Energy Secretary Chris Wright signed two emergency orders under Section 202(c) of the Federal Power Act on June 30, directing PJM Interconnection — the grid operator serving 65 million people across thirteen states and the District of Columbia — to dispatch backup generation at AI data centers and large industrial loads through July 3 [1]. The threshold was blunt: any facility drawing at least 50 megawatts of peak load was required to switch to its own backup generators within 15 minutes of an emergency signal, freeing public grid capacity for the residential air conditioning units running across a region baking in 100-plus-degree heat [4].

The orders expired July 3. What did not expire is the legal architecture they demonstrated.

This paper's July 1 reporting on FERC's technical conference on data-center load forecasting established that the contest at the July 7–8 FERC proceedings was over which load projections the grid would be allowed to believe. The July 2 story on FERC's Section 206 show-cause orders moved that argument from adjectives into a dated public file, giving data centers 30 days to justify their tariff positions or face reform. The 202(c) order is the first enforcement receipt for that regulatory build-up: the federal government tested data-center "self-supply" in real time and found it wanting.

The heat emergency that triggered the orders was not a statistical anomaly. PJM forecast a peak load of 166,304 megawatts on July 3 — which would have surpassed 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]. PJM's demand is now growing at its fastest pace in decades, and the Department of Energy's own filing cited the North American Electric Reliability Corporation's finding that the primary driver is data centers, followed by electrification and manufacturing loads [4].

The heat emergency ran alongside 29 deaths in New Jersey and more than 900,000 households losing power across the region, with an estimated three million users of home medical equipment exposed to outages [2]. Data centers were drawing evaporative cooling water and generator exhaust from the same stressed grid that was failing residential customers. The backup-generator mandate did not sever that overlap; it papered over it.

Here is the structural novelty buried in the emergency language. Backup generators do not draw from the solar panels or dedicated fuel storage that the phrase "self-supply" implies. They run on natural gas piped from a shared regional distribution network. When PJM directed data centers to switch to their own generators during an emergency in which gas demand was already spiking, the order did not remove those facilities from the shared infrastructure — it changed which infrastructure they were sharing [1]. The Department of Energy has not published data on aggregate generator gas consumption during the emergency period, and no facility has disclosed its individual draw.

This is the 202(c) fiction the order itself created: it defines data centers as capable of operating independently of the grid, then relies on a shared gas network that it does not regulate.

The legal consequence is more durable than the policy gap. This was the third Section 202(c) order issued to PJM in 2026 alone [1]. Three orders in one year establishes a pattern. FERC's show-cause orders to all six regional transmission organizations now have a concrete 202(c) use case attached to them — which means the next round of proceedings will be conducted with everyone knowing what a failure state looks like. Regulators negotiating with data center operators over grid-adequacy demonstrations have a documented emergency to invoke.

The orders also exempted specific facility categories: hospitals, 911 call centers, water treatment plants, air traffic control towers, and defense installations [4]. Data centers are not on that list. That exclusion is not administrative oversight; it is a policy judgment about which infrastructure the federal government considers essential. Hospitals cannot be curtailed. Data centers can be. That distinction is now in the regulatory record.

The question FERC's show-cause process will eventually have to answer is whether the 202(c) instrument remains a one-time emergency tool or gets codified as a standing mandate. If the backup-generator requirement becomes a tariff condition rather than an emergency waiver, AI infrastructure operators face a material new operating cost: maintaining dispatchable backup generation at 50-MW-plus facilities is not free, and the fuel and maintenance obligations would belong to the data center, not the grid [3].

Al Jazeera's coverage framed the orders as evidence of AI's growing energy appetite threatening grid reliability [2]. CNN described the orders as the federal government protecting residential customers from data-center competition for power [5]. Neither framing is wrong. Both miss the operative fact: for the first time, a specific class of private infrastructure actors was told, in writing, by a cabinet secretary, that their access to the public grid is conditional on federal satisfaction with their emergency operating posture.

That is not a heat footnote. It is the first debit entry in a ledger AI infrastructure did not know it was carrying.

-- DARA OSEI, London