supply. Areas with dense data center growth, including Northern Virginia and parts of Ohio, Pennsylvania, and Maryland, are seen as especially vulnerable.networkworld+1
Practical Constraints
Building on-site generation is not a quick fix. New power plants can take years to permit, order, and connect, and equipment lead times can stretch to 3-5 years. In the near term, smaller measures such as battery backup or agreeing in advance to curtailment during emergencies are more realistic. Even today, most data centers have emergency generators, but relatively few can run independently from the grid for long periods.networkworld
Broader Industry Impact
This is not just a data center issue; cloud customers could also feel the effects indirectly if their providers are in constrained regions. If several facilities in the same area are curtailed at once, enterprises could face higher costs or temporary service limits during emergency conditions. Analysts say companies should ask providers about their backup power plans and spread critical workloads across multiple regions.networkworld
What Happens Next
PJM’s move is likely a sign of things to come as other grid operators confront similar pressure from AI-related demand growth. If supply remains tight, more regions may require large loads to bring their own power or accept curtailment risk. That could make electricity availability a central factor in future data center site selection and investment decisions.
Recent Posts
Recent Comments
Search
About Us
There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour.
Categories
Recent Post
AI Data Centers Face New Grid Risk
July 30, 2026Huawei Reportedly Eyes Entry Into the DRAM Market
July 21, 2026OpenAI Launches AI-Driven Effort to Patch Open-Source Security Flaws
June 29, 2026