Emerald AI Reaches $1.05B Valuation on $150M Series A Round

A unicorn built on turning AI's power use down

Emerald AI is a Washington, D.C. startup whose software lets AI data centers cut or shift their electricity use when the grid is strained. On August 25, 2026, it announced a $150 million Series A at a $1.05 billion valuation, and it said the round was oversubscribed.1 Energize Capital and DCVC co-led the financing. The company says its total funding now exceeds $220 million.27

The company is roughly two years old, founded in 2024 by CEO Varun Sivaram, a former climate policy official in the Biden administration.6 One outlet dates the founding more precisely to November 2024.14 Emerald came out publicly in July 2025 with a $24.5 million seed round.15 It took about 13 months from that launch to reach unicorn status.4

The valuation is less telling than what it says about where AI money is going. The investors are betting that electricity, not chips, is now the tightest limit on AI growth, and that software can loosen it faster than new power plants and transmission lines.

What the company actually sells

Emerald's product is called Emerald Conductor. It coordinates a data center's computing jobs with onsite energy resources so the facility can lower its power draw during grid stress without disrupting critical work.513 In practice, the system can slow, pause, cap or relocate selected jobs when a utility asks for less demand.14 The company describes two kinds of flexibility: slowing non-urgent work over time, and moving workloads between locations.18

Sivaram says the approach is software-only and works on any hardware, so it can scale across many facilities without redesigning them. He also says it does not need access to sensitive model or training data.

The company's headline claim is large. It says that if this kind of flexibility were applied across the AI buildout, it could unlock more than 100 gigawatts of unused capacity on the existing U.S. grid, years before new generation and transmission could be built.213 The company's own newsroom summary says "up to" 100 gigawatts rather than "more than," a small difference that hints at how loose the number is.15 Analysts have pointed out that the 100-gigawatt figure is Emerald's own technical estimate. No grid operator has certified it, and it depends on utilities and hyperscalers actually adopting flexible operation.4

The investor list

The group of backers stands out more than the round size. Participants included NVIDIA, Samsung Ventures, Siemens, Aramco Ventures, Salesforce Ventures, GE Vernova, RWE, JERA Ventures, In-Q-Tel, Radical Ventures, Energy Impact Partners and Lowercarbon Capital. John Doerr and Tom Steyer invested personally.213 Emerald says 12 Fortune Global 500 companies are now investors or sit on its Strategic Advisory Board.217

Coverage broadly agrees on why. The backers include chipmakers, utilities, equipment makers and cloud buyers, and each has a direct commercial reason to want the power problem solved. Commentators called that mix of climate funds, energy companies, industrial groups and tech platforms unusual for a Series A.1612 Nvidia's interest is the easiest to read. More grid capacity means more of its GPUs can be plugged in sooner, and it also backed Emerald's 2025 seed round.4 One analysis put it bluntly: Nvidia is paying to remove a limit on its own revenue.11

The investors frame it the same way. Energize Capital's John Tough called electricity the emerging constraint on AI and said software flexibility is one of the fastest ways to add capacity.8 DCVC's Zachary Bogue said the platform turns data centers into "grid-responsive assets instead of energy-hogging liabilities."1

The evidence so far

Emerald has completed five demonstrations at commercial data centers in Arizona, Illinois, Virginia, Oregon and London. Partners included NVIDIA, EPRI, Oracle, Nebius and National Grid.2 The clearest published result is from Phoenix in May 2025. A cluster of 256 Nvidia A100 GPUs cut its power draw by 25% from average for three hours, with 15-minute ramps down and back up.14

The company says it has now moved into commercial deployment. It recently ran its software across an entire California data center during peak demand.213 With Silicon Valley Power, it launched what both parties call the first U.S. Flexible Load Interconnection Program. Data centers get more grid access in exchange for verified reductions in demand that the utility can call on when needed.1219 The bigger test is the nearly 100-megawatt Vera Rubin AI Research Factory in Manassas, Virginia, built with Digital Realty and NVIDIA. Dominion and the regional grid operator PJM are involved.12

The sources disagree on how much this has been proven. Some say the company is selling a working product, not a roadmap.16 A more skeptical analysis says the Santa Clara pilot had not yet published results. It also says key questions remain open: who controls the switch, how operators are paid, and what the effect on emissions is.14 Another report found that as of late June there was still no standard binding agreement between a utility and a data center. It also found that bill credits are often too small to justify delaying profitable computing work, so faster grid access is the stronger incentive.17

On balance, the skeptics have the better case for now. One 25% cut over three hours on a 256-GPU cluster shows the idea works technically. It does not show that a 100-megawatt AI factory can reliably serve as a grid resource for years.

Inconsistencies in the numbers

The funding details are less settled than the press release suggests. A Form D filed on August 3 recorded $90,229,639 sold toward a planned $150 million offering. It did not name the valuation or the lead investors.1417 One outlet noted that the final round terms had not been separately confirmed in a filing.14

The totals also don't fully add up across sources. In March 2026, Emerald announced a $25 million strategic expansion round led by Energy Impact Partners, which it said brought total funding to about $68 million.18 Adding $150 million to that gives roughly the "more than $220 million" figure now reported.10 One investor-focused breakdown, however, says only $49.5 million came before the Series A, from the seed and the strategic round. That would leave the total short of $220 million.4 The missing piece may be an $18 million round that was reported earlier, in which Nvidia reportedly invested. Another outlet described a "$100M round" anchored by Nvidia, which matches no figure in the company's announcement.11 Anyone tracking this deal should rely on the company's $150 million and $220-million-plus figures and treat the other versions with caution.

Accounts of the business model differ too. One report says Emerald earns money by taking a share of the extra power sales it enables for utilities, plus a share of added revenue when it helps data center operators shift workloads.6 Another says the company has not publicly disclosed revenue or pricing.4 Both can be true: the general structure is known, but the actual economics are not.

Where the deal fits in AI funding

The round lands in a crowded market for AI infrastructure deals. Reuters-syndicated coverage noted that investor demand for these startups has grown over the past year as tech companies race to add data center capacity. That has pushed valuations up and sped up funding rounds for companies working on bottlenecks such as power.9 Morgan Stanley projects that global electricity demand will rise by more than 1 trillion kWh a year through 2030, with data centers driving nearly 20% of that growth.10 The company says data centers could account for nearly half of U.S. demand growth over the same period.13

By those standards, Emerald's raise is modest. Comparable 2026 power-related rounds were much larger: Base Power raised $1 billion at a $13 billion valuation, and Valar Atomics raised $1 billion at $6 billion.4 Those, however, are later-stage or hardware-heavy businesses. Emerald is pricing an early-stage software company at a billion dollars on the strength of pilots and partnerships.

That leads to the main concern. One outlet pointed to a pattern of AI-infrastructure startups reaching unicorn status at the Series A stage before showing that their businesses last. It warned that a billion-dollar valuation this early leaves little room for mistakes.6

The bottom line

In our view, the most important factor is regulation, not technology. Grid operators and FERC are rewriting the rules for connecting large loads. If those rules require large facilities to accept curtailment in exchange for faster connections, Emerald stops selling something optional and starts selling compliance.11 The Silicon Valley Power program is an early example of that trade.12

If other utilities copy that model and the Manassas facility performs at scale, the valuation will look cheap in hindsight. If flexibility remains voluntary and paid only through small bill credits, Emerald will have a hard time growing into a valuation set by investors whose main interest is getting more power to data centers. For now, the round mainly signals that grid flexibility has become an investable category of its own.