NERC's Level 3 alert on computational loads: seven actions and how they are becoming interconnection requirements

Why NERC escalated

NERC alerts come in three levels. Level 2 is a recommendation; Level 3 is an Essential Action. NERC issued a Level 2 alert on large loads in 2025 and found from the responses that entities “generally did not have sufficient processes, procedures, or methods” for computational load. The event record made the case for urgency. In Dominion's Virginia territory, normally cleared faults led to about 1,500 MW of data center load transferring to backup in July 2024, 1,800 MW in February 2025, and 3,800 MW on July 22, 2026. ERCOT has analyzed 32 events of 100 MW or more since January 2023.

The Level 3 alert went out on May 4, 2026 to Transmission Planners, Planning Coordinators, Transmission Owners, Balancing Authorities, Reliability Coordinators, and Transmission Operators. Acknowledgement was due May 11 and status reports August 3. It is explicit that it is not a Reliability Standard and carries no penalties—but every recipient had to report its status, NERC aggregates those reports for FERC, and the alert says entities should act before receiving a computational load interconnection request, including those that merely could feasibly receive one within two years.

The seven Essential Actions

Read as a whole, the alert treats a data center the way the industry learned to treat inverter-based generation a decade ago: model it separately, study its disturbance behavior, record what it does, and coordinate its protection with the grid's.

From alert to mandatory standard

The alert was never meant to stand alone. The pieces have moved quickly:

• March 18, 2026 — NERC's Standards Committee authorizes Project 2026-02, Computational Loads, recommending one near-term “bridge” standard rather than reopening the MOD, PRC, TOP/IRO, and COM families in parallel.
• April 1, 2026 — draft registry criteria for a “Computational Load Entity” at 20 MW, 60 kV, and more than 1 MW of IT load.
• July 16, 2026 — FERC orders NERC to file Phase I “essential actions” standards and registry criteria by December 31, 2026, and a Phase II workplan by March 1, 2027.
• August 19, 2026 — revised registration splits the entity into a Computational Load Owner (CLO) and Computational Load Operator (CLOP), at sites of 50 MW or more connected at 100 kV or above. Draft standards CLO-001, CLO-002, and CLO-003 post for comment.
• September 2026 — the three CLO drafts and revised FAC-001 and FAC-002 pass their first industry ballot. The posted schedule runs to a final ballot in November and Board adoption in December.
• Next — a Large Loads Working Group draft SAR for Phase II would set minimum voltage ride-through, frequency ride-through, RoCoF withstand, and post-fault active power recovery for computational load. Today, the SAR notes, PRC-024 and PRC-029 cover generators and IBRs, and nothing covers load.

Table 1: How the CLO and FAC drafts turn the actions into requirements

The Phase I package maps closely onto the alert. The table follows the three themes that matter most for interconnection—models, ride-through, and monitoring—and cites the Draft 1 requirement numbers from the August 2026 posting.

Three things stand out:

• The model becomes a compliance artifact. Under CLO-001 R11, the data center owner must verify that model parameters match the installed equipment within a year of in-service and again within 180 days of any firmware or setting change that alters dynamic response. For a site whose UPS and power-supply behavior is defined in firmware, that is a standing obligation.
• Ride-through arrives in two steps. Phase I requires the parameters and the protection coordination; Phase II will require the performance. Interconnection agreements signed in between will be governed by ISO and utility rules.
• EMT is not in Phase I. The drafts require “dynamics” data and models the TP accepts, nothing more specific. Whether a site must deliver a PSCAD model is being decided region by region—which is why the second table matters.

Table 2: How ISOs and utilities are implementing EMT requirements

NERC leaves model type to the planner, so the real EMT requirements are being written by ISOs and Transmission Owners. They are converging on the same shape but at very different speeds.

ERCOT is the leading edge. PGRR 144 makes PSCAD universal for large loads and, for electronic loads, requires the UPS converter model to be benchmarked against hardware tests—voltage ride-through and a subsynchronous frequency scan—with an as-built consistency check before energization. That is the same evidence inverter-based resources have had to produce for years, and it implies EMT models that are validated, not just delivered. MISO's approach is screening-driven: EMT where system strength or TO criteria call for it, with the stricter of the two governing. PJM has put its detail into ride-through and has not yet said anything about models for loads. Utility practice outside the ISO rules mostly reads “EMT may be requested”—which, after the Level 3 alert, increasingly means it will be.

It is also worth seeing what a mature EMT model specification looks like. ATC's planning criteria for inverter-based resources require a PSCAD model that compiles in ATC's Fortran version, runs accurately at a 10 µs step or larger, supports multiple instances in one case, avoids global variables and extra layers, and initializes without external input. Expect large-load EMT requirements to borrow that template.

The ride-through envelopes, side by side

EMT models exist to prove ride-through, so the envelopes they will be tested against matter. ERCOT's NOGRR 282 is approved and in effect; the OCP hyperscaler specification says it mirrors NOGRR 282 closely and adopts MISO's convention. Here is how the OCP commitment compares with the MISO and PJM drafts:

The differences are small in the table and large in hardware:

• MISO wants recovery twice as fast—90% of pre-disturbance power within one second, against two seconds for PJM and the hyperscalers' OCP commitment.
• PJM goes straight at the July 2024 failure mode. That event involved six fault applications in about 82 seconds; PJM's draft bans transfer schemes that act only on a count of sags, and where equipment forces one, requires surviving six sags or swells in 90 seconds.
• Cooling is the open seam. PJM and OCP let cooling trip at deep sags and exclude it from the recovery baseline; the MISO draft does not separate it. A site designed only to the OCP commitment could fall short in MISO.

A developer siting in more than one region should design to the strictest envelope and the fastest recovery, then prove it once with a validated EMT model—ideally with the real controllers in the loop, as covered in our notes on real-time simulation and AI load smoothing tests.

What changes for interconnection from here

The direction is clear even where the details are still moving. By the time Phase I is filed at year end, a large data center will be a registered entity with model, monitoring, and communication obligations; within a year or two, it will have ride-through performance obligations as well. The interconnections being studied today will live under those rules.