NVIDIA and Emerald AI Want AI Data Centers to Power the Grid, Not Just Drain It

NVIDIA and Emerald AI announced Monday at CERAWeek 2026 that they are working with AES, Constellation, Invenergy, NextEra Energy, Nscale Energy & Power, and Vistra to build what they are calling “flexible AI factories” — data centers designed to connect to the power grid faster and operate as demand-response assets that can actively support grid reliability.

The technical centerpiece is NVIDIA’s Vera Rubin DSX AI Factory reference design, which includes a new software layer called DSX Flex. That library allows AI factories to send flexibility signals to power grids — essentially letting the data center dial back compute load during peak demand periods and redirect co-located generation and storage resources back to the grid.

Jensen Huang framed it in characteristically direct terms: “AI factories are the engines of the intelligence era, and like any great engine, every system must be designed together — energy, compute, networking and cooling as one architecture.”

The practical problem this solves is significant. Gigawatt-scale AI projects are increasingly being built with co-located generation — natural gas, solar, or batteries on-site — because conventional grid interconnection queues are measured in years, not months. That keeps AI capacity coming online faster, but it also means stranding generation assets that could otherwise support the broader grid.

Emerald AI’s Conductor platform is the orchestration layer that sits between compute tenants and those energy assets. It promises to ensure AI workloads remain protected during grid-response events, while simultaneously squeezing value out of on-site generation when the grid needs help. Varun Sivaram, Emerald’s CEO, put it plainly: “AI factories produce tremendously valuable AI tokens, and with DSX Flex, they can also provide measurable relief back to the grid.”

NVIDIA says it trialed DSX Flex-style flexibility at five commercial data centers worldwide over the past year. The first commercial-scale deployment is planned for its AI Factory Research Center in Virginia, built around Vera Rubin infrastructure, later this year.

The coalition’s timing is deliberate. U.S. grid operators and regulators are under pressure from surging electricity demand, and AI data center developers are increasingly in the crosshairs of state-level permitting fights. Framing AI factories as grid assets rather than grid burdens is a reputational and regulatory play as much as a technical one.

For Nvidia, it is also a way to make its Vera Rubin architecture — the successor to the Hopper and Blackwell generations — the default design for the next wave of large-scale AI infrastructure buildout. The energy companies in the coalition collectively represent hundreds of gigawatts of generation capacity. If they standardize on Vera Rubin DSX, that is a substantial lock-in.