The Startup Betting AI's Power Problem Is Software, Not Steel

Emerald AI raised $150 million to prove that AI data centers can flex their power draw on command — and that doing so is worth more than building new power plants.

The consensus fix for AI's electricity problem has been to build more of it: more gas turbines, more transmission lines, more nuclear restarts. Emerald AI's $150 million Series A is a bet on the opposite move — teach the data centers to want less, at exactly the moments the grid needs them to.

The Washington, D.C.-based startup announced on August 25 that it had raised an oversubscribed $150 million Series A at a $1.05 billion valuation, co-led by Energize Capital and DCVC, according to Emerald AI's own announcement. The round brings its total funding past $220 million and pulls in an investor list that reads like a cross-section of the entire AI power stack: Nvidia, Samsung Ventures, Siemens, GE Vernova, RWE, Aramco Ventures, Salesforce Ventures, and In-Q-Tel among them. Twelve Fortune Global 500 companies are now investors and sit on the company's Strategic Advisory Board.

What Emerald AI's software actually does

The company's Emerald Conductor platform dynamically adjusts how much power a data center draws in response to real-time grid conditions — without simply shutting workloads down. Instead of a binary on/off switch, the software orchestrates AI computational workloads alongside onsite energy resources, throttling non-critical compute during periods of grid stress while protecting the performance of workloads that can't tolerate interruption.

The pitch, in Emerald's own framing, is that this can unlock more than 100 gigawatts of capacity on the existing U.S. grid — power that's already there, sitting unused as headroom utilities won't commit to new large industrial loads without, without any new transmission or generation being built. Given that building new grid infrastructure routinely takes a decade or more, and the International Energy Agency projects data centers will account for nearly half of U.S. electricity demand growth through 2030, software that makes existing capacity usable years earlier is a fundamentally different kind of bet than building new capacity from scratch.

From demonstration to commercial deployment

What separates this round from Emerald's earlier funding is a change in what the company can actually point to. Over the past year it completed five live demonstrations at commercial data centers in Arizona, Illinois, Virginia, Oregon, and London, working with partners including Nvidia, EPRI, Oracle, Nebius, and National Grid. With that phase complete, the company says it has moved into commercial deployment at full data-center scale in California, and it's now working with Digital Realty and Nvidia on what it calls the world's first power-flexible AI factory — a roughly 100-megawatt facility in Manassas, Virginia, tested alongside EPRI, Dominion, and the PJM Interconnection, expected online later this year.

That progression — from pilot to paid, multi-site commercial deployment — is exactly the traction investors have been demanding across AI infrastructure this year, and it shows in the cap table. Emerald AI also partnered with Silicon Valley Power on the first-in-the-nation Flexible Load Interconnection Program, which grants data centers expanded grid access in exchange for verified, dispatchable flexibility — a regulatory template that, if it spreads to other utilities, would turn Emerald's software from a nice-to-have efficiency play into a prerequisite for getting a data center connected to the grid at all.

The stakeholder breakdown

For AI labs and cloud operators building new data centers, Emerald's pitch is speed: interconnection queues in many U.S. markets now stretch years, and a facility that can credibly promise flexible, dispatchable power draw can jump that queue or secure capacity a conventional inflexible facility couldn't get approved for at all.

For utilities and grid operators, the calculus is about avoiding an ugly choice between throttling AI growth and passing enormous new infrastructure costs onto residential ratepayers. Software that lets them approve large new loads without guaranteeing those loads will strain the system at peak hours is, functionally, a way to say yes to AI-driven demand growth without inheriting all of its downside risk.

For the strategic investors on Emerald's cap table — chipmakers, industrial conglomerates, oil majors, sovereign-linked funds — the bet is different again. Nvidia has an obvious interest in anything that gets more AI accelerators plugged in faster. Siemens and GE Vernova sell the industrial equipment inside power-flexible facilities. Aramco Ventures and RWE have direct exposure to how energy markets absorb an AI-driven demand shock. Backing Emerald gives all of them a seat at the table shaping how that transition actually happens, rather than reacting to it after the fact.

What's still unproven

Five demonstrations and one full-scale commercial deployment is real progress, but it's a small sample against a claim of unlocking 100-plus gigawatts nationally. The hardest test for Emerald's model isn't a single flagship facility in Virginia — it's whether utilities across dozens of different regulatory jurisdictions, with different market structures and different appetites for risk, will actually grant the kind of flexible interconnection terms Silicon Valley Power did. Regulatory approval, not the underlying software, is the real bottleneck standing between a good demonstration and the gigawatt-scale outcome the company's pitch depends on.

Why it matters

Every AI capex story eventually runs into the same wall: the grid. Emerald AI's round is notable less for its size than for who's writing the checks — a coalition spanning the company that makes the chips, the companies that make the industrial hardware, the energy majors exposed to demand growth, and government-linked strategic capital, all betting on the same underlying idea. If AI's power constraint gets solved primarily through software that makes existing infrastructure more efficient, rather than through a decade-long buildout of new generation and transmission, it changes the timeline for how fast AI infrastructure can actually scale — and who profits from getting there first.

Frequently Asked Questions

How much did Emerald AI raise and at what valuation?

Emerald AI raised $150 million in an oversubscribed Series A financing, co-led by Energize Capital and DCVC, at a $1.05 billion valuation. The round brings the company's total funding to more than $220 million.

What problem does Emerald AI's technology solve?

Emerald AI's Emerald Conductor software dynamically adjusts how much power an AI data center draws based on real-time grid conditions, without shutting workloads down entirely. The company says this approach can unlock more than 100 gigawatts of existing U.S. grid capacity that would otherwise sit unused.

Is Emerald AI's technology already deployed commercially?

Yes. After completing five demonstrations in Arizona, Illinois, Virginia, Oregon, and London, Emerald AI says its software is now running commercially at full data-center scale in California, with a roughly 100-megawatt power-flexible facility in Virginia expected online later in 2026.

Who are Emerald AI's investors?

The Series A was co-led by Energize Capital and DCVC. Other investors include Nvidia, Samsung Ventures, Siemens, GE Vernova, RWE, Aramco Ventures, Salesforce Ventures, JERA Ventures, In-Q-Tel, and investors John Doerr and Tom Steyer, among others. Twelve Fortune Global 500 companies are now investors overall.

Editor's note — sources: Emerald AI official announcement (emeraldai.co); TechFundingNews; Fortune coverage of AI data-center power constraints.