Emerald AI Raises $150M to Flex Data-Center Power
The two-year-old Washington, D.C. startup closed an oversubscribed $150 million Series A at a $1.05 billion valuation, co-led by Energize Capital and DCVC. Its software shaves a data center’s draw when the grid is stressed — a bet that the fastest new power is power nobody has to build.
Emerald AI has raised $150 million in an oversubscribed Series A that values the two-year-old company at $1.05 billion. Energize Capital and DCVC co-led the round, announced August 25, which takes the Washington, D.C. company past $220 million raised in total.
The product is a software layer, not a generator. Emerald Conductor sits between a data center’s AI workloads and its energy resources and adjusts how much power the building draws in response to grid conditions — trimming or shifting demand when the system is stressed, and soaking up headroom when it is available. The company describes the scheduling as agent-driven, with software deciding how operators and their tenants shuffle load rather than a human dispatcher making the call. The point is that the trimming happens without stopping the compute that matters.
That distinction is the whole thesis. Utilities have historically had to treat a large data center as an inflexible block of demand — a load that pulls roughly the same megawatts whatever else is happening on the network — which means new capacity has to be sized for the worst case and built before the campus can connect. Emerald AI estimates that making those loads flexible instead would free up more than 100 gigawatts of untapped capacity on the existing U.S. grid, years ahead of anything new coming online. Founder and chief executive Dr. Varun Sivaram frames the constraint plainly: software, he argues, is the fastest way through.
The company has moved past pilots. It ran five demonstrations at commercial sites in Arizona, Illinois, Virginia, Oregon and London, and now has an operational deployment at a full-scale California data center. The most visible project is the Vera Rubin AI Research Factory in Manassas, Virginia — a roughly 100-megawatt build with Digital Realty and NVIDIA that the partners are billing as the first power-flexible AI factory of its kind. Emerald is also part of Silicon Valley Power’s Flexible Load Interconnection Program, the sort of tariff arrangement that turns demand response from a favor into a contract.
The cap table tells you who thinks this is infrastructure rather than a feature. Twelve Fortune Global 500 companies are now investors, spanning chips, industrials and energy: NVIDIA, Samsung Ventures, Siemens, GE Vernova, RWE, JERA Ventures, Aramco Ventures, Salesforce Ventures and ADI Ventures, alongside Radical Ventures, Energy Impact Partners, Lowercarbon Capital, In-Q-Tel, Sabanci Climate Ventures, Marunouchi Innovation Partners, Emerson Collective, The Olayan Group and Temerty Group. John Doerr and Tom Steyer invested as individuals.
Grid interconnection has quietly become the binding constraint on AI expansion, and the local politics of new data centers have hardened along with it. A company selling utilities a reason to say yes — and selling operators a way to connect sooner without waiting on new generation — is positioned at exactly that friction point. Whether flexibility scales from one Virginia factory to the tens of gigawatts the pitch deck implies is the question the next two years will answer.
Want AI news before everyone else?
The morning's most important AI stories, straight to your inbox. No fluff.