Japan-backed gas power for AI sounds decisive—until you ask what happens if the demand arrives late. This is The Data Center Daily. Today: who carries the risk on the newest AI power bet—and why water rules and grid software may decide what gets built. First, the agreement that moved the fight from city hall to Washington and Tokyo. From PR Newswire:
NextEra Energy, Inc. (NYSE: NEE) today announced it has executed definitive agreements with the U.S. Department of Commerce and the Government of Japan to fund the development and operation of up to 10 gigawatts of natural gas-powered generation in Texas and Pennsylvania. Today's announcement follows President Donald J. Trump's approval of the projects, which were selected in connection with Japan's $550 billion investment commitment to the United States as part of the U.S.-Japan trade agreement, in March.
Up to 10 gigawatts of gas in Texas and Pennsylvania, with Commerce and Japan putting up an initial $3.3 billion. Washington and Tokyo are putting real money behind data-center power—and they’d better be clear on who carries the risk if that large-load demand arrives late. These are definitive agreements with identified states, a funding milestone, and turbine deposits—well beyond the usual capacity announcement. “Up to 10 gigawatts” still isn’t 10 gigawatts energized, so watch for committed capacity and in-service dates. And the first money goes to long-lead turbines and choosing EPC contractors. Good—because a press release cannot spin a turbine. But the ratepayer-protection language needs an actual cost allocation, not just a pledge stapled to a trade deal. NextEra says this is explicitly for large-load and digital-infrastructure hubs. Supply has moved into executed paperwork; now we need to see how much of that load is equally firm. From Corey Olson at Schneider Electric Blog:
Across the United States, states are introducing legislation that governs how data centers use, report, and manage water. While early proposals focused on transparency, newer measures regulate cooling methods, requiring hydrologic impact assessments and, in some cases, encouraging or mandating cooling approaches that minimize water withdrawals. For colocation providers, hyperscalers, neocloud players, and the ecosystem of engineers, specifiers, and contractors that support them, cooling architecture has become part of the permitting and compliance strategy.
Schneider’s point is that water is now part of the permit file. States are moving beyond reporting requirements toward hydrologic impact assessments and cooling-method rules before a data center breaks ground. Good. At 227 kilowatts per rack for NVIDIA's Vera Rubin platform, cooling isn't a facilities footnote. If a project plans to pull heavily from a stressed watershed, that belongs in the approval process—not in a sustainability brochure. And yes, this is Schneider and Motivair advocating closed-loop liquid cooling. But the regulatory shift is bigger than the product pitch: architecture can now determine whether a campus clears compliance at all. Closed-loop can reduce withdrawals, fine—but make developers show their hydrologic math. Then show us the backup-water plan and the actual cooling design. “Water-efficient” has become a very roomy marketing term. Brookhaven National Laboratory, writing in TechXplore:
In late 2024, IBM Research and its collaborators laid out a vision for foundation models for the electric grid: fast, physically consistent, task-agnostic neural solvers that could give grid operators and planners a new tool to keep pace with an ever-changing system. Now, that vision is on its way to becoming a reality. Today, we're releasing our first contribution for steady-state grid analysis through the OpenGridFM project at Linux Foundation Energy (LF Energy).
GENCO targets the studies—power flow, optimal power flow, state estimation—that turn an interconnection queue into a multi-year traffic jam. Great. Now get it onto an operator’s desk before it becomes another very elegant GitHub repository. Brookhaven and IBM Research, working through LF Energy, are releasing an open-source steady-state solver; they’re not saying it’s deployed. The pitch is one shared model across three planning tasks. The milestone that matters is utility validation and adoption. TechXplore puts roughly 2.6 terawatts of generation and storage in U.S. interconnection queues. If this can cut study time without cutting safety margins, planners need it yesterday. If it produces a prettier answer nobody will sign off on, it changes nothing. We just covered 10 gigawatts of proposed gas buildout aimed at digital load. Faster grid math won’t pour concrete or win a permit. It could tell developers much earlier whether that load-and-generation story survives contact with the network. If you’re enjoying The Data Center Daily, please subscribe and leave us a review wherever you’re listening. Reviews help more people find the show, and your support helps us keep bringing you the day’s data-center news.
Links to every story are in the show notes if you want to dig into any of them. Thanks for listening, and we’ll be back tomorrow. That’s it for today’s Data Center Daily. This is a Lantern Podcast.