DOE Orders AI Data Centers Onto Backup Generators as PJM Forecasts Record 166,304 MW Demand
Energy Secretary Chris Wright issued two emergency orders on June 30 under Section 202(c) of the Federal Power Act, authorizing PJM to force AI data centers and large loads onto backup diesel generators through July 3 as the grid approaches a projected all-time peak of 166,304 MW.
The federal government conscripted hyperscaler diesel generators into grid service for the third time in 2026. The pattern is no longer anomalous.
Key takeaways
- Energy Secretary Chris Wright issued two emergency orders on June 30 under Section 202(c) of the Federal Power Act, directing PJM to dispatch backup generation at AI data centers and large loads through July 3, 2026.
- PJM is projecting a potential all-time peak load of 166,304 MW on July 3, which would surpass the prior record of 165,563 MW set in 2006, before a single significant AI data center existed in the PJM footprint.
- This is the third Section 202(c) backup-generator order issued to PJM in 2026 alone, establishing a legal and operational precedent for federal commandeering of private generation assets whenever grid reserves tighten.
Energy Secretary Chris Wright issued two emergency orders on June 30, 2026, directing PJM Interconnection to deploy backup generation resources at AI data centers and other large commercial loads through 11:59 PM EDT July 3, per the DOE press release. PJM is forecasting a potential new all-time peak load of 166,304 MW on Thursday July 3, covering 13 states and Washington D.C., a grid serving more than 67 million people.
The orders were issued under Section 202(c) of the Federal Power Act. PJM submitted the underlying applications on June 27 and June 29. Wright's statement in the release: "Maintaining affordable, reliable, and secure power in the PJM service territory is non-negotiable."
What the Orders Actually Authorize
The first order directs PJM to dispatch specified generation units to maintain reliability. The second, more consequential order authorizes PJM, in coordination with Transmission Owners and Electric Distribution Companies, to direct backup generation resources at data centers and large industrial and commercial sites to operate as a last resort before or during an Energy Emergency Alert Level 3, the point at which firm load gets interrupted and rolling blackouts may begin.
Under new PJM Manual 13 procedures adopted June 24, large loads with backup generation must be able to transfer to that backup within 15 minutes of an emergency dispatch signal. The orders also authorize PJM to direct power plants to operate beyond normal pollution limits if necessary, with potential exceedances of sulfur dioxide, nitrogen oxide, carbon monoxide, and ammonia. PJM must notify DOE when this occurs.
This is the third time in 2026 DOE has issued a Section 202(c) order to PJM specifically, following orders on January 26 and May 18. The May 18 order covered a window when PJM had less than 5,800 MW of reserves and 40,400 MW of generation offline for planned maintenance. At that time, Wright noted in the May 18 order: "Currently, there are tens of gigawatts of readily available backup generation that have remained largely untapped."
The full 2026 202(c) order index tracks the escalating cadence.
The Delta That Explains Everything
The projected July 3 record of 166,304 MW sits 741 MW above the prior all-time peak of 165,563 MW, set in 2006. That 2006 record was set with no material AI data center presence in the PJM footprint. Virginia now hosts the largest data center market in the United States. The delta between 2006 and 2026 is not residential air conditioning growth. It is AI infrastructure.
The NRDC's Tom Rutigliano put it plainly: "It's the data centers that got us to that point, that we're in a place where luck matters."
In December 2025, PJM's capacity auction fell short of its reliability reserve target for the first time in the auction's history, driven by data-center-led demand growth, per PJM's own auction results. Its new Capacity Advisory procedure, also effective June 24, now provides up to five days advance notice of approaching shortfalls. That institutional adaptation confirms what the emergency orders already signal: grid stress from AI load is no longer a weather-event anomaly. It is the operating environment.
This is where the Stargate AI buildout story and the grid-reliability story converge. Non-dispatchable anchor loads are being built faster than transmission and generation can absorb them, and when the system tightens, the government's tool is a Depression-era statute that lets it commandeer private assets.
The Rationing Precedent and What It Means for Mining
Each invocation of Section 202(c) does two things simultaneously: it solves the immediate reliability problem, and it builds out the legal and operational machinery for managing large loads under federal direction. That machinery does not get dismantled between emergencies.
Bitcoin miners are voluntarily dispatchable in ways hyperscalers structurally cannot be. Miners curtail instantly when grid prices spike, sell demand response into wholesale markets, and locate where stranded generation exists rather than concentrating load in congested urban corridors. The power economics are different in every dimension that matters for grid reliability.
The political risk is that regulators do not make that distinction. Bitcoin mining is an easier target than Microsoft Azure or Google infrastructure. If the mining industry does not make the demand-response case loudly and in policy venues before this rationing framework hardens, it will be administered the same way as the loads actually causing the problem.
The falsifiable version of this thesis: if PJM successfully interconnects enough new dispatchable generation to rebuild reserve margins above 20% within three years, and if subsequent heat events pass without Section 202(c) orders, this is a temporary friction story, not a structural rationing regime. Watch the reserve margin trend and the order cadence. Three orders in six months is the current data point.
What to Watch
PJM's July 3 peak will either set a new record or fall short depending on actual temperatures and load. More important than the number itself is whether DOE issues a fourth Section 202(c) order before summer ends. If it does, the "temporary anomaly" framing collapses entirely. Watch also for FERC rulemaking on a voluntary, market-compensated demand-response framework for large loads. If that exists, the emergency-commandeering approach becomes unnecessary. If it does not materialize, the emergency orders become the de facto policy.
Sources
Frequently Asked Questions
Section 202(c) of the Federal Power Act authorizes the Secretary of Energy to order emergency measures to maintain grid reliability during a shortage or supply disruption. Under the June 30 orders, the government can direct PJM to require data centers and large commercial loads to switch from grid power to their own backup generation resources, including diesel generators and battery arrays, within 15 minutes of an emergency signal. This happens only after other reliability tools are exhausted, as a last resort before voltage reduction or rolling blackouts.
Bitcoin miners can reduce load to near zero almost instantaneously and do so voluntarily in exchange for demand-response payments. That flexibility is built into the economics of mining: when power prices spike or the grid tightens, miners curtail and sell back capacity. AI data centers run continuous, non-negotiable compute workloads with no equivalent flexibility. That structural difference is why emergency orders target data center backup generators rather than anything miners are doing, but the regulatory category of "large load" catches both.
The prior PJM all-time peak of 165,563 MW was set in 2006, primarily driven by residential and commercial air conditioning demand during a heat event. The forecasted 2026 record of 166,304 MW occurs in a fundamentally different demand environment: Virginia's Northern Virginia corridor is the largest data center market in the world since 2006, and AI infrastructure buildout has added load that runs 24 hours a day regardless of weather. The incremental demand is structural, not seasonal.
