Crusoe bets 5 GW on medium voltage AI UPS

Crusoe will deploy 5 GW of ON.energy medium-voltage battery UPS to smooth AI power swings and meet ERCOT ride-through rules.

Outdoor modular battery energy storage units at a hyperscale AI data center campus
Crusoe plans 5 GW of ON.energy medium-voltage AI UPS from 2026· Crusoe Blog

Crusoe plans to roll out 5 GW of medium voltage AI UPS from ON.energy across its hyperscale campuses beginning in 2026, a system built to ride through grid disturbances rather than trip offline.

July turned that risk into a hard lesson. A downed line in northern Virginia knocked more than 3 GW offline in roughly 30 seconds, twice the size of a 2024 incident, and the voltage swing rippled all the way to Chicago. PJM needed 10 minutes to rebalance the grid.

Crusoe casts its choice as a design decision, not a compliance checkbox. AI clusters swing from tens to hundreds of megawatts in under a second as thousands of GPUs bounce between compute and communication, and traditional low-voltage UPS systems answer that by disconnecting entirely.

At gigawatt scale, those simultaneous disconnects start to become the disturbance themselves.

How the battery replaces dummy compute

Running dummy matrix multiplies to flatten load chews through close to 6% of energy, about 480 GWh a year at 1 GW, plus another 192 GWh for cooling, for roughly 672 GWh wasted. Firmware power floors at 80 to 90% of TDP still bleed around 7%, roughly 600 GWh a year, and leave residual swings.

ON.energy's battery sits at medium voltage between the grid and the campus, soaking up the dip from 100% to 30% and discharging on the spike so the grid sees a flat profile without any fake compute.

Why staying connected is the real test now

ERCOT's NOGRR 282 now requires large loads to ride through faults instead of tripping, and a standalone battery storage system can still get capped on injection to dodge generator reclassification. Crusoe says its setup decouples grid support from campus continuity with ramp-rate-controlled inverters, keeping the site online while holding the meter steady.

Commissioning from 2026 into 2027 will show whether this actually trims total draw at gigawatt scale, as US data centers head toward 9% to 17% of national electricity by 2030.

What is still missing is price, round-trip efficiency, and cycle life. Without those numbers, outsiders can't weigh the waste avoided against battery degradation, even if the grid benefit looks obvious.

© 2026 StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training requires a license. See our terms.
Daniel Singer

Written by

Daniel Singer

Editor, StartupHub.ai

Daniel Singer is the editor of StartupHub.ai, a technology expert and thought leader on AI and its applications across sectors, from fintech and healthcare to developer tooling and consumer software. He writes and tests the tools covered here thoroughly and regularly, and built StartupHub.ai to give founders, operators and buyers a clearer read on what they are actually being sold.

More from Daniel Singer

Startups in this story

Profiles for the companies named above.