ExSitu

The Grid Wasn't Built for This: The US Regulator Gives Grid Operators 60 Days to Prepare for AI

FERC ordered six US grid operators to show they are handling the huge electricity demand from AI data centers, and to pass the cost of upgrades on to whoever connects.

4 min read

The Grid Wasn't Built for This: The US Regulator Gives Grid Operators 60 Days to Prepare for AI
Unsplash · Yuan Yang

The US electricity regulator (FERC) ordered six major grid operators — PJM, MISO, SPP, CAISO, ISO New England and NYISO — to show that their existing interconnection rules properly handle large electricity consumers, or to file revised tariffs within 60 days, a deadline that passed on August 17, 2026.

Why This Is Happening Now

Electricity demand from AI data centers in the US is expected to grow from 75.8 gigawatts in 2026 to 134.4 gigawatts by 2030 — a gap of almost 60 gigawatts that the grid must absorb in just four years. Interconnection queues that once processed a single request over years now face requests on the scale of entire gigawatts every month from large cloud providers.

Who Pays for the Upgrade

FERC's key decision: new data centers will bear the full cost of the grid upgrades that their specific connection requires, instead of spreading it across the general public of consumers. Texas (ERCOT), which is outside FERC's jurisdiction because of its separate power grid, is not subject to this decision.

What This Signals

The decision reflects a broader trend: power infrastructure, not computing power itself, is becoming the main bottleneck of AI growth. It is a point that matters even to those not operating in the US — every global cloud provider faces the same problem, to varying degrees, in every market where it builds infrastructure.

What This Means for Your Organization

  • If you are considering hosting heavy AI workloads in a cloud abroad in the near term — check with your provider whether its specific data facility is in a region already facing a bottleneck in connecting to the local power grid. A delay in building such infrastructure could translate in the near future into more limited availability or a significantly higher price than expected today.
  • Electricity demand from AI data centers in the US is expected to grow from 75.8 gigawatts in 2026 to 134.4 gigawatts by 2030 — a gap of almost 60 gigawatts that the US grid must absorb in just four years. This is not a theoretical problem for planners in committees; it is already affecting the timelines of real projects.
  • The US regulator (FERC) ruled that new data centers will bear the full cost of the grid upgrades their specific connection requires, instead of spreading that cost across the general public of consumers as is customary today. This trend is expected to spread gradually outside the US as well, and it is worth factoring it in now when pricing any large-scale AI project over the long term, not only at the initial rollout stage.
  • Although this specific decision is entirely American, the pattern behind it is relevant to every market, including Israel: a flood of interconnection requests on the scale of entire gigawatts from large cloud providers is overwhelming aging infrastructure that was never designed for this pace of growth. An organization that depends on a continuous, stable power supply for a constant AI workload should examine this as part of its business continuity considerations, not only as a question of ongoing cost.

Sources