ON.energy

ON's AI UPS™ exceeds ERCOT's voltage ride-through compliance requirements. Validated at the National Laboratory of the Rockies   North American Electric Reliability Corporation (NERC) has identified ride-through as a key reliability issue for large loads. ERCOT is actively advancing voltage and frequency ride-through requirements for new interconnections. When the grid faults, traditional UPS systems disconnect. At gigawatt-scale AI infrastructure, that's not a power quality issue. It's a grid reliability issue.   Left: A conventional low-voltage UPS. Graphed against ERCOT's Table A requirements. It disconnects where the grid needs it most.   Right: ON's AI UPS™. It rides through faults and exceeds the compliance envelope.   AI needs power. We have the solution.

The New Standard for Datacenter Power Solutions. This week, our CTO Ricardo de Azevedo presented at CERAWeek, sharing ON’s validated results from live AI load testing on the ARIES platform at the National Laboratory of the Rockies Flatirons Campus: •⁠ ⁠100% ripple mitigation •⁠ ⁠Full IEEE 2800 and ERCOT LLVRT compliance •⁠ ⁠75% more cost-effective than a combined LV UPS + standalone BESS stack. AI load volatility can be contained before it reaches the grid with infrastructure designed for what AI actually demands.

"We have the opportunity to build the grid of the future. It’s often framed in terms of what could go wrong, but we don’t talk about what could go right.” – ON CEO, Alan Cooper. Explosive AI-driven data center demand is colliding with the realities of timelines required to build physical infrastructure, secure permits, expand transmission, and plan generation. Yesterday, Alan joined Patrick Ryan (Meta), Bill Brown (New Water Capital, L.P.), and moderator Ashutosh Singh (S&P Global) at CERAWeek to discuss how we can power AI at scale. Voltage ride through failures and AI load ripple are real risks when you bring GW scale data centers online, but there are real solutions. Facilities can demonstrate ride-through, offer credible grid services, and absorb risk. That’s our vision for the future: flexible data centers that can keep the grid safe.

Most coverage of AI power asks: can the grid handle it? Our CEO, Alan Cooper has been asking a different question: what does it take to build data centers that are actually safe for the grid? He put it in writing.

On the agenda for Tuesday, March 24 at CERAWeek.    12:30–1:00 PM CST | "Powering AI at Scale": CEO Alan Cooper joins Patrick Ryan (Meta), Bill Brown (New Waters Capital LP), and moderator Ashutosh Singh (S&P Global) to discuss what it actually takes to source power as a competitive advantage—bringing your own capacity, compressing timelines, and designing for grid-safe requirements before they become mandates. 3:30 PM CST | Pod 4 — CTO Ricardo de Azevedo presents "The New Standard for Grid-Safe AI Power": a technical deep dive on the AI UPS™, including validated NLR test results showing 100% ripple mitigation and industry-leading ride-through performance under real AI load profiles.  The power demands behind AI are accelerating. We'll be in talking about the solution.  See you in Houston. 

Last week, our team was at Shoals Technologies Group's Mega manufacturing campus for the ribbon cutting on a new U.S. production line built to support our work together. AI and hyperscalers aren't asking for more power someday. They need it now, at scale, and the grid wasn't built to deliver it. That pressure isn't just reshaping demand. It's reshaping how power infrastructure gets engineered, sourced, and built. U.S.-based manufacturing. Serious power architecture. Built for AI. 

In February we completed live testing of ON's MV AI UPS™ at the National Laboratory of the Rockies (NLR), the only facility in the Western Hemisphere with both a grid simulator and a data center simulator. During testing we introduced load swings from 30% to 100% of capacity in 10-millisecond to 1-second intervals. Blink-of-an-eye fast. The upstream grid profile remained flat. The system absorbed the transients in real time while maintaining a controlled 0.33% per-second ramp rate and stable state of charge. The AI UPS will remain on-site at NLR's Flatirons Campus for at least two years. More test scenarios. More data. More validation. AI load volatility can be contained before it reaches the grid. This is what grid-safe data centers look like. This is the new standard for AI power. Read more: https://lnkd.in/gtmfEGdA

The left side of this graphic is what happens when you connect conventional energy storage to an AI data center load. The right side is what happens when you connect ON’s MV AI UPS™. Same load. Same grid. Fundamentally different outcome. Here's what you're looking at: ▪️GPU workloads cycle constantly, rapidly and at high frequency. That cycling creates a power ripple that has to go somewhere. ▪️With conventional energy storage, it goes back to the grid. That oscillating utility meter on the left isn't the load being unpredictable. It's the grid absorbing instability it was never designed to handle. ▪️With ON's AI UPS, the ripple is suppressed at the source. The utility meter stays flat. The grid never sees what the GPUs are doing. GPU ripple suppression isn't a feature on a spec sheet. It's the difference between a grid-safe data center and one grid operators won't interconnect. The grid doesn't have to see what the GPUs are doing. That's what grid-safe data centers look like.

The graphs below illustrate ERCOT’s Low-Voltage Ride-Through (LVRT) envelope alongside representative performance characteristics of different UPS architectures. The comparison highlights the difference between: • What ERCOT requires from large loads • What traditional low-voltage UPS systems were designed to do • What ON.energy’s medium-voltage AI UPS™ platform enables The gap isn’t marginal. It’s structural. Earlier this week, Katherine Blunt and Jennifer Hiller reported in the The Wall Street Journal that nearly 2,000 MW of Virginia data center load dropped offline during transmission faults. Not a simulation. Not a stress test. A real grid event that highlighted what happens when large loads disconnect during disturbances. Grid operators are taking notice. ERCOT already requires large transmission-connected loads to meet defined ride-through performance envelopes. Other ISOs are moving in the same direction. The technical mandate is converging on one principle: When the grid faults, large loads stay connected. ON.energy’s medium-voltage AI UPS™ platform exceeds even the most stringent ride-through requirements. Traditional low-voltage UPS systems were designed to disconnect. The MV AI UPS™ platform enables ride-through across the full voltage-time envelope, eliminates GPU ripple at the grid interface, and enables coordinated campus-scale fault response. Grid-safe architecture is no longer a differentiator. It is becoming the new standard.

Designing grid-safe power for hyperscale AI sites means managing megawatt-scale swings in milliseconds. GPU-driven volatility is redefining system design, and conventional architectures weren’t built for it. We recently completed live testing of ON.energy’s medium-voltage AI UPS™ at the National Laboratory of the Rockies. During testing, we introduced >70% instantaneous load swings to simulate GPU volatility. The upstream grid profile remained flat. The system absorbed the transients in real time while maintaining a controlled 0.33% per-second ramp rate and stable state of charge. Real technology. Real-world conditions. AI load volatility can be contained before it reaches the grid. The new standard for AI power has arrived.