Nvidia, Oracle trials show AI data centers cut power 30% during grid stress
In brief
- GPU clusters reduced power 25-35% within 30 seconds of grid signal
- AI workloads maintained performance despite significant power cuts in trials
- Data center flexibility could support 100 GW additional US grid load
How the trials worked
The first trial took place on May 3, 2025, at an Oracle Cloud data center in Phoenix, Arizona, where a cluster of 256 Nvidia GPUs achieved a 25% reduction in power consumption sustained over three hours. AI workloads continued running without compromising service quality or throughput despite the reduction — the flexibility didn't degrade performance.
A follow-up trial in December 2025 at a Nebius data center in London with 96 Nvidia Blackwell Ultra GPUs achieved a load reduction exceeding 30%, reaching approximately 35% at its peak. The speed was remarkable. Power draw decreased within just 30 seconds of receiving a grid signal from National Grid and was sustained for up to 10 hours.
The technology behind the shift
Emerald Conductor is a platform developed by Emerald AI that dynamically shifts or pauses tasks across GPU clusters without human intervention. When grid operators signal peak demand periods, the software rebalances workloads across available infrastructure — keeping AI applications running while reducing the total draw. The orchestration happens in real time.
The concept being tested falls under EPRI's DCFlex initiative, which explores how data centers can function as dynamic grid resources rather than static energy sinks. This reframing is crucial. Instead of viewing data centers as fixed loads, grid planners can now model them as flexible participants in demand management.
Grid capacity gains and operator incentives
According to Duke University estimates cited in the research, even modest flexibility from data centers could accommodate roughly 100 GW of additional data center load on the US grid without constructing new power plants. That's equivalent to avoiding billions in infrastructure investment.
Data center operators themselves stand to benefit from demand response programs, which typically pay participants for reducing load during stress events. The economics align: operators earn revenue from flexibility, utilities avoid costly peaker activations, and the grid becomes more resilient.
