AI Infrastructure Crunch
Leading AI labs are locking up long‑term compute while real‑world constraints — power, water, zoning and a shortage of ‘AI‑ready’ data centers — are becoming the binding limits on growth. Anthropic has signed multiyear capacity deals, including renting CoreWeave capacity and reported arrangements tied to Google, even as JLL says less than 10% of U.S. data‑centre capacity is AI‑ready and communities are pushing back on large projects over water and power use (bloomberg.com, x.com, datacenterknowledge.com, whsv.com, wpbf.com).
Anthropic just signed a multiyear deal to rent computing capacity from CoreWeave so it can keep building and serving Claude, which tells you the bottleneck in artificial intelligence is no longer just chips but access to whole buildings full of them. CoreWeave is not a chipmaker; it is a cloud company that buys huge clusters of Nvidia processors and leases them out, so Anthropic is effectively reserving factory floor space for artificial intelligence before someone else does. That scramble is happening because an ordinary data center is not automatically fit for artificial intelligence work, which packs far more processors into one room and demands denser power feeds, heavier cooling, and faster networking than older server halls were built for. Jones Lang LaSalle, the real estate firm known as JLL, said this week that less than 10 percent of United States data-center capacity is “AI-ready,” so most existing buildings cannot simply be repurposed for production-scale model training. JLL’s 2026 outlook says global data-center capacity could rise from 103 gigawatts to 200 gigawatts by 2030, with artificial intelligence workloads reaching about half of all capacity, which means the industry is trying to add nearly another entire world-sized fleet in five years. The problem is that a new artificial intelligence data center is constrained by utility hookups, not just money, because the biggest projects can demand hundreds of megawatts of electricity and years of grid upgrades before the first server is switched on. Cooling turns electricity into a water and land fight, which is why local opposition is now shaping artificial intelligence buildouts town by town instead of this being only a Silicon Valley financing story. In Strasburg, Virginia, residents are questioning Project Tallmadge, a proposed data center on an 87-acre site in the industrial park near Interstate 81, with debate focused on water use, power demand, and environmental effects. In Palm Beach County, Florida, officials postponed a vote on the hyperscale project called Project Tango until July 15 after public criticism over electrical demand, water use, and noise. So the race has shifted from “who has the best model” to “who has guaranteed access to power, cooling, permits, and an artificial-intelligence-ready building,” and that is why labs are signing multiyear infrastructure deals before those physical slots disappear.
