daily · Clean Energy
PJM tells FERC that always-on AI data centers are straining the grid, citing a 3,900 MW July load drop
At FERC's annual Reliability Technical Conference on October 21, PJM operations planning chief Christopher Pilong warned that flat, round-the-clock data center demand is crowding out shoulder-season grid maintenance. PJM is writing ride-through standards after roughly 3,900 MW of large loads abruptly shifted to backup generators on July 22, and is piloting 'compute flexibility' with Emerald AI and at an NVIDIA AI factory in Manassas, Virginia.
By Greener Numbers Editorial Team · Published October 23, 2026 · Updated 10/23/2026
PJM Interconnection used the Federal Energy Regulatory Commission's annual Reliability Technical Conference on October 21 to press the case that the accelerating buildout of AI data centers is changing how the nation's largest grid has to be run. Christopher Pilong, PJM's senior director of operations planning, told commissioners that data centers' nearly constant energy consumption "really is like a rising sea level ... every hour of the day," eroding the spring and fall shoulder seasons when generators and transmission owners traditionally take equipment offline for maintenance.
That pressure is already showing up in operations. Pilong said hot weather up to 20 degrees above normal forced emergency procedures across PJM and neighboring grids on October 12-14, conditions he said would not have triggered such steps two or three years ago. "It really is about getting things built," he told the conference, pointing to the lag in new generation and transmission relative to large-load growth.
PJM is developing enforceable voltage and frequency ride-through requirements for large loads, work informed by a July 22 event in which about 3,900 MW of large loads unexpectedly transferred to backup generation in response to routine transmission events. The disruption was managed by PJM and transmission owners but caused concerning voltage fluctuations - the kind of uncoordinated tripping the standards are meant to prevent by keeping facilities connected through disturbances. PJM is also participating in North American Electric Reliability Corp. efforts to draft continent-wide standards for Large Computational Loads, and cited the risk of coordinated, malicious disconnections as further reason for incident reporting and recovery requirements.
In a prefiled statement, Pilong laid out four reliability priorities: timely and continuously updated technical information from large load operators; ride-through requirements for new and existing loads; interconnection and energization decisions that reflect how long generation and transmission actually take to build; and coordinated electric-gas planning, cybersecurity and performance accountability.
On the flexibility side, PJM is working with Emerald AI and other data center customers on "compute flexibility" - throttling non-urgent computational jobs during grid stress - with practices being or to be studied at multiple sites, including an NVIDIA AI factory in Manassas, Virginia. Pilong said PJM is also preparing its response to FERC's call for interim, non-firm transmission service frameworks. "There is no other solution," he said of the coordination push.
Official source: insidelines.pjm.com ↗