The projection lands somewhere between alarming and inevitable: data centers worldwide may consume 945 terawatt-hours of electricity by 2030, roughly double today's draw, if the International Energy Agency's latest forecast holds. Goldman Sachs Research sees an even steeper climb—perhaps 165% growth from 2023 levels, translating to some 122 gigawatts of new capacity humming away before the decade closes. The culprit, or catalyst depending on your vantage point, is artificial intelligence. And the infrastructure keeping these machines cool and fed? It wasn't designed for this.
Investors, sensing both crisis and windfall, have responded with checkbooks open. Over the past year, a cohort of startups promising smarter energy management for data centers has collectively pulled in hundreds of millions of dollars. The bet is straightforward: whoever solves the power bottleneck gets to ride the AI wave without the grid collapsing beneath it.
Follow the Money
Emerald AI has emerged with the largest seed round in this pack—$67.5 million assembled across three separate raises between July 2025 and February of this year. The initial $24.5 million tranche closed July 1, 2025, led by Radical Ventures, with checks from NVentures (Nvidia's venture arm), AMPLO, CRV, and Neotribe. By October, the company had tacked on an $18 million extension. Then, on February 19, another $25 million arrived, according to Axios Pro. Emerald's pitch centers on turning AI data centers into what it calls "grid allies"—essentially, making compute loads flexible enough to dance with utility demands in real time rather than against them.
Phaidra takes a different tack, deploying AI to optimize the cooling and energy systems inside data centers. The company closed a $50 million Series B on October 1, 2025, with Collaborative Fund leading and Index Ventures, Nvidia, and Sony Innovation Fund joining. That followed a $12 million seed raise back in February 2024, when the AI boom was already well underway but the power conversation hadn't reached fever pitch.
Then there's Exowatt, which has gone the hardware route: thermal energy storage using what it describes as modular "hot rock" batteries—heat captured and released on demand. The company raised a $70 million Series A on April 22, 2025, led by Felicis, then added a $50 million extension in November, per TechCrunch reporting. Exowatt claims a 90-gigawatt-hour pipeline of projects already tied to data center deployments, though the timeline for those remains somewhat vague.
At the heavier end of the infrastructure spectrum sits Crusoe Energy, a company that made its name monetizing stranded energy sources, including flared natural gas from oil fields. Crusoe announced the first close of a $1.375 billion Series E on October 24, 2025, co-led by Valor Equity Partners and Mubadala, with Nvidia, Fidelity, and Founders Fund chipping in. That came on the heels of a $600 million Series D at a $2.8 billion valuation in December 2024—a funding velocity that suggests investors see durable margins in solving energy access for compute.
Governments Take Notice
Policy is starting to catch up, if unevenly. On March 13 of this year, the UK floated a proposal to prioritize AI data centers for national grid connections and hinted at discounted electricity rates—an acknowledgment, as grid regulator Ofgem put it, of "severe connection bottlenecks." Translation: the queue for power hookups is long, and AI is jumping the line.
Across the Atlantic, the U.S. Department of Energy identified 16 federal sites in April 2025—including Los Alamos, better known for its nuclear legacy—for accelerated data center construction. The rationale cited existing energy infrastructure already in place, a rare advantage in a sector where lead times for new substations can stretch years.
The Uptime Institute, which tracks data center operations globally, noted in a January field report that "giant data center power plans reach extreme levels"—dry language for what amounts to a land grab for electricity. An Axios analysis from February 6 suggested that data centers could account for roughly half of all U.S. power-demand growth through 2030, extrapolating from IEA modeling. Half. The implications for grid planners, utility executives, and ratepayers are just beginning to come into focus.
The Question Left Hanging
Academic research has started proliferating around what might be called coordinated control strategies—co-optimizing GPU utilization with thermal management, tying voltage regulation to compute load, orchestrating workloads based on carbon intensity of available power at any given hour. Solutions are emerging at every layer: software, thermal systems, storage, generation.
Whether the capital and innovation arrive fast enough to keep pace with the training runs—those multi-week computational marathons that birth the next generation of AI models—remains the industry's central question. Perhaps more than the founders expected when they first started pitching "smart cooling" or "grid flexibility," the answer may determine not just which startups succeed, but how quickly the AI buildout can continue without buckling the infrastructure beneath it.
