Emerald AI Raises $150M Series A at a $1.05B Valuation With Nvidia, Siemens and GE Vernova; Its 100-Gigawatt Grid Claim Carries No Cited Source

by Vikram Desai · August 26, 2026

Funding & Deals Reporter, Frontier Tech Wire

Emerald AI said Tuesday it had closed an oversubscribed $150 million Series A at a $1.05 billion valuation, co-led by Energize Capital and DCVC, bringing total capital raised to more than $220 million. The company sells software that throttles and reschedules AI workloads inside data centers so a facility can cut its draw when the grid is stressed, turning a fixed load into a flexible one.

The investor list is the most striking part of the announcement and is unusual for a Series A. According to the company's Aug. 25 release, participants include Nvidia, Samsung Ventures, Siemens, GE Vernova, RWE, Aramco Ventures, Salesforce Ventures, JERA Ventures, ADVentures, Sabanci Climate Ventures, In-Q-Tel, Radical Ventures, Energy Impact Partners, Lowercarbon Capital, Marunouchi Innovation Partners, Emerson Collective, The Olayan Group and the Temerty Group, Earthshot Ventures and Collective Global, along with individual investors John Doerr and Tom Steyer and General Catalyst's scout fund. The company says twelve Fortune Global 500 companies are now on its cap table.

That composition tells you what the round is. A chipmaker, a turbine manufacturer, a European utility, a Japanese power company, a Gulf energy investor and an industrial conglomerate are not writing venture checks at this stage for financial return alone; they are buying a position in whether power flexibility becomes a standard feature of data-center interconnection. The strategic-heavy roster is also why the valuation should be read carefully — a $1.05 billion mark set by a syndicate with commercial reasons to want the product to exist is not the same signal as one set by financial investors competing on price.

The company's product is called Emerald Conductor. The pitch, in the words of Energize Capital managing partner John Tough in the release, is that "the binding constraint on AI is no longer chips or capital; it is power, and software is the fastest way through it." DCVC co-founder and managing partner Zachary Bogue said the platform makes flexibility a permanent feature of how data centers are powered. Founder and chief executive Varun Sivaram said the company was founded on the conviction that the intelligence driving the AI revolution could solve its own greatest bottleneck, power.

Now the part this desk checks. The release carries two headline statistics, and it treats them very differently.

The first is attributed. The release states that data centers are projected to account for nearly half of the growth in U.S. electricity demand through 2030, and says explicitly that this comes from the International Energy Agency. That is how it should be done: a claim, a named institution behind it, and a reader who can go look.

The second is not. The release says the approach can unlock more than 100 gigawatts of untapped capacity on the existing U.S. power grid. No study, institution, methodology or internal analysis is named as the basis for it. It is the number in the headline of much of the coverage, it is the number that sizes the entire opportunity, and as published it is an assertion. There is academic work on how much headroom curtailable load could free up on the U.S. system, and the figure is not implausible; the point is that the release does not tell you which work, if any, it is drawing on, or what curtailment tolerance and time horizon the estimate assumes. Those assumptions are what a 100-gigawatt number lives or dies by.

The deployment record has the same shape. The release says Emerald AI has completed five successful commercial data-center demonstrations, in Arizona, Illinois, Virginia, Oregon and London. It gives no megawatt size for any of them, no percentage of load actually shed, and no duration over which the reduction was sustained. Those three numbers are the entire technical content of a demand-flexibility demonstration. A grid operator evaluating this technology would want all three; the release publishes none.

What it does publish is names, and the names are checkable in their own right. Partners cited include Nvidia, Oracle, Nebius, National Grid and EPRI, the utility industry's research institute. The release says Emerald AI partnered with Silicon Valley Power to launch what it calls a first-in-the-nation Flexible Load Interconnection Program, and that in Manassas, Virginia it is working with Digital Realty and Nvidia to bring online the nearly 100-megawatt Vera Rubin AI Research Factory, slated to come online later this year. That "nearly 100-megawatt" figure is the only capacity number attached to any Emerald AI site anywhere in the announcement, and it refers to a project that has not launched.

The release also discloses no revenue, no customer count and no contracted megawatts. It describes customers spanning AI firms, data-center operators and electric utilities without saying how many. For a company that has just been valued above a billion dollars, the absence of a contracted-capacity figure is the single most informative omission — contracted flexible megawatts is the unit this business would ultimately be measured in.

The round was one of several disclosed in Tuesday's venture activity. It was not the day's largest — a $200 million Series D for the autonomous-freight company Gatik was bigger — but it was the largest with a direct line to AI data-center infrastructure. A separate deal in the same day's flow, a $30 million Series B for Celera Semiconductor led by Maverick Silicon, sits on the chip-design side of the same problem: its Nesto design-automation technology combines digital twins and AI-driven methods to cut the time and cost of designing analog chips.

Emerald AI is private, so there is no share price here and no filing to read against the release. That makes the discipline more important rather than less. What the company has actually disclosed is a funding amount, a valuation, a partner list, a project that has not opened yet and one sourced macro statistic. What it has asserted without support is the number that defines the size of its market. Those are two different categories of fact, and a release that sources one of them while leaving the other bare invites the question of why.