Nvidia Says AI Data Centers Waste Up to 40% of Grid Power as Power Becomes Main Bottleneck

For a one-gigawatt data center, Evans said 20% is lost through cooling, power distribution and conversion, and a further 10% is lost to incorrectly configured racks, failures and restarts. He said: "You sign a contract for 100 megawatts, but you never actually get that. With a bit of luck, you'll get 93 megawatts, and at certain times of the day, even less."

Nvidia believes some of that 30%, which can reach 40%, can be recovered. Evans said data centers are designed for peak load and outages rather than for what they typically use. That means unused capacity can be tapped and power and cooling configurations can be revised to find more efficient settings, according to the report.

The comments were shared with data center builder Vertiv at its recent annual EMEA event. Vertiv and Nvidia have been collaborating since early 2026 on co-designing next-generation data centers, with Vertiv providing power and infrastructure management for Nvidia's GPUs. The two companies plan to assemble sustainable facilities. Evans said: "Together, we want to build sustainable AI data centers where the focus isn't on using as much power as possible, but on getting the most out of every watt."

A key point from the press conference was that the main bottleneck for AI data centers is no longer a shortage of chips but a lack of power, TechRadar reported. The constraint is producing underpowered data centers, and some operators are waiting several years to receive the electricity needed to match their capacity.

Location can affect the problem. Vertiv EMEA president Paul Ryan explained that in some places power is available but there is no permit. In another place, he said, a permit is granted quickly but there is no power. Sometimes both are available, he added, but the location is so remote that equipment can barely be delivered there.