Blowing Off Steam: How Power-Flexible AI Factories Can Stabilize the Global Energy Grid

Emerald AI released a white paper in collaboration with NVIDIA, National Grid, EPRI, and Nebius demonstrating how "power-flexible" AI factories can autonomously adjust electricity consumption during grid stress events. The technology addresses growing concerns about data center power demands overwhelming electrical grids worldwide.

The concept draws inspiration from the "TV pickup" phenomenon observed during Euro 2020, when millions of UK viewers simultaneously turned on kettles during halftime. This caused a sudden 1-gigawatt demand spike equivalent to a nuclear reactor's output in just minutes.

AI factories equipped with this technology can act as shock absorbers for abrupt power surges by temporarily reducing consumption. Grid operators can signal facilities to ramp down during stress events like lightning strikes or low wind periods.

The London demonstration at Nebius' AI factory used 96 NVIDIA Blackwell Ultra GPUs connected through the NVIDIA Quantum-X800 InfiniBand platform. The Emerald AI Conductor Platform ensured high-priority AI workloads maintained peak throughput while flexible jobs slowed during power reductions.

The system achieved 100% alignment with over 200 power targets set by EPRI and National Grid during testing. The London trials went beyond U.S. demonstrations by measuring total IT equipment consumption including CPUs and supporting infrastructure.

For AI facilities, this technology unlocks significantly faster grid connections without requiring massive infrastructure upgrades that typically take years. For everyday consumers, it helps limit expensive grid buildouts and keeps electricity rates affordable.

National Grid executives see this as enabling the UK to attract hyperscaler investments and unlock economic growth despite operating at a smaller scale than U.S. data center markets.