Largest U.S. Power Grid Operator Plans Temporary Blackouts for Large Data Centers
0xBroomberg
PJM Interconnection, the largest U.S. grid operator, will allow temporary power cuts to data centers of 50 MW or above starting June 2027 — generation capacity simply cannot keep pace with surging data-center demand, signaling that AI's physical bottleneck is shifting from chips to electricity.
Who gets cut off, and why now?
PJM Interconnection — the regional grid spanning Virginia to Illinois and serving roughly 67 million people — announced temporary curtailments for data centers drawing 50 MW or more, effective June 2027.
The trigger: a recent PJM capacity auction fell short of its target, confirming that new generation cannot be built fast enough to match data-center load growth.
This means → the grid operator is conceding a basic math problem: supply-side build timelines have fallen behind demand-side expansion.
How big is the data-center power appetite?
Over the past year, PJM wholesale electricity prices have nearly doubled. PJM's independent market monitor pinned the main cause on surging data-center demand.
Forecasts project data-center electricity consumption will reach four times its current level by 2035.
In plain terms = prices doubled in a year, and the biggest new source of demand still has a 4× ramp ahead — the grid shortfall is arithmetic, not a surprise.
How will curtailments work, and what do data centers get?
The curtailments fall under a demand-response mechanism — a program where large users voluntarily shed load when supply runs short. Manufacturers and other big consumers have participated in similar programs for years.
Affected data centers will be compensated and given advance notice ranging from 30 minutes to several days, depending on demand forecasts.
This means → this is not a blackout — it is a priced, rules-based dispatch tool. But for facilities that market themselves as "always on," agreeing to be curtailed is a significant concession.
Can backup power fill the gap? At what cost?
The policy is expected to push more data centers toward on-site diesel generators — diesel fuel is readily available and storable on-site, but running costs are high and emissions are significant.
Federal rules allow diesel generators to run for demand-response events up to 50 hours per year, rising to 100 hours when emergencies and maintenance are included.
A real-world case has already surfaced: operator Vantage Data Centers was recently criticized for allegedly coordinating with Virginia environmental regulators to challenge a report showing that diesel backup generators near a 96 MW data center in Northern Virginia could cause tens of millions of dollars in annual health damage.
What is the real unresolved question here?
PJM has faced sustained pressure over capacity management and large-user interconnection, and says it will press ahead with a new capacity auction.
The core tension: if curtailments are enforced, data centers accelerate diesel self-generation → bigger emissions controversy. If curtailments are not enforced, grid overload risk stays.
This reflects a physical constraint the AI supply chain had underpriced — the bottleneck is no longer computing power, it is electrical power.
Content is for reference only, not financial advice.