AI data centers can stabilize the grid if built right

The explosive growth of AI computing facilities has triggered alarm bells across the U.S. electric grid. When a transmission line failed in Northern Virginia this July, roughly 3,100 MW of data center demand instantly vanished as facilities switched to backup power—a voltage disturbance felt from Washington to Chicago. Two years earlier, approximately 60 data centers in the same region disconnected simultaneously during another incident.

These events prompted the North American Electric Reliability Corp. to issue its highest-level alert in May, reserved for urgent threats to the power system. Federal regulators have since issued new directives for connecting large loads, while communities question whether AI infrastructure will drive up their energy costs and compromise reliability.

But four industry executives argue the technology already exists to make AI facilities beneficial rather than burdensome to the grid. Drew Baglino of Heron Power, Vlad Troy of NVIDIA, Michael Polsky of Invenergy, and Varun Sivaram of Emerald AI propose three obligations that define responsible data center development.

Do no harm through ride-through capability

Today's data centers fail a basic test: when grid voltage dips briefly, backup systems disconnect the entire facility. This transforms a routine fault into a multi-gigawatt shock. The problem compounds as AI workloads create rapid power swings—tens of megawatts in seconds—that stress surrounding infrastructure.

The solution involves batteries that provide backup power to GPUs while simultaneously exporting power to maintain grid stability during disturbances. Facilities that ride through voltage events make the system more resilient, while those that disconnect amplify problems. Building this capability into the infrastructure itself, rather than relying on operator promises, creates reliability the grid can actually count on.

Flex consumption during system stress

For a century, the power system has operated on a one-way assumption: demand does what it wants, and supply must chase it. Utilities build for the hottest afternoon of the year, which is why the grid runs far below capacity most of the time—on average, barely above half.

AI facilities can break this pattern by flexibly controlling power use when the grid is strained. According to Duke University research, if large new loads trim consumption for less than half a percent of their hours each year—during heat waves or cold snaps—the existing grid could absorb roughly 100 GW of new demand years sooner than interconnection queues currently allow. The Brattle Group estimates that a 10% improvement in grid utilization would cut average retail rates by nearly 4%.

Add generation and transmission to the system

The third obligation extends beyond careful power use to actively strengthening the surrounding infrastructure. Many AI projects will begin with co-located generation and storage to achieve faster power delivery. The critical question is whether these resources become permanent islands—assets sized for one customer and walled off from the grid—or bridge power that enables fuller grid integration over time.

In areas with high-priority interregional transmission needs identified by the Department of Energy, AI facilities can catalyze infrastructure investments that deliver reliability and affordability benefits to millions of ratepayers. Customer-funded interregional transmission projects can directly serve AI demand while simultaneously strengthening the shared system.

Why it matters

The U.S. is in the middle of one of the largest infrastructure buildouts in its history. Without intentional design, AI data centers could strain the grid and leave communities bearing the cost. But facilities built according to these principles can steady voltage in their neighborhoods, absorb available capacity in their regions, and lower bills for families and businesses sharing their wires. FERC's June orders directing all six regional grid operators to justify or reform their large-load tariffs put this framework on a national timeline, explicitly rewarding ride-through capability, flexibility, and grid enhancement.

Regulators and utilities can accelerate progress with performance-based rules that reward demonstrated behavior rather than mandating specific equipment. Silicon Valley Power and the Electric Reliability Council of Texas have begun developing such frameworks, showing what good citizenship looks like in practice.

These details were first reported by Utility Dive in an opinion piece by the four industry officials.