Medium-Voltage UPS for Data Centers

ON.energy’s medium-voltage UPS protects AI data centers from power disturbances and shields the grid or onsite generation from AI load swings. Currently deploying at gigawatt scale, ON’s MV UPS is proven to meet emerging large-load voltage ride-through requirements.

3 GWDeployed
9.6 MWNLR Testbed Validation
ZeroTransfer time
1–8+ hrsBackup duration
Consult the Datasheet

AI UPS Frequently Asked Questions

An MV UPS acts as an electrical firewall by continuously decoupling the facility load from the grid. In ON.energy’s inline, double-conversion architecture, power passes through a DC link and controlled power converters; integrated energy storage supplies or absorbs transient energy. Grid disturbances remain upstream, while rapid AI load changes remain downstream.

Faults, sags and swells stop at the UPS input. Because all power is continuously converted through a DC link, the load never sees the disturbance and there is no transfer event. Integrated energy storage carries the facility through outages for 1 to 8+ hours, depending on configuration, with zero transfer time.

AI workloads can swing power by ±70% of interconnection capacity within milliseconds. Placing integrated energy storage between the facility and the grid absorbs those swings, with less than 1% battery state of charge variation, and presents a smooth, ramp-rate compliant profile to the grid regardless of what the GPUs are doing.

Voltage ride-through is the ability to stay connected and keep operating through grid voltage disturbances instead of tripping offline and risking cascading failures. Large-load rules such as ERCOT’s LLVRT standards are making it a requirement. AI UPS™ rides through 0 V for the duration of the BESS (1+ hour) and sustains 120% voltage for 12 seconds.

A hall-level UPS protects one room from inside the building, so the campus still looks like a large, volatile load to the grid. A site-boundary UPS sits upstream at medium voltage, where the site meets the grid, decoupling the whole facility and handling grid compliance in one place. It is outdoor rated (IP55) and scales in 3.5 MW lineups to multi-GW sites.

Talk with our engineering and sales teams to go over specs, ride-through performance, and your data center’s power architecture and interconnection plan. and we’ll connect you with a team member.

The AI / Grid Paradox

As data center workloads shift toward AI and HPC, operators face a dual challenge: protecting sensitive, volatile compute loads from grid disturbances while ensuring the facility itself does not destabilize the local grid.

AI UPS™ solves this paradox — decoupling the load from the grid so that massive compute transients never disrupt the grid, and grid faults never disrupt the compute.

Converging Power Requirements

Rapid Load Transients

Rapid Load Transients

GPU clusters ramp from idle to full load in milliseconds, creating power swings beyond traditional system tolerance.

Zero Tolerance for Interruption

Zero Tolerance for Interruption

Training runs worth millions in compute time can be lost to momentary power disturbances.

Unprecedented Scale

Unprecedented Scale

Giga-campus deployments demand multi-GW power systems that break the traditional low-voltage UPS architecture mold.

Voltage Ride-Through

Voltage Ride-Through

Facilities must remain connected during grid disturbances rather than disconnecting and causing cascading failures.

Frequency Ride-Through

Frequency Ride-Through

Similar requirements apply during frequency excursions across both 50 Hz and 60 Hz systems.

Ramp Rate Control

Ramp Rate Control

Large loads cannot appear or disappear instantaneously from the grid's perspective.

Traditional Approaches Fall Short

Conventional architectures rely on low-voltage UPS systems without grid-interactive capabilities, static transfer switches with non-zero transfer times, generators with seconds-long startup delays, and separate battery storage that doesn't provide UPS functionality. This fragmented approach creates protection gaps, compliance challenges, and missed economic opportunities.

Unified Power Protection
and Grid Integration

AI UPS™ delivers a comprehensive solution through a single integrated system — continuously conditioning all power flowing to the facility while presenting a compliant profile to the grid.

CapabilityPerformance
Transfer TimeZero (continuous power conversion)
Power QualityComplete isolation from grid disturbances
Backup Duration1–8+ hours (configurable)
Scalability3.5 MW per lineup to multi-GW per site
RedundancyN+1 to 2N+1 configurations
InstallationOutdoor-rated (IP55), upstream of facility
0 ms
Transfer Time
IP55
Environmental Rating
24/7
Monitoring & Support
>99.9%
Uptime Performance

Proven at the National
Laboratory of the Rockies

AI UPS™ has been validated at full scale through the ARIES (Advanced Research on Integrated Energy Systems) platform - a world-class research infrastructure operated by the National Laboratory of the Rockies at its Flatirons Campus in Colorado.

ON.energy AI UPS™ system deployed at the NLR Flatirons Campus

The ARIES Flatirons Campus provides hardware-in-the-loop testing at up to 20 MW with voltage levels reaching 13.2 kV - representing the real-world distribution and bulk power system interface. The facility offers two controllable grid interfaces: a 7 MW grid simulator that replicates utility grid behavior including disturbances and voltage ride-through events, and a 20 MW load simulator with 400-microsecond response times that reproduces real-world demand dynamics. AI UPS™ connects at medium voltage to both interfaces.

This enables ON.energy to prove AI UPS™ transient response and load smoothing performance under the most demanding conditions — including voltage ride-through events, frequency excursions, and rapid load transients — under controlled, repeatable conditions with GPS-synchronized medium-voltage data acquisition at every node.

7 MW
Grid Simulator (grid disturbances & ride-through)
20 MW
Load Simulator (400 µs response time)
9.6 MW
Three Full Lineups Under Testing
13.2 kV
Medium-Voltage Connection

Why This Matters

Unlike lab-scale demonstrations, the Flatirons Campus tests AI UPS™ at the actual voltages, power levels, and grid conditions it will face in production deployments. ON.energy is running a full 3.2 MW branch through the most stringent load profiles in the industry, proving transient response and load smoothing performance at scale. The facility's digital real-time simulators - the largest in any R&D environment - scale test results to represent community and regional grid configurations, validating performance across every scenario before a single unit ships to a customer site.

Voltage & Frequency
Ride-Through

AI UPS™ meets and exceeds the most stringent ride-through requirements currently proposed or in effect, including ERCOT's Large Load Voltage Ride-Through (LLVRT) standards.

AI UPS™ (MV UPS)

AI UPS™ dramatically exceeds ERCOT compliance - riding through 0V continuously for the duration of the BESS (1+ hour), and sustaining 120% voltage for 12 seconds.

Exceeds Compliance
AI UPS™
ERCOT LL
Nominal

LV UPS (Competitor)

Traditional low-voltage UPS systems fail to meet ERCOT's Large Load ride-through requirements - leaving critical areas of non-compliance at low voltages.

Non-Compliance
LV UPS
ERCOT LL
Nominal

Real data from our NLR full-scale testbed

During a voltage ride-through event, the grid voltage drops to zero (true ZVRT event) while the AI UPS™ maintains perfectly stable load voltage at ~1.06 p.u. throughout. 10 ms timestep. Drag to zoom, double-click to reset.

Grid Voltage (p.u.)
Load Voltage (p.u.)

Grid Voltage on graph doesn't drop to 0 due to location of the meter at Medium Voltage.

Grid Power (MW)
Load Power (MW)

AI Workload Transient
Management

Modern AI workloads create power swings of ±70% of interconnection capacity within milliseconds. AI UPS™ completely decouples the facility's internal load behavior from its grid-facing power profile.

Absorb

Absorb

Rapid load changes absorbed using integrated energy storage with minimal (<1%) battery state of charge variation.

Smooth

Smooth

Presents a smooth, predictable power profile to the grid regardless of internal workload behavior.

Comply

Comply

Maintains ramp rate compliance automatically, protecting grid infrastructure from stress.

Real data from our NLR full-scale testbed

During rapid AI workload transients, the AI UPS™ absorbs load power swings using integrated energy storage, presenting a smooth grid-facing power profile. 10 ms timestep. Drag to zoom, double-click to reset.

Grid Voltage (p.u.)
Load Voltage (p.u.)
Grid Power (MW)
Load Power (MW)
Load Frequency (Hz)

Ready to Transform Your Power Architecture with AI UPS™?

For technical inquiries, system sizing, or project discussions, our engineering team is ready to design a solution for your facility.

Modular Design
at Any Scale

Each data hall receives an independent AI UPS™ system with built-in redundancy - ensuring fault isolation, independent maintenance, and scalable deployment.

Utility Grid
Utility Grid
Medium Voltage Feed
→
AI UPS™
AI UPS™
3.5 MW Lineups
N+1 to 2N+1
→
Facility
Facility
Clean, Uninterrupted Power
SpecificationValue
Environmental RatingIP55 (outdoor rated)
Installation LocationUpstream of facility switchgear
DesignCompact modular
CoolingSelf-contained, no building integration
MaintenanceFull access from grade level
Base Module3.5 MW lineup
Site ScalabilityUnlimited - multi-GW supported
Backup Duration1–8+ hours per configuration

Beyond Backup:
Grid Services Revenue

AI UPS™ transforms what is traditionally a cost center into a revenue generator through participation in grid services.

Peak Shaving

Reduce facility demand charges by limiting grid draw during peak periods. Store low-cost energy during off-peak hours.

15–30% demand charge reduction

Demand Response

Participate in utility demand response programs and provide grid support during system emergencies.

50–200 /kW-year

Energy Arbitrage

Charge during low-price periods, discharge during high-price periods. Particularly valuable in ERCOT and CAISO markets.

Market-dependent

Ancillary Services

Frequency regulation, spinning reserves, and voltage support across multiple market structures.

$20–100 /kW-year

Total Cost of Ownership

FactorTraditional UPS + BESSAI UPS™
GPU RippleBuffering with ms delay100% filtering
Transfer Time10–20 ms (STS)Zero
Grid ComplianceRequires additional equipmentIntegrated
Revenue GenerationLimited (BESS only)Multiple streams
ScalabilityLimited by LV architectureLinear to multi-GW
MaintenanceSeparate UPS & BESS systemsSingle integrated system
Space RequirementsIndoor UPS + outdoor BESSOutdoor only

Industry Standards
Compliance

Grid Interconnection

  • IEEE 1547 - Distributed Energy Resources
  • ERCOT NOGRR282 - Large Load Requirements
  • PJM, CAISO, MISO regional requirements

Data Center

  • Uptime Institute Tier III/IV
  • TIA-942 Infrastructure Standards
  • ASHRAE TC 9.9 Environmental Guidelines
  • IEC 62040-3 Class 1 UPS Performance

Electrical Safety

  • UL 1741 - Inverters & Interconnection
  • UL 9540 - Energy Storage Systems Safety
  • IEEE 1584 - Arc-Flash Hazard
  • NFPA 70 - National Electrical Code

Proven Technology
Foundation

ON.energy builds and operates hyperscale power systems to solve the toughest resilience challenges. AI UPS™ is built on grid-scale power conversion and energy storage technology deployed in the hundreds of gigawatts worldwide.

3 GW+
Operating & Under Construction
10+ GW
Production Capacity
75+
Projects Across Americas
>99.9%
Uptime Performance
20 yr
Tier-1 Component Warranties
U.S.
Manufacturing, FEOC Compliant