Data centers are already a material component of electricity demand. Multiple analyses now estimate that data centers consume roughly 4% of U.S. electricity today, with projections reaching 7-9% by the early 2030s under high-growth scenarios.

The International Energy Agency has been explicit about the role AI plays in this acceleration:

“AI-optimized data centres are significantly more energy-intensive than conventional data centres, and their rapid deployment is a key driver of electricity demand growth.”
— International Energy Agency, Energy and AI

This is not limited to a handful of hyperscale campuses. Growth is being driven by:

• Cloud migration across regulated industries
• Increased redundancy and uptime requirements
• Latency-sensitive workloads moving closer to population centers
• AI inference workloads that scale with usage, not training cycles

Once built, data centers rarely disappear. They become embedded infrastructure.

The stress this demand creates is already visible in planning processes. PJM Interconnection, which coordinates power for 65 million people across 13 states and Washington, D.C., has publicly acknowledged that large, concentrated data-center loads are reshaping near-term reliability planning.

As one PJM official put it recently:

“The challenge is not just how much load is coming, but how fast—and where.”

This distinction matters. Grid constraints are increasingly local and temporal, not just system-wide. Even if aggregate demand forecasts are revised downward, localized congestion and peak reliability risks remain.

It is reasonable—and healthy—for markets to question whether today’s AI investment pace is sustainable. Financial media and analysts have begun drawing parallels to past technology cycles, suggesting demand forecasts could be overstated if capital spending cools.

But from a grid perspective, the downside scenario is often misunderstood.

If AI investment slows:

• Growth rates may moderate
• Some speculative data-center proposals may not materialize
• Timelines may stretch

What does not happen:

• Existing data centers do not shut down
• Cloud and enterprise compute demand does not reverse
• Electrification trends (transport, heating, industrial loads) do not stop

As one utility executive recently noted:

“Even the ‘low’ case today is higher than our ‘high’ case five years ago.”

The risk to the grid is not massive overbuild. It is under-building resources that can be delivered on time, especially near load.

Across both aggressive and conservative demand outlooks, one constraint dominates: delivery speed.

Long-lead assets—new gas, large transmission projects, nuclear—remain essential but slow. In contrast, energy storage offers:

• Short development and interconnection timelines
• Capacity value during peak and shoulder hours
• Flexibility to respond to uncertain load growth

This is particularly relevant for data centers themselves. Behind-the-meter or non-export storage:

• Reduces stress on local transmission
• Improves reliability during peak events
• Allows load growth without waiting for system-wide upgrades

These benefits exist whether AI growth accelerates—or simply persists at a lower rate.

AI may prove cyclical. Venture funding, valuations, and deployment curves almost certainly will be.

But data center energy demand is structural.

It reflects a deeper shift in how economies store data, deliver services, and run critical systems. The grid challenge would exist even without AI—AI simply compresses the timeline and raises the stakes.

The real planning error would be assuming that demand uncertainty justifies inaction.

Several signals will matter more than headline forecasts:

• How quickly storage and flexible resources are deployed near load
• Whether data centers are required—or incentivized—to bring capacity with them
• How grid operators adapt planning models to shorter lead-time assets
• Whether permitting and interconnection reform translates into actual delivery

The grid does not respond to narratives.
It responds to assets, locations, and timelines.

AI may ebb and flow.
Data centers are here to stay.
And the infrastructure decisions made in the next 24–36 months will define reliability outcomes for the next decade.

Watch this space.