ON.energy Deploys Medium-Voltage AI UPS at National Laboratory
- Data Centers Product Standards & Compliance
Marin Holt
- 0
- 5 minutes read
- First medium-voltage UPS designed specifically to absorb AI data center GPU workload spikes at utility scale
- AI UPS handles power surges from 30% to 100% of capacity occurring in 10-millisecond to 1-second intervals
- NLR ARIES platform at Flatirons Campus, Boulder, CO is the only grid simulator worldwide that replicates hyperscale data center power conditions
- A single hyperscale data center consumes energy equivalent to Boulder, CO; load can shift from 0% to 100% in under one second
ON.energy completed construction of a medium-voltage AI uninterruptible power supply at the National Laboratory of the Rockies (NLR) in Boulder, Colorado, in January 2026. The device is the first medium-voltage UPS engineered specifically for the power behavior of AI data centers – facilities whose GPU clusters can surge from near-zero to full load in under one second and repeat that swing multiple times per minute. Testing began on NLR’s Advanced Research on Integrated Energy Systems (ARIES) platform early in 2026, simulating both the data center load and the utility grid it connects to.
What the AI UPS Does
A single hyperscale data center consumes as much energy as a small city like Boulder or 1.5 million laptops charging simultaneously. Unlike those loads, however, a data center’s draw is not steady – it spikes and collapses in milliseconds, a behavior driven by GPU workloads that ramp aggressively during AI training and inference jobs. Standard UPS designs buffer against outages, not against the grid impact of rapid load swings from a facility that scales into the gigawatts. ON.energy’s AI UPS stores energy and actively shapes the load signal so the utility grid sees a smooth, steady demand rather than the actual jagged profile of the data center behind it. During initial testing at ARIES, the system absorbed surges from 30 percent to 100 percent of its rated capacity occurring in 10-millisecond to 1-second intervals. Dax Kepshire, president of ON.energy’s data center division, described the goal: to help hyperscale data centers become good grid citizens by smoothing fast, compute-driven swings so the grid sees a steady, reliable load at gigawatt scale.
The Testing Platform
NLR’s ARIES platform at the Flatirons Campus is described by its operators as the only facility in the world where both a hyperscale data center cluster and a power grid can be simulated simultaneously. As of 2025, ARIES can replicate the electrical characteristics of large-scale GPU compute clusters, including the fast ramp events that make AI workloads difficult for grid operators to manage. Andrew Hudgins, acting program manager for ARIES, stated that this test environment lets ON.energy validate performance scenarios that are not feasible on a live grid. NLR published a follow-up feature on the project in March 2026 describing the testing scope: full-range AI data center power conditions including GPU workload spikes and grid fluctuations in both grid-connected and islanded modes.
Critical Perspective
Laboratory validation and live deployment are different problems. ARIES simulates a data center and a grid; it does not replicate the fault modes, protection systems, and regulatory interfaces of an actual interconnection. The test platform at NLR also operates at a scale considerably smaller than a gigawatt-class hyperscale facility. ON.energy has not announced a commercial deployment with a named utility or data center operator. Whether the AI UPS’s performance at NLR translates to the physical hardware and system integration demands of a real interconnection will require a grid-connected pilot at utility scale – work that has not yet been publicly announced.
Why It Matters
This deployment at the National Laboratory of the Rockies, if proven effective, could significantly alter how massive AI data centers interact with the power grid. By smoothing out the rapid 30 percent to 100 percent load swings observed in testing, it promises to benefit utility providers by reducing grid instability, though the long-term market impact and widespread adoption remain to be seen.
