The power challenge surrounding artificial intelligence is increasingly about more than finding enough megawatts.
AI data centers can also introduce rapid changes in electricity demand as large clusters of accelerators ramp workloads up and down. Those swings create a different kind of infrastructure problem: how to serve highly dynamic compute loads without passing that volatility directly onto the electric grid.
That challenge is helping move battery energy storage deeper into data center power architecture.
In an onsite podcast interview recorded live at the Data Center Frontier Trends Summit 2026, DCF Contributing Editor Doug Black spoke with Asser Elsamahy, P.E., vice president of engineering at ON.energy, about the emerging role of battery-based power quality infrastructure for AI data centers.
Elsamahy said battery power systems themselves are hardly new. Energy storage has been deployed at gigawatt scale around the world for roughly two decades. What is new is the way the technology is being adapted to the operating characteristics of large AI facilities.
“They’re new to the data center industry, but they’re not necessarily new in the market,” Elsamahy said. “They’ve been deployed at gigawatt scale already, multiple gigawatts all over the world.”
The difference now is the load.
Major swings in AI computing demand can create additional stress for grid operators already confronting rapid growth in large-load interconnection requests. Elsamahy said that dynamic is accelerating interest in energy storage as a way to manage the interface between AI infrastructure and the grid.
From Battery Storage to an “AI UPS”
ON.energy’s approach is built around what the company calls an AI UPS, or medium-voltage uninterruptible power supply.
The architecture differs from the parallel battery energy storage system, or BESS, configuration commonly used for standalone grid storage.
ON instead uses a double-conversion design with two sets of inverters.
One inverter set faces the grid and can respond to utility or grid-operator requirements. Those requirements can include voltage ride-through, frequency behavior and limits on how rapidly a large facility ramps power consumption.
The second inverter set faces the data center load.
That allows the load-facing system to deliver the power demanded by the computing environment — including peaks, surges and rapid changes — while the grid-facing side behaves according to the requirements of the utility system.
The result, according to Elsamahy, is a degree of electrical separation between what happens inside the AI data center and what is presented to the grid.
“You have two sets of inverters,” he said. “One set that’s connected to the grid does what the grid wants, follows the grid requirements.”
The other, he continued, is “facing the load” and supplies what that load demands.
That distinction becomes particularly important when compared with a conventional parallel BESS architecture using a single set of inverters.
With one inverter interface, Elsamahy said, the storage system can face competing requirements from the data center and the grid.
“It kind of is in a constant struggle of, do I support the grid or do I support the load?” he said. “I have to pick one or the other. It’s very difficult to do both.”
For data center designers, that points toward a larger change in how batteries may be considered in future AI facilities.
Rather than serving primarily as stored energy for backup operation or standalone grid services, batteries can become part of the facility’s power-quality architecture — shaping the electrical behavior of the data center itself.
Using AI’s Load Swings to Recharge
Any architecture that routinely draws from batteries must also manage state of charge.
Elsamahy described two ways the system can replenish stored energy.
The first is straightforward: during favorable operating periods, the data center can draw slightly more power from the grid than the compute load itself requires, directing the excess into the batteries.
For a hypothetical 1-gigawatt load, for example, the facility might draw modestly above 1 GW when grid conditions and electricity pricing make doing so advantageous.
“Typically you do that during off-peak, when energy price … is usually the lowest,” Elsamahy said.
But the variability of the AI load itself can also provide charging opportunities.
Because compute demand naturally rises and falls, the battery system can absorb energy during downward portions of that load profile and discharge when demand rises again.
“That’s kind of the beauty of the batteries,” Elsamahy said. “It’s basically a reservoir of energy.”
The concept effectively turns battery storage into a buffer between two environments with very different preferences: AI infrastructure that can change power demand abruptly, and an electric grid that operates most reliably when large loads behave predictably.
A New Meaning for “Grid Citizen”
That buffering capability also intersects with one of the industry’s most difficult emerging issues: how rapidly growing data center loads affect utilities and surrounding communities.
Elsamahy described energy storage as one tool for becoming a better “citizen” of the grid.
Peak shaving offers one example.
Residential and commercial electricity consumption tends to rise during predictable periods of the day. A large AI data center drawing its maximum requirement at the same moment can add further pressure to the system.
A sufficiently sized battery reservoir gives the data center another option.
Elsamahy used a hypothetical 1-GW facility to illustrate the point. During a period of grid stress, stored energy might allow the facility to reduce its grid draw to 800 MW while continuing to support the compute load internally.
The numbers would depend on the system design and operating requirements, but the principle is important: storage provides operators with a controllable lever for reducing grid demand without requiring the data center to stop computing.
“We’re able to reduce the power demand that the load is currently asking,” he said, helping “keep the grid stable” while preserving capacity for other consumers.
That does not mean a gigawatt-scale AI campus simply disconnects from the utility whenever demand spikes.
When Black raised the idea of temporarily “getting off the grid,” Elsamahy drew a distinction.
Because of the enormous scale of these facilities, he said, the objective is not necessarily to leave the grid. Instead, batteries allow operators to reduce how much power they demand from it at critical moments.
“You have that lever to reduce your power demand,” he said.
Designing for Tougher Grid Requirements
Elsamahy expects that capability to become increasingly valuable as AI data centers represent a larger share of electricity demand.
He said ON.energy is seeing strong interest from hyperscalers and data center developers looking for ways to manage AI load swings and meet increasingly detailed interconnection requirements.
Among the requirements emerging around large loads are voltage ride-through, frequency ride-through and ramp-rate controls — all designed to ensure that enormous electrical consumers do not amplify disturbances elsewhere on the grid.
Elsamahy expects those standards to tighten.
“My view is that these requirements will only get stronger and stricter,” he said.
That expectation is already affecting projects several years away from operation. Elsamahy said customers planning facilities two or three years into the future are considering these requirements during design rather than waiting until facilities approach energization.
ON.energy is also selling into the hyperscale market, although Elsamahy did not identify individual customers.
The larger signal is that power quality is moving upstream in AI data center planning.
For much of the current infrastructure boom, the overriding question has been where developers can secure enough electricity to support hundreds of megawatts — and increasingly gigawatts — of new compute.
That question is not going away.
But as those loads grow large enough to influence the grids around them, another requirement is moving alongside it: not simply securing the power, but controlling how a data center takes it.
For the next generation of AI infrastructure, batteries may increasingly sit directly at that boundary.
