NERC's reliability discipline for large loads: how AI data centers are being treated by the grid

The power problem for AI data centers is not only about long transformer lead times or power architecture. How reliability regulators treat data centers as a new kind of load is starting to determine where facilities can be sited and whether they can connect. At the center is NERC(North American Electric Reliability Corporation), which oversees bulk power reliability in North America.

Why NERC is treating "computing load" differently

The starting point is a change in the nature of load itself. NERC has determined that computing loads, including data centers, AI, and crypto assets, create BPS(Bulk Power System) reliability challenges distinct from conventional industrial loads. NERC's definition of "large loads" explicitly lists data centers, cryptocurrency mining, and hydrogen electrolyzers.

The background is rapid demand growth. The share of US electricity consumed by data centers is projected to rise to as much as 12% by 2028, and NERC describes this as load growth without recent precedent. In NERC's forecast, North American summer peak demand increases by more than 224GW and winter peak demand by 245GW from 2026 to 2035. This is a scale traditional grid planning did not assume.

What counts as a large load: threshold debate

The line that defines a regulated "large load" is not yet settled. In NERC's survey, candidate thresholds above 50MW and at 75MW were cited often, and ERCOT(Texas) already treats loads of 75MW or more as large loads. At the same time, NERC notes that 500MW-class and gigawatt-class large loads require treatment different from loads in the hundreds of megawatts. The direction is not a single threshold, but requirements that vary by size band.

Why it is risky: steep changes and real examples

Computing loads disturb the grid because of the speed and size of their changes. In AI training data centers, 50MW blocks can change steeply in about 250 milliseconds. Such rapid changes have been confirmed to harm frequency control, voltage regulation, and system stability.

This is not only a theoretical concern. NERC issued industry guidance in September 2025 on a large-load issue that caused sudden load loss above 1,000MW. Cases have also been identified in Virginia and Texas where data-center large loads amplified grid instability.

NERC's tightening timeline

In response, NERC is strengthening discipline in stages.

• August 2024: established the Large Loads Task Force, now the Large Loads Working Group, and developed a work plan for integrating computing loads into the grid.
• April 1, 2026: published draft new registration criteria for computing-load operators and opened a 45-day public comment period.
• May 2026: issued a Level 3 essential actions alert on large computing loads, requiring seven actions essential to BPS reliability.
• End-2026 target: plans to develop the first Reliability Standard for large computing loads.

The direction of requirements is also becoming concrete. NERC has positioned accurate and validated dynamic models for large loads as an essential requirement for reliability. Operators will need to quantify the variability characteristics of their own loads.

The core mismatch is timing

Even as discipline tightens, a fundamental difficulty remains. Transmission projects often take five to ten years or more from planning to completion, while some large loads seek grid interconnection within one year. This timing mismatch overlaps with stalled interconnection queues(DOE/FERC large-load interconnection reform) and long transformer lead times, delaying data-center energization.

Business implications

NERC's stricter discipline directly affects data-center site and interconnection strategy. First, large-load rules at candidate sites, including thresholds, registration criteria, and reliability requirements, need to be checked early in site selection. Second, operators need the capability to show their load-variation characteristics through validated dynamic models. Third, designs should assume long lead times for transmission and transformers and include onsite power or storage to soften abrupt load changes on the grid. The grid is no longer something that can simply be connected and used. Whether and how quickly a site connects now depends on how well it satisfies reliability requirements.

Reference FactCards