Energy AI Meets Flexible Compute: NVIDIA and Emerald AI Build Data Centers That Bend to Grid Demand

At CERAWeek 2026, NVIDIA and Emerald AI announced a partnership with six major energy companies — AES, Constellation, Invenergy, NextEra, Nscale, and Vistra — to build AI data centers designed to operate as flexible grid assets. Instead of consuming electricity at a fixed rate regardless of grid conditions, these facilities ramp compute workloads up during periods of surplus generation and throttle down when demand peaks or supply tightens.

The technical foundation rests on NVIDIA's Vera Rubin DSX reference design and the accompanying DSX Flex software stack. DSX Flex continuously monitors grid frequency, pricing signals, and renewable generation forecasts, then automatically adjusts the facility's power draw in response. When wind and solar output surges beyond what the grid can absorb, the data center increases its load and processes deferred AI training batches. When a heat wave drives residential air conditioning demand to record levels, the facility curtails non-critical workloads within minutes. Jensen Huang described the approach as turning AI factories into "the battery the grid has always needed."

The partnership identifies 100 gigawatts of potential flexible capacity across the participating companies' generation portfolios. Nscale alone plans to deploy 2 gigawatts of flexible AI compute initially, scaling to 8 gigawatts as the model proves out. For grid operators running energy AI and AI grid management platforms, these facilities represent a new category of demand-response resource — one that responds faster than traditional industrial curtailment programs and carries no penalty for frequent cycling.

Renewable energy AI stands to benefit directly from this architecture. Intermittent generation from wind and solar creates surplus periods that currently force curtailment or negative pricing. Flexible AI data centers absorb that surplus productively, improving the economics of renewable projects and reducing the curtailment waste that has challenged grid planners. For energy companies and compute operators alike, surplus power becomes processing capacity rather than lost revenue.

Sources

NVIDIA Newsroom