Quantum AI framework to reduce data centre energy consumption

A new quantum computing-based optimisation framework could reduce energy consumption in large AI workload data centres by up to 12.5%.

The new framework, which was developed by researchers at Cornell University, is also calculated to reduce the data centre’s carbon emissions by almost 10%.

Data centre energy consumption is currently relatively small, estimated by the IEA in mid-2023, a little over 1% of global electricity use.

However, the percentage is expected to grow rapidly as AI and generative AI in particular becomes more widely adopted.

Have you read?
Quantum computing demonstrated for grid battery placement
The good, the bad and the safety: AI and data in the energy sector

The quantum computing-enabled optimisation framework, detailed in the journal Advances in Applied Energy, aims to address this challenge by integrating variational quantum circuits with classical optimisation to enable efficient and uncertainty-aware control of energy systems.

These include uncertainties associated with weather conditions and renewable energy generation while optimising the energy consumption in the AI data centres.

“By developing quantum computing-based AI methods, we are tackling the pressing energy and climate challenges faced by AI data centres, significantly enhancing their sustainability and efficiency,” explains Fengqi You, professor in Energy Systems Engineering and co-director of the Cornell University AI for Science Institute, who developed the framework with doctoral student Akshay Ajagekar.

“Quantum AI methods for molecular design highlight the transformative potential of AI for science, driving scientific discoveries, enabling sustainable solutions and fostering new innovations.”

Computational experiments were conducted at data centres at various locations in the US, demonstrating the framework’s ability to significantly reduce power consumption and carbon emissions associated with AI data centre operations.

Additionally, the researchers report that the superiority of the framework is confirmed from a comparative study with other methods and offers a computationally efficient and scalable approach for control computation in contrast to its classical counterparts.

Based on this work, several US patent applications have been filed.