The largest power grid in the United States is about to start pulling the plug on data centers when supply runs tight. PJM Interconnection, the operator that serves 67 million customers from Virginia to Illinois, says it will curtail power to large data centers and other big users during shortages, according to TechCrunch AI. The decision comes after an auction to add new generating capacity fell short, leaving the grid with less cushion than the AI build-out demands.
This is a direct collision between the AI infrastructure boom and the physical limits of the power system. And it’s the clearest signal yet that the grid, not the chips, may become the real bottleneck.
What PJM actually decided
- Curtailments start in June 2027, not immediately.
- Cuts apply only to data centers rated 50 megawatts or larger.
- Affected customers get advance notice, from 30 minutes to a few days depending on demand forecasts.
- Customers who have their power cut will be compensated, the same way older demand response programs treat manufacturers.
- PJM is running another capacity auction to try to close the gap.
The mechanism isn’t new. Demand response programs, where large users agree to power down in exchange for payment, have existed for decades. What’s new is the scale of who’s now on the receiving end. Data centers have become some of the biggest loads on the system, and PJM’s own independent market monitor blamed them for much of the increase as wholesale electricity prices nearly doubled over the past year.
Why this matters
Data centers are expected to use 4x more electricity by 2035 than they do today, TechCrunch AI reports. That growth is landing faster than operators can build generation, and PJM has taken heat in recent months for how it’s managed both new capacity and the flood of large new users.
What stands out here is the shift in posture. For years, the assumption was simple: if a hyperscaler wanted power, the grid would find a way to deliver it. PJM is now saying the opposite. During a shortage, data centers are the first to get throttled, and they’ll have to plan around that.
The likely fallout
Expect operators to respond in two ways:
- On-site power. Many new data centers, and possibly existing ones, will move to build their own generation so they’re not exposed to grid cuts. This accelerates a trend already underway with gas turbines, fuel cells, and even talk of nuclear.
- Backup generators. Sites that don’t build dedicated power will lean on backup units, which cost more to run and pollute more. Diesel is the common choice because the fuel is easy to store and widely available.
That second path carries real friction. Federal rules cap diesel backups at 50 hours per year for demand response events and 100 hours for emergencies and maintenance. And the environmental cost is already drawing scrutiny. This week, Vantage Data Centers came under fire for its apparent coordination with Virginia regulators to cast doubt on a report estimating that diesel backups near a 96-megawatt data center in Northern Virginia could cause tens of millions of dollars in annual health damage.
What to watch next
The 2027 start date gives operators a runway, but the direction is set. If you’re building or siting AI infrastructure in PJM territory, power strategy just moved from a line item to a core design decision. A few things to track:
- Results of PJM’s next capacity auction, which will show whether new supply is actually coming.
- How hyperscalers structure on-site generation deals, and whether nuclear and gas commitments speed up.
- Regulatory fights over backup generator emissions, especially in Virginia’s data center corridor.
- Whether other grid operators outside PJM adopt similar curtailment rules.
The AI industry has spent the last two years racing to secure compute. The next race is for electrons, and PJM just made that explicit. Full details are available at the original TechCrunch AI report.