Artificial intelligence is famously power-hungry, and the data centers that run it are straining electric grids. A new joint study from chipmaker Nvidia and cloud provider Oracle offers one possible pressure valve.
According to a report from Crypto Briefing, the study found that AI data centers can cut their power draw by as much as 30% during periods of grid stress. In other words, when the wider electricity system is under strain, these facilities could dial back their consumption rather than pushing the grid closer to failure.
The reporting frames this as a flexibility measure: instead of drawing a constant, heavy load around the clock, an AI data center could temporarily reduce demand when the grid most needs relief. Crypto Briefing attributes the 30% figure directly to the Nvidia and Oracle study.
Beyond that headline number, further details from the study — such as how the reduction is achieved, over what time periods, and at what cost to computing performance — are not specified in the available reporting.
Why it matters: as AI workloads keep expanding, the electricity they consume is becoming a real constraint on both the industry and the power grids that serve everyone else, so any credible way to make these data centers more flexible during peak stress could ease tension between AI growth and grid reliability.