AI Data Centers: FERC Forces Grid Operators to Charge the True Cost of Interconnection

The Federal Energy Regulatory Commission (FERC) unanimously adopted a show-cause order on June 18, 2026 (docket RM26-4-000), requiring six US regional grid operators — PJM, MISO, SPP, CAISO, ISO New England and NYISO — to either demonstrate their interconnection rules meet the "cost causation" standard or file revised tariffs within roughly 60 days. ERCOT (Texas), outside federal jurisdiction, is not covered.

The core of the order: large new grid users — chief among them data centers built for AI training and inference — must now pay the full cost of grid upgrades directly tied to their own interconnection, rather than having that cost socialized across the broader ratepayer base through distribution tariffs.

The numbers explain the urgency: US AI data center power demand is estimated at 64.4 GW in 2025, 75.8 GW in 2026, and projected to reach 134.4 GW by 2030 — 9 to 17% of total US electricity load. Several hyperscalers have already bypassed saturated interconnection queues through direct power deals: Microsoft (835 MW, Three Mile Island restart), Amazon (1,920 MW of nuclear agreements through 2042), Meta (1,100 MW, Clinton, Illinois plant).

The deadline for revised tariff filings fell around August 17, 2026. The order says nothing about demand itself: it redefines who — the broader ratepayer base or the data center driving the need — bears the marginal cost of grid congestion, an instrument choice with direct implications for energy and O&G players exposed to AI's electricity demand.