Power-flexible Artificial Intelligence factories tested as grid stabilizers

Power-flexible Artificial Intelligence factories are being tested as responsive grid assets that can cut power use during demand spikes while keeping priority workloads running. Trials in the U.K. suggest the approach could speed grid connections for large computing sites and reduce pressure for costly infrastructure expansion.

Millions of viewers in the U.K. turned on kettles at halftime during the UEFA EURO 2020 round of 16 match between England and Germany, creating a sudden surge in electricity demand. National Grid saw a demand spike of about 1 gigawatt in a matter of minutes, highlighting how quickly grid operators must respond to keep the system stable as large new electricity users come online.

Emerald AI, working with NVIDIA, EPRI, National Grid and Nebius, presented a model for “power-flexible” Artificial Intelligence factories that can automatically reduce power use during periods of grid stress. Following proof-of-concept trials at Artificial Intelligence factories in Arizona, Virginia and Illinois, the group brought the Emerald AI Conductor Platform to Nebius’ new Artificial Intelligence factory in London. The team ran production-grade Artificial Intelligence workloads on a cluster of 96 NVIDIA Blackwell Ultra GPUs connected through the NVIDIA Quantum-X800 InfiniBand platform, while NVIDIA System Management Interface provided seconds-level GPU power telemetry.

EPRI and National Grid simulated grid stress events ranging from lightning strikes to extended periods of low wind power, then instructed the facility to temporarily cut consumption. One test recreated the football match “TV pickup” event. As millions of simulated kettles were about to switch on, the Artificial Intelligence cluster reduced its power use and absorbed the abrupt demand swing without disrupting the highest-priority workloads. In the Nebius demonstration, higher-priority tasks maintained peak throughput while more flexible jobs were slowed temporarily.

Emerald AI recorded 100% alignment with over 200 power targets that EPRI and National Grid instructed the Artificial Intelligence cluster to follow for this experiment. The companies said the result shows that flexible computing loads can help grids use existing capacity more efficiently, reducing the need to overbuild infrastructure for worst-case peaks and helping keep consumer electricity costs manageable.

National Grid said the tests extended beyond earlier U.S. work by covering not only GPUs, but also CPUs and the total power consumption of the surrounding IT equipment. The approach is positioned as a way to help London connect large new customers despite infrastructure constraints, while supporting broader economic growth tied to computing investment in the U.K. After four demonstrations, Emerald AI and NVIDIA are preparing for real-world deployment at the Aurora AI Factory in Virginia, set to open this year.

57

Impact Score

Intel launches Xeon 600 workstation processors

Intel has launched its Xeon 600 series Granite Rapids-WS processors for workstations and high-end desktops, with a focus on Artificial Intelligence development, AVX-512, and large PCIe connectivity. The lineup scales up to 86 Redwood Cove performance cores and supports high memory and I/O capacity.

Open and proprietary Artificial Intelligence are converging

NVIDIA framed the future of Artificial Intelligence as a mix of open and proprietary models, with orchestration, specialization and collaboration shaping how systems are built. New coalition efforts aim to expand open frontier models while supporting enterprise and research use cases.

Agentic commerce depends on trusted context

Agentic commerce shifts digital systems from recommendation to execution, raising the bar for identity, permissions, and data quality. Organizations need deterministic entity resolution and real-time context to let agents act safely at machine speed.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.