U.S. data center power demand is expected to climb from 31 gigawatts in 2025 to 41 GW in 2026 and 66 GW the year after , according to Goldman Sachs Research. That's not a forecast. It's a crisis dressed up in spreadsheets. Average grid interconnection timelines exceed 2-4 years, with full transmission upgrades often taking 7-10 years , Redwood Materials notes. AI infrastructure, by contrast, needs power in months—not years. The mismatch has created a scramble that is reshaping the energy sector faster than regulators, utilities, or even the hyperscalers themselves anticipated.
Britain's National Grid said on Wednesday it will invest $1.75 billion for a 35% stake in Joulent, a U.S. energy platform developing power infrastructure for data centres , Reuters reported. The deal will fund Joulent's first project—a 2.67 GW co-located power facility in West Texas that will provide dedicated electricity to a Microsoft-operated data center under a 20-year power purchase agreement . The project, known as Kilby, targets first power delivery by 2028. It's a bet that the future of AI infrastructure isn't about waiting for the grid—it's about bypassing it entirely.
Can Batteries Replace the Grid?
Not replace. Buffer. GPU racks now consume 140 kilowatts versus traditional 2-10 kilowatt CPU systems in data centers , according to Battery Tech Online. That sevenfold jump in power density has turned data centers into something closer to industrial smelters than server farms. The problem isn't just volume—it's volatility. A facility may operate at 2 MW under normal conditions, only to spike to 3 MW when a training job initiates , notes EticaAG. Diesel generators can't spin up fast enough to catch those sub-second surges. The grid strains. Silicon waits.
Enter battery energy storage systems. Energy storage is projected to support 9.8 gigawatt-hours of gas generation at data centers through 2030 , according to BloombergNEF. The combo has become so popular that companies such as Caterpillar Inc. and GE Vernova Inc. have announced products or partnerships pairing energy storage with gas generation , Fortune reported. At xAI's Colossus facility in Memphis, rows of Tesla Inc. Megapacks are being installed next to gas turbines as part of a 1.2 gigawatt off-grid power plant .
The economics are stark. By cutting the cost of battery storage roughly in half, Redwood has made 24/7 renewable power "economic and scalable," said Crusoe cofounder Cully Cavness, speaking to Fast Company. Redwood Materials—founded by Tesla's former CTO JB Straubel—pioneered the first solar- and EV-battery-powered data center in Nevada in 2025, using second-life EV batteries to power modular AI infrastructure. The 12 MW / 63 MWh microgrid was built and commissioned in under four months .
What Happens When Meta Becomes a Utility?
Meta is building a cloud computing business — internally called Meta Compute — to sell access to its surplus AI infrastructure to outside companies , Bloomberg reported Tuesday. The announcement sent Meta shares up roughly 9%, according to CNBC. It also cratered the stocks of CoreWeave and Nebius—two AI cloud providers that had each signed multi-billion-dollar contracts to supply Meta with compute capacity. Meta is projected to spend as much as $145 billion on AI infrastructure this year , a figure that now looks less like a cost center and more like a platform play.
One potential plan includes selling access to various AI models that are hosted on Meta's existing AI infrastructure, an approach similar to AWS's Bedrock offering . Meta would run the data centers and chips that power the models, including its own Muse Spark models, and charge developers to access them. The move positions Meta alongside Amazon Web Services, Microsoft Azure, and Google Cloud in the enterprise AI infrastructure market—a space that had, until now, seemed safely beyond the social media company's ambitions.
The shift mirrors Amazon's own trajectory. AWS launched in 2006 as a way to monetize Amazon's internal e-commerce infrastructure. It became the company's most profitable business unit. Meta's cloud entry follows the same playbook: build massive internal infrastructure to support AI-driven products, then monetize excess capacity by renting to external customers. At Meta's shareholder meeting in May, Zuckerberg had said entering cloud computing was "definitely on the table," noting that firms were approaching Meta "almost every week" to buy access to its AI models or spare computing power .



