1.35GW Vera order announced
Nscale and Microsoft announced a collaboration to deploy 1.35GW of NVIDIA Vera Rubin NVL72 GPUs at a West Virginia AI campus — one of the largest single‑site GPU commitments to hit public view. That scale locks in huge centralized GPU demand and raises the bar for alternative accelerator suppliers competing on cost and power.
Nscale acquired) American Intelligence & Power Corporation, securing the Monarch Compute Campus site that spans roughly 2,250 acres in Mason County, West Virginia. The Monarch site is described as the United States’ first state‑certified AI microgrid with a power runway scalable to more than 8 GW, and Nscale said it will spin up a new division, Nscale Energy & Power, headquartered in Houston. (nscale.com) Nscale said) the campus buildout and compute delivery will be staged across multiple tranches beginning in late 2027 under an initial multi‑year compute services term and long‑term data‑centre lease structure. (nscale.com) AIPCorp previously outlined a strategic alliance with Caterpillar to pair utility‑grade generation and battery storage — a plan described by Caterpillar as a repeatable model to deploy up to 2 GW of dedicated on‑site power per platform and manage rapid load variability for hyperscale AI loads. (investors.caterpillar.com) NVIDIA’s Vera Rubin NVL72 rack‑scale system is specified as integrating 72 Rubin GPUs and 36 Vera CPUs with NVIDIA’s NVLink‑6 fabric and a claimed 3.6 exaFLOPS of NVFP4 inference performance per rack, along with multi‑terabytes of HBM4 memory at rack scale. (nvidia.com) Independent coverage and vendor briefings peg NVL72‑class rack power in the order of 120–130 kW for early Rubin deployments (with roadmap designs that target substantially higher rack densities later), a figure that drives the need for specialized liquid cooling and medium‑voltage distribution architectures. (videocardz.com) Microsoft’s leadership team announced) that Azure has brought up an NVL72 system for validation, and NVIDIA’s launch materials list major cloud and cloud‑adjacent partners — AWS, Google Cloud, Oracle, CoreWeave and others — as Vera Rubin ecosystem participants slated to ship partner systems in the second half of the year. (convergedigest.com) Nscale’s prior large‑scale commitments, including an earlier contract for roughly 200,000 GB300 GPUs with Microsoft, and consultancy forecasts that AI‑related data‑centre capacity could reach into the hundreds of gigawatts by 2030, frame this deal as part of a broader hyperscaler buildout trend. (nscale.com)
