Energy Secretary Chris Wright signed two emergency orders under Section 202(c) of the Federal Power Act on June 30, authorizing PJM Interconnection to direct data centers and other large loads onto diesel backup generators within 15 minutes of an emergency signal. The orders ran through July 3. PJM's demand forecast for the period peaked at 166,304 MW, against the grid's all-time record of 165,563 MW set in 2006 — before a single significant AI data center existed in the PJM footprint. [1]

This paper reported on PJM's third consecutive capacity-price record when prices moved to $329.17 per megawatt-day — an 11-fold increase from the 2024-25 level of $28.92. The July 5 edition advanced that story when a live DOE order arrived in the middle of a heat emergency, converting a forecast problem into an emergency instrument. Today's story is what the instrument does: it does not remove data centers from the grid's problem. It transfers the problem to a different medium and a different set of people.

The transfer mechanism

The 202(c) order did not curtail data-center operations. It required campuses with at least 50 megawatts of peak load to disconnect from the shared utility grid and switch to their own backup generation resources — diesel generators and battery arrays — within 15 minutes of an emergency signal. [1] The AI compute continues. The electricity comes from on-site diesel engines instead of transmission lines. The difference, from the grid's perspective, is load reduction. From the community's perspective, the difference is air quality.

The backup generators deployed under the order are primarily Tier 2 diesel units. PJM officials acknowledged in their emergency filing that the order could result in exceedances of sulfur dioxide, nitrogen oxide, and carbon monoxide limits within the communities hosting the data campuses. [1] Those communities are not the same communities that hosted data centers when they were first built; the hyperscale campuses that dominate PJM's load growth are concentrated in Northern Virginia, central Ohio, and the outer-suburban rings of Philadelphia and Chicago — communities with existing air-quality vulnerability and disproportionate concentrations of populations that bear elevated baseline health risk from particulate pollution.

Price and cost

Day-ahead settlement prices on the PJM Western Hub exceeded $1,222 per megawatt-hour during the emergency period. [2] Wholesale prices across the PJM footprint crossed $2,000/MWh at peak. [2] Ordinary ratepayers on variable-rate electricity plans pay day-ahead settlement prices when they use electricity during emergency periods. The rate is not capped for residential customers during a 202(c) order.

The capacity price context matters. PJM capacity prices have risen from $28.92 per megawatt-day in the 2024-25 planning year to $329.17 in the 2026-27 planning year, an increase that reflects the grid's need to procure enough reserve capacity to meet the load that AI data centers have added. [3] That cost is paid by all PJM customers — residential, commercial, industrial — through their electricity bills. The data centers add load. The capacity market responds. The costs distribute across the full customer base. The 202(c) order shifts the heat-wave marginal cost to diesel exhaust. The structural capacity cost remains.

What the order did not resolve

PJM has asked states to develop rules to allocate reserve-margin costs specifically to hyperscalers — to make the connection between the load that data centers add and the costs that load generates visible in the rate structure. Those state rules do not exist. The 60-day FERC deadline for grid operators to justify or reform their cost-allocation rules for large loads falls in August — the same month the August 1 tariff deadline and the next potential heat wave peak collide. [1] The structural question — who pays for the grid that AI builds — is in regulatory process. The emergency is live now.

The third 202(c) order of 2026 is evidence that the emergency is not a single heat-wave event. It is the new operating condition of a grid that was not designed for current AI data-center load, managed by a cost structure that does not yet assign that load's cost to the entities that create it.

-- DARA OSEI, London