Blowing Off Steam: How Power-Flexible AI Factories Can Stabilize the Global Energy Grid
Emerald AI collaborated with NVIDIA, EPRI, National Grid, and Nebius to demonstrate how power-flexible AI factories can automatically adjust energy consumption during periods of high electrical demand. By slowing flexible jobs while maintaining high-priority tasks, these facilities can help stabilize power grids without interrupting critical operations. Tests conducted on an AI cluster in London demonstrated complete alignment with grid instructions during simulated stress events.
Key Takeaways
- Emerald AI worked alongside NVIDIA, EPRI, National Grid, and Nebius to test software capable of modulating power usage in AI data centers during times of grid stress.
The management platform successfully throttled secondary tasks to reduce total power draw while keeping critical AI computations operating at full performance.
- Demonstrating this level of load flexibility helps operators manage extreme demand fluctuations without building costly additional power infrastructure.
For developers and system architects, this approach illustrates how hardware monitoring and software scheduling can integrate with utility management to secure faster grid connections for massive compute clusters.
- Dynamic power adjustment allows AI facilities to connect to existing power grids faster while shielding essential workloads from disruption.
- During the experiment, the cluster achieved 100% alignment with over 200 individual power targets issued by energy operators.
- Emerald AI demonstrated how AI data centers can dynamically lower energy usage during peak grid demand.
Stats & Key Facts
- #During the experiment, the cluster achieved 100% alignment with over 200 individual power targets issued by energy operators.
- #A London test cluster utilizing 96 NVIDIA Blackwell Ultra GPUs achieved 100% alignment across more than 200 grid power targets.

Emerald AI worked alongside NVIDIA, EPRI, National Grid, and Nebius to test software capable of modulating power usage in AI data centers during times of grid stress. In a London facility housing a cluster of 96 NVIDIA Blackwell Ultra GPUs, researchers simulated severe energy surges such as sudden consumer demand spikes. The management platform successfully throttled secondary tasks to reduce total power draw while keeping critical AI computations operating at full performance.
During the experiment, the cluster achieved 100% alignment with over 200 individual power targets issued by energy operators. Demonstrating this level of load flexibility helps operators manage extreme demand fluctuations without building costly additional power infrastructure. For developers and system architects, this approach illustrates how hardware monitoring and software scheduling can integrate with utility management to secure faster grid connections for massive compute clusters.
Emerald AI demonstrated how AI data centers can dynamically lower energy usage during peak grid demand. A London test cluster utilizing 96 NVIDIA Blackwell Ultra GPUs achieved 100% alignment across more than 200 grid power targets. Dynamic power adjustment allows AI facilities to connect to existing power grids faster while shielding essential workloads from disruption.
For more details please read the original article at NVIDIA Blog.
Continue Learning
Comments
Comments appear only after moderation. Your email identifies your submission to the moderator and is never displayed here.
No approved comments yet.