On a brutally hot Phoenix afternoon in May 2025, a rack of 256 NVIDIA GPUs inside a commercial Oracle data center did something data centers are not supposed to do: it turned itself down. As air conditioners across the Salt River Project territory drove demand toward a system peak, software from a then-obscure Washington startup cut the cluster's power draw by 25 percent and held it there for three straight hours, without breaking the AI workloads running on it.
That demonstration is now worth a billion dollars. Emerald AI announced on August 25, 2026 that it has raised $150 million in an oversubscribed Series A at a $1.05 billion valuation, vaulting the roughly two-year-old company into unicorn territory and bringing its total funding to more than $220 million. The round was co-led by Energize Capital and DCVC, with a strategic investor list that reads like a census of everyone with money riding on the AI buildout: NVIDIA, Samsung Ventures, Siemens, GE Vernova, RWE, Aramco Ventures, Salesforce Ventures, JERA Ventures, In-Q-Tel, Radical Ventures, Energy Impact Partners, Lowercarbon Capital, Emerson Collective, and individual checks from John Doerr and Tom Steyer. Twelve Fortune Global 500 companies now hold stakes and sit on the company's Strategic Advisory Board.
The pitch is deceptively simple. Emerald's Conductor platform sits between an AI facility's workload scheduler and its onsite energy resources, throttling, shifting, and rerouting compute so that a data center's power draw becomes a dial the grid can turn rather than a fixed block of demand. Deferrable training runs get paused or slowed; latency-sensitive inference is protected. The company says its software can now flex a full data center at multi-megawatt scale, and that applied broadly the approach could unlock more than 100 gigawatts of untapped capacity on the existing U.S. grid.
The queue is the bottleneck, not the megawatts
To understand why investors paid unicorn prices for what is, functionally, an orchestration layer, you have to understand the interconnection queue. Utilities size their systems for peak demand, and a new large load has to prove the grid can serve it at that peak, in the worst hour of the worst day, before it gets a connection. That study-and-upgrade process has collapsed under AI-driven demand: the U.S. interconnection backlog has swelled past 2,600 GW, median time from request to commercial operation is approaching five years, and in Northern Virginia — the world's densest data center market — waits stretch to seven. In ERCOT, large loads filed 198 GW of interconnection requests in the first quarter of 2026 alone, with data centers accounting for the overwhelming majority.
The perverse part is that the grid is mostly idle. A widely cited Duke University Nicholas Institute study found that 76 GW of new load — about 10 percent of current U.S. aggregate peak demand — could be absorbed today if those loads accepted curtailment just 0.25 percent of their annual uptime. At a 0.5 percent curtailment rate, the number rises to 98 GW, with PJM alone able to take 18 GW. Sivaram has put it more bluntly.
> The power grid is like a large-scale freeway that only faces rush hour two times a month. That's how underutilized our power grid is. > — Dr. Varun Sivaram, Founder and CEO, Emerald AI
Emerald's business is converting that statistical headroom into a contract a utility will actually sign. The company partnered with Silicon Valley Power in Santa Clara on what it calls the first-in-the-nation Flexible Load Interconnection Program, which grants data centers expanded grid access in exchange for verified, dispatchable flexibility — capacity now, in return for a commitment to step back when the system is strained. In Manassas, Virginia, Emerald is working with Digital Realty and NVIDIA on the nearly 100-megawatt Vera Rubin AI Research Factory, billed as the world's first power-flexible AI factory, tested alongside EPRI, Dominion, and PJM and slated to come online later this year.
Why this matters for the buildout
The financing lands at the exact moment the AI industry's constraint has migrated from silicon to substations. The IEA projects data centers will account for nearly half the growth in U.S. electricity demand through 2030, while new transmission and generation take a decade to build. Nobody is going to close that gap with steel.
> The binding constraint on AI is no longer chips or capital; it is power, and software is the fastest way through it. > — John Tough, Managing Partner, Energize Capital
There is also a political dimension the investor syndicate clearly priced in. Ratepayer backlash against data centers is now a live issue in Virginia, Georgia, Ohio, and Texas, and utilities are increasingly unwilling to socialize the cost of infrastructure built for a single hyperscaler's peak. Flexibility is the compromise that lets both sides claim a win: developers get faster interconnection, utilities avoid capital they'd have to recover from households. DCVC's Zachary Bogue framed it as turning data centers into grid-responsive assets rather than, in his phrase, energy-hogging liabilities.
Emerald has completed five commercial demonstrations — Arizona, Illinois, Virginia, Oregon, and London — with partners including NVIDIA, EPRI, Oracle, Nebius, and National Grid. It was named to the 2026 TIME 100 Most Influential Companies and a World Economic Forum Technology Pioneer.
The open question is durability. Demonstrations prove a facility can flex; commercial scaling requires AI companies to accept contractual curtailment on assets costing billions, and requires regulators to write flexible-load tariffs that hold up when the grid is genuinely short. Watch three things over the next two quarters: whether the Vera Rubin factory in Manassas actually energizes on schedule and under a PJM-blessed flexible interconnection, whether other utilities copy the Silicon Valley Power tariff structure, and whether any frontier lab publicly commits training capacity to a curtailment agreement. If those land, Emerald's $1.05 billion will look conservative. If they don't, the grid will keep waiting.
“The power grid is like a large-scale freeway that only faces rush hour two times a month. That's how underutilized our power grid is.”— Dr. Varun Sivaram, Founder and CEO, Emerald AI