JUNE ISSUE OF THE CURRENT
ERCOT’s planned transmission outage restriction period opened four weeks ago. Through September 15, no 345kV line can come offline during peak hours. NYISO’s baseline reserve margin this summer is 417 MW, against a reserve requirement of 2,620 MW. A three-day heat event at 95 degrees drops that margin to negative 1,679 MW.
We’re a month into a four-month window. In ERCOT, the maintenance work that didn’t get done before May 15 waits until September. And September is arriving with a different grid than the one those outage schedules were built around.
Planning outages was supposed to be the tractable part
Interconnection queue reform, IBR compliance, large load jurisdiction. These have been the dominant storylines. Planned outage coordination rarely makes the conference agendas. It’s internal work, unglamorous, handled in scheduling systems and spreadsheets and phone calls between transmission operators who’ve been doing this together for years.
That familiarity is also the problem. The workflows function because the grid changed slowly, with a known generation mix, predictable peak windows, and load profiles that shifted gradually season to season. Schedulers knew which lines could come down in June without stressing the network because they’d done it the June before under similar conditions.
The grid those schedules were built for doesn’t exist anymore.
A new variable in the window
In May, NERC issued a Level 3 alert, its highest tier, regarding large computational loads dropping off the bulk power system without warning. Not tripping during faults the way an inverter-based resource might, but simply and rapidly reducing demand by 1,000 MW or more. NERC’s review of entity responses to its prior warnings found that most operators didn’t have sufficient processes, procedures, or methods to address these loads. Transmission operators must respond with specific actions by August 3, now seven weeks away. Alongside the alert, NERC released proposed registration criteria for a new functional entity type it’s calling the Computational Load Entity, covering any facility hosting 1 MW or greater of computational load. The compliance architecture for this load class is being built in real time.
Outage scheduling is a contingency analysis problem. You model the network with the planned outage in place, run the contingencies, verify nothing exceeds its limits, approve or deny. The whole thing rests on the load model.
“A data center that can shed several hundred megawatts in seconds isn’t a stable load in any sense the outage model was built for.”
Traditional industrial loads don’t behave this way. A furnace, a compressor, a motor: none of these reconfigure themselves in response to internal software logic. A data center does. It can shed workloads, switch to backup generation, or respond to thermal management controls in ways that change its grid footprint in seconds. If that happens during a contingency the operator was already managing, the power flow picture the approved outage plan was built around no longer describes the system the operator is actually running.
None of this is in the static load model. And most of it happened after the last time the model was updated.
What this summer actually tests
NERC’s draft 2026 Summer Reliability Assessment is more optimistic than last year. Elevated risk regions are down from six to three. MISO, ERCOT, and others have added significant capacity. This is a genuinely better position on paper.
But adequate capacity and adequate situational awareness aren’t the same thing. A region can have sufficient generation in aggregate and still run into trouble if a large load drops at the wrong moment during a contingency the operator was already managing. It’s not a capacity problem. It’s a modeling problem. Those don’t get solved by adding megawatts.
“Adequate capacity and adequate situational awareness are not the same thing. This summer tests both.”
When we ask planning teams how current their large load models are, the answer is almost always some version of: we’re working on it. The data centers that came online in the last eighteen months, the ones in commissioning now, the ones whose operating characteristics haven’t been fully characterized, are already on the system. The window is open. The studies that were supposed to happen are competing with everything else on the schedule. A useful place to start is simple: when was your large load model last updated, and how many data centers have come online since?
The outage window doesn’t wait for the model to catch up. Neither does summer.
If your team is navigating this right now, we’d like to hear what you’re running into.
Subscribe to The Current and follow Simple Thread on LinkedIn for fresh analysis, project highlights, and event updates, delivered straight from the front lines of utility software.
Loved the article? Hated it? Didn’t even read it?
We’d love to hear from you.
