AIComputers

Nvidia: Only 60-70% of Data Center Power Reaches AI

Nvidia and Vertiv said at Vertiv's EMEA event that 30-40% of the power an AI data center draws is lost to cooling, conversion, misconfigured racks and restarts.

Nvidia and Vertiv said at Vertiv's annual EMEA event on October 14, 2026 that AI data centers use only about 60% to 70% of the electricity they draw from the grid for AI processing, according to techradar.com.

Rod Evans, Nvidia's Vice President of Supercomputing, AI Cloud Infrastructure and AI/HPC for EMEA, said at most 70% of drawn power goes to AI work and it can be as low as 60%. Nvidia attributes 20% of a one-gigawatt data center's power use to cooling, power distribution and conversion, with a further 10% lost to incorrectly configured racks, failures and restarts.

Evans also said that of a contracted 100 megawatts a facility may only receive around 93 megawatts, and less at certain times of day. He added that the main bottleneck for AI data center capacity is now a facility's connection to the grid rather than a chip shortage.

Vertiv EMEA president Paul Ryan said power availability, permits and remote locations each pose separate obstacles. The two companies said they plan to build more efficient and sustainable AI data centers focused on getting the most out of every watt.

Existing Collaboration

Vertiv and Nvidia have been working together since early 2026 on co-designing next-generation data centers, with Vertiv providing power and infrastructure management for Nvidia's GPUs.

Quick answers

How much of an AI data center's power is used for AI processing?

Nvidia's Rod Evans said at most 70% of the power drawn is used for AI work, and it can be as low as 60%.

What happens to the rest of the power an AI data center draws?

Nvidia attributes 20% of a one-gigawatt data center's power use to cooling, power distribution and conversion, and a further 10% to incorrectly configured racks, failures and restarts.

What did Nvidia and Vertiv say they plan to do?

The two companies said they plan to build more efficient and sustainable AI data centers focused on getting the most out of every watt.

Source