The NVIDIA-backed startup says its software can cut data center power consumption by up to 40% during peak grid stress without sacrificing AI performance.
Picture this: millions of English soccer fans flip on their electric kettles at halftime, and somewhere in London, an AI program quietly tells a data center to ease off the throttle. No blackouts, no fried transformers, no drama. That’s the pitch from Emerald AI, a Virginia-based energy-tech startup that wants to make data centers behave less like power-hungry monsters and more like good neighbors on the electrical grid.
The company’s Conductor software platform dynamically manages workloads, onsite energy resources, and real-time grid signals to reduce power consumption by 25-40% during peak stress events. The kicker: it claims to do this without any hardware changes and without sacrificing AI computational performance.
What Conductor actually does #
In a May 2025 test conducted in Phoenix, the Conductor platform achieved a 25% power reduction sustained over three hours while utilizing 256 NVIDIA GPUs during peak grid loads.
Emerald AI has also run demonstrations in the UK, where the software delivered over a one-third power reduction in under 60 seconds. For context, that UK test was reportedly tied to the kind of sudden demand spikes caused by events like halftime tea-making surges during soccer matches, where millions of kettles clicking on simultaneously can strain the grid.
The company has launched additional pilots with Silicon Valley Power in Santa Clara, suggesting the technology is being validated across multiple grid environments and geographies.
The Virginia bottleneck #
Northern Virginia is the data center capital of the world, home to the densest concentration of computing facilities on the planet. It’s also increasingly grid-constrained, with new data center projects facing extensive delays just to get connected to the electrical grid.
Emerald AI’s flagship project, the Aurora AI Factory, is a 96 MW facility in Manassas, Virginia, built in collaboration with NVIDIA and Digital Realty. Commercial operations are expected to begin in the first half of 2026. The facility is designed from the ground up to be power-flexible, meaning it can modulate its grid consumption in real time based on local conditions.
Who’s backing this #
Emerald AI raised $24.5 million in seed funding in July 2025. The investor list includes NVIDIA’s venture arm NVentures, along with notable figures including Jeff Dean and Fei-Fei Li.
The company was founded in 2024 by Varun Sivaram, who serves as CEO and operates from headquarters in Clarendon, Virginia.
What this means for crypto and energy-intensive computing #
Emerald AI doesn’t operate in the crypto space directly. There are no tokens involved, no blockchain components, no DeFi integrations. But the implications for energy-intensive sectors like cryptocurrency mining are hard to ignore.
Bitcoin mining operations face the same fundamental challenge: they consume enormous amounts of electricity, and their relationship with local grids is often contentious. Some miners have already experimented with demand-response programs, voluntarily curtailing operations during peak grid stress in exchange for lower energy rates or other incentives. Conductor’s approach, automating that flexibility at the software layer, represents a more sophisticated version of what progressive mining operations have been doing manually.
A data center or mining operation that can credibly promise to reduce its draw by 25-40% during emergencies is a very different regulatory conversation than one that insists on constant peak consumption.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our