Emerald AI's $150 million Series A, which values the company at $1.05 billion less than a year after it exited stealth, arrives at the precise moment Wall Street is discovering that the data-center buildout it is financing carries risks it has never underwritten before1,2,3. ANALYSIS The juxtaposition frames a single question: whether software that makes AI facilities flexible enough to coexist with the grid can de-risk an asset class that lenders are already struggling to price.
Why it matters
The U.S. AI data-center buildout is posing complex challenges to major lenders as they stretch themselves to finance, insure, and underwrite what the Financial Times calls "a novel asset class". Wall Street players extending themselves to underwrite that asset class are looking to limit exposure. ◆ Any technology that credibly reduces the power volatility of a data center, the variable that most complicates grid interconnection and insurance pricing, speaks directly to the risk that lenders are trying to contain.
The big picture
Emerald AI was founded in 2024 by Varun Sivaram, who previously held a senior climate position in the Biden administration. Its core product, Emerald Conductor, is software that schedules AI workloads against batteries and automatically adjusts the power consumed by AI clusters in response to grid requirements while meeting promised performance levels. The company's pitch: AI data centers can act as flexible assets for the power grid rather than merely serving as constant power consumers.
The $150 million round was led by DCVC and Energize Capital and was oversubscribed. The investor roster is unusually broad: Nvidia, Samsung Ventures, Siemens, Aramco Ventures, RWE, JERA Ventures, Energy Impact Partners, Lowercarbon Capital, Emerson Collective, General Catalyst's scout fund, John Doerr, and Tom Steyer all participated. Twelve Fortune Global 500 companies now sit on Emerald AI's Strategic Advisory Board, and total funding exceeds $220 million.
"We founded Emerald AI on the conviction that the intelligence driving the AI revolution could solve its own greatest bottleneck: power," the company said.
Between the lines
The technical evidence is early but specific. In a field test published in Nature Energy, Emerald AI's software reduced the power consumption of a commercial cluster equipped with 256 GPUs by 25% over a three-hour period in Phoenix without compromising performance guarantees. Over the past year the company completed five demonstrations at commercial data centers in Arizona, Illinois, Virginia, Oregon, and London, and deployed across an entire data center in California, sustaining grid-responsive power flexibility during peak grid strain.
Two partnerships point to where the company is headed commercially. Emerald AI is working with Digital Realty and Nvidia to bring online the nearly 100-megawatt Vera Rubin AI Research Factory in Manassas, Virginia. Its partnership with Silicon Valley Power launched what the company calls the first-in-the-nation Flexible Load Interconnection Program. The New York Times reported that Emerald AI uses software to keep power demand at computing facilities "from getting out of control"4.
ANALYSIS The investor composition tells its own story. Energy majors (RWE, Aramco Ventures, JERA), industrial conglomerates (Siemens, Samsung), and the chip supplier whose hardware fills these facilities (Nvidia) all participated alongside climate-focused venture funds. That breadth suggests the power-flexibility thesis resonates across the full stack of parties exposed to data-center energy risk, from fuel suppliers to GPU vendors.
The financing-risk strand reported by the Financial Times adds context. If lenders are already looking to limit exposure to data centers as an asset class, any mechanism that makes a facility's power draw more predictable, or that accelerates grid interconnection through flexible-load programs, could become a prerequisite rather than a nice-to-have for project finance.
What's next
Emerald AI said it will put the new capital into scaling commercial deployments worldwide. The Vera Rubin AI Research Factory in Manassas, a nearly 100-megawatt facility with Digital Realty and Nvidia, will be an early test of whether Conductor can operate at that scale. ANALYSIS Whether the company's grid-flexibility software can move from demonstration-scale results to a standard layer in data-center financing packages will determine how much of the lender anxiety the Financial Times described it can actually absorb. The company was recently named a 2026 Technology Pioneer by the World Economic Forum and one of the 2026 TIME100 Most Influential Companies.