FERC Pushes Grid Operators to Fast-Track Data Center Power Connections as AI Demand Surges

The Federal Energy Regulatory Commission is pushing grid operators to update rules for connecting massive power users like AI data centers, citing grid stress and affordability concerns. In a move announced in late 2024, FERC gave six grid operators 60 days to justify or revise their tariffs for large-load interconnections and 30 days to explain how they will handle surging demand from AI and other industrial users.

FERC's order follows a surge in interconnection requests from data centers and AI factories. According to the commission, the volume of large-load interconnection requests has doubled in the past year, with AI-related facilities accounting for roughly 40 percent of new demand.

New Deadlines and Transparency Requirements

FERC's directive applies to six independent system operators and regional transmission organizations across the United States. Each must respond within the specified timeframe:

• PJM Interconnection: Must revise tariffs for data center co-location and behind-the-meter generation.
• Midcontinent Independent System Operator (MISO): Must justify how it handles large-load requests from AI and manufacturing.
• Southwest Power Pool (SPP): Required to report on interconnection queue processing times.
• California ISO (CAISO): Needs to address generator interconnection agreement backlogs.
• New York ISO (NYISO): Must clarify rules for co-located load and generation.
• ISO New England (ISO-NE): Required to update procedures for large-load interconnection studies.

The order also mandates that each grid operator submit a plan detailing how it will ensure cost allocation for network upgrades is transparent and not unfairly shifted to residential ratepayers.

Energy Efficiency Innovations Gain Attention

As grid regulators act, researchers are pursuing novel cooling technologies that could reduce data center power and water use. MIT nuclear engineering researchers have adapted a two-phase cooling concept originally developed for nuclear reactors. The method uses engineered bubbles in a dielectric fluid to remove heat more efficiently than traditional air cooling, potentially cutting cooling energy consumption by 30 percent and water use by 80 percent.

The approach, described in a paper published in October 2024, targets the thermal management of AI chips that can draw 1,000 watts or more per chip. The researchers claim the bubbles create a thin film that improves heat transfer, a technique they say could be retrofitted into existing data center liquid cooling loops.

These innovations may ease pressure on grid operators facing FERC's scrutiny. However, deployment at scale remains years away. For now, the regulatory push is focused on near-term fixes: faster interconnection studies, transparent cost accounting, and a clearer pathway for large loads like AI data centers to connect without destabilizing the grid.

NVIDIA has publicly supported FERC's actions, calling them a major milestone for AI infrastructure deployment. The outcome of the 60-day tariff review window will set a precedent for how the next generation of hyperscale data centers interacts with the U.S. power grid.