06/09/2024
How much power is AI consuming?
Compute clusters — essential for building and running AI models — require significant energy, leading to increased power capacity demands and challenges for data center operators, particularly in power distribution and cooling.
However, some AI power consumption estimates circulating in the media and at conferences are notably high. If accurate, they may indicate serious infrastructural and regulatory challenges ahead, but exaggerated forecasts could also provoke unnecessary regulatory responses.
For example, Schneider Electric projects AI power demand to reach 4 gigawatts (GW) in 2023, translating to 35 terawatt-hours (TWh) annually, and increasing to 15 GW (131 TWh annually) by 2028. Similarly, Alex de Vries of Digiconomist estimates that AI workloads could consume between 85 TWh and 134 TWh by 2027, potentially boosting global data center power demand by 30% to 50% or more.
On the other hand, the Uptime Institute offers a more conservative outlook. They estimate that in the first quarter of 2024, Nvidia systems supporting generative AI will use around 5.8 TWh annually. However, this figure is expected to rise rapidly if Nvidia meets its sales targets, with the generative AI infrastructure potentially consuming 21.9 TWh annually by the first quarter of 2025. This supports the view that the initial impact of AI is limited.
What do you think?
PS: If you’re looking for a place to host your computing cluster, feel free to reach out to us.
In the past year, Uptime Intelligence has been asked more questions about generative AI and its impact on the data center sector than any other topic. The questions come from enterprise and colocation operators, suppliers of a wide variety of equipment and services, regulators and the media. Most of...