
Using AI UPS Systems to Tame AI Load Swings
AI data centers are creating a power challenge that goes beyond simply securing enough megawatts. Large AI computing clusters can introduce rapid swings in electricity demand, creating new power-quality and grid-stability concerns for utilities and data center operators. In this interview recorded live at the 2026 Data Center Frontier Trends Summit, DCF Contributing Editor Doug Black speaks with Asser Elsamahy, P.E., vice president of engineering at ON.energy, about how battery-based power systems are emerging as one response.
Elsamahy says battery energy storage itself is not new. Systems have been deployed globally at gigawatt scale for roughly two decades. What is changing is their role inside the data center. As AI workloads introduce larger and faster fluctuations in power demand, batteries and power electronics can increasingly act as a buffer between the compute environment and the electric grid. ON.energy’s approach is what the company calls an AI UPS, or medium-voltage UPS, built around a double-conversion architecture.
Instead of relying on a single set of inverters, the system uses two. One side faces the grid and can respond to utility requirements such as voltage ride-through, frequency ride-through and ramp-rate limits. The other faces the data center load and supplies what the computing equipment requires, including rapid surges and changes in demand. The result is an architecture designed to separate the behavior of the AI load from the behavior presented to the grid. Elsamahy contrasts that approach with a conventional parallel battery energy storage system using a single inverter set.
“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.”
The batteries can also be recharged in multiple ways. A facility can draw modestly more power than its compute load requires during off-peak periods, or it can use downward portions of the AI load profile to replenish stored energy.
Elsamahy describes the battery as a “reservoir of energy” that gives operators greater flexibility over when they draw power and how they respond to changing conditions. That flexibility can also support peak shaving.
Using a hypothetical 1-GW data center, Elsamahy explains that stored energy could allow the facility to reduce its instantaneous grid draw during a period of system stress — potentially lowering a 1-GW demand to 800 MW while continuing to support the internal load.
The goal is not necessarily to disconnect from the grid, he notes. At these scales, the more practical value is having a lever to reduce demand when grid conditions require it. That capability is becoming increasingly relevant as utilities and grid operators scrutinize large-load interconnections more closely.
Elsamahy says ON.energy is seeing strong demand from hyperscalers and data center developers looking several years ahead and designing around requirements such as voltage and frequency ride-through and limits on how quickly loads can ramp. His expectation is straightforward: those requirements will become stricter as AI data centers account for a larger share of electricity demand.
For an industry focused heavily on securing power, the conversation points to another emerging requirement for AI infrastructure: controlling not only how much power a data center consumes, but how that demand appears to the grid. In that environment, battery storage may increasingly become part of the core power-quality architecture of the AI data center rather than simply another backup resource.
More episodes from "The Data Center Frontier Show"



Don't miss an episode of “The Data Center Frontier Show” and subscribe to it in the GetPodcast app.








