On hot afternoons in California, when grid operators plead for every megawatt, a data center running Emerald AI’s software does the opposite of what its tenants expect: it throttles back. Compute jobs wait, training runs pause, and the building draws less power at the exact moment the grid needs relief. When demand falls and electricity turns cheap, the same software turns the machines back on, deploying spare capacity that would otherwise sit idle.
That flexibility is the business model. Emerald AI, a start-up founded by Sivaraman, said it has completed a $150 million funding round at a valuation of $1.05 billion. The round was led by DCVC and Energize Capital, with participation from NVIDIA, Samsung Venture Investment, GE Vernova, and Salesforce Ventures. The haul vaults Emerald AI into the ranks of well-funded companies working at the intersection of the AI build-out and the electricity grid.
The company’s pitch is straightforward. AI’s explosive growth has collided with an electricity system that was never designed for it. Data centers are among the fastest-growing sources of power demand in the United States, and utilities in several states have warned that new connections could be delayed by years. Emerald AI’s software gives data-center operators a way to be useful to the grid instead of a burden on it: at peak times the building sheds load, and in off-peak hours it consumes more, soaking up cheap or surplus power.
The company has already signed contracts in California and Virginia, two of the most constrained markets for data-center electricity, and it is testing its technology with NVIDIA and Oracle, according to the announcement. Sivaraman, the founder, said the goal is to turn data centers from a grid burden into a flexible resource, easing the local opposition that has stalled construction projects from Northern Virginia to Phoenix.
The round is the latest sign of how the AI boom is reshaping energy markets. For years, demand response, the practice of paying large electricity users to cut consumption during shortages, was a niche tool used by industrial plants and big-box retailers with fleets of air conditioners. Now the fastest-growing participants are server farms, whose owners are discovering that a few minutes of flexibility can be worth millions of dollars when power prices spike.
Investors say the economics are improving as grids tighten. DCVC, an early investor in the company, has backed hardware and infrastructure start-ups for more than a decade and sees AI’s power problem as one of the defining investment themes of the decade. Energize Capital, which focuses on the energy transition, said flexibility markets are expanding as renewable generation makes supply more variable and as utilities struggle to build transmission fast enough.
The competition is growing too. Tesla sells batteries that let data centers ride through peaks, and a crop of start-ups is offering software that shifts workloads across regions or time zones. What distinguishes Emerald AI, according to analysts, is its focus on working with the grid itself, participating in wholesale markets and helping utilities avoid the most expensive hours of the day, rather than simply shifting demand between facilities.
The company’s backers also include heavyweights from the AI supply chain, a sign of how intertwined the sectors have become. NVIDIA, the dominant maker of AI chips, has a direct interest in making sure that data centers can actually get the power they need to run its processors. GE Vernova, which builds the turbines and transformers that underpin the grid, sees software like Emerald AI’s as a complement to its hardware. Salesforce Ventures’ participation reflects a broader corporate push to make AI more sustainable.
For the communities where data centers cluster, the pitch is political as well as technical. Local officials in Virginia, California, and elsewhere have heard complaints about noise, water, and the strain new facilities place on residential power bills. A data center that helps keep the lights on during a heat wave is a different political object from one that only consumes. Whether that argument wins over skeptical neighbors remains an open question, but the company’s investors are betting that flexibility, not resistance, is the path forward.
The company’s software also addresses a quieter problem: power procurement. Negotiating with utilities and grid operators is slow, and the terms determine a data center’s economics for decades. Emerald AI’s pitch to operators is that its software makes those negotiations easier, because a building that can flex its load is a better neighbor to the grid than one that cannot.
The financing also signals where the market believes the value lies in the AI energy chain. Power purchase agreements, grid interconnection rights, and software that manages both have become hot commodities, and investors are placing bets across the stack. Emerald AI’s particular wager is that the software layer will matter most: the intelligence that decides when a building should burn and when it should idle, measured in seconds and priced in megawatts.
