NextFin News - Nscale has signed a letter of intent with Microsoft to deploy a staggering 1.35 gigawatts of AI compute capacity in Mason County, West Virginia, marking the first large-scale commercial rollout of NVIDIA’s next-generation Vera Rubin architecture. Announced on March 16 at the GTC 2026 conference, the deal encompasses approximately 430,000 Vera Rubin NVL72 GPUs. The project, dubbed the "Monarch" campus, represents a decisive shift in how hyperscalers secure the massive power and specialized hardware required for the next era of generative AI.
The scale of the West Virginia facility is difficult to overstate. At 1.35 GW, the site will consume more electricity than a large nuclear reactor can produce, necessitating a bespoke energy strategy. To solve the power bottleneck that has stalled data center projects across Northern Virginia and Ohio, Nscale has partnered with Caterpillar to utilize natural gas-fired power generation. The campus is built on what is described as the first state-certified AI microgrid in the United States, a self-contained energy ecosystem that could eventually scale to an 8 GW footprint. This "off-grid" approach allows Microsoft to bypass the years-long queues for utility grid interconnections that have become the primary constraint on AI expansion.
For NVIDIA, the partnership serves as the flagship demonstration of its Vera Rubin DSX AI Factory reference architecture. While the previous Blackwell generation focused on maximizing throughput for existing large language models, the Vera Rubin chips—scheduled for rollout starting in 2027—are designed for "industrial-scale intelligence." The NVL72 configuration used in this deal suggests a liquid-cooled rack design that treats the entire rack as a single, massive GPU. By securing nearly half a million of these units, Microsoft is effectively pre-empting the supply chain for the late 2020s, ensuring that its Azure cloud remains the primary destination for the most compute-intensive frontier models.
The choice of West Virginia is a calculated move into the "Rust Belt" of the AI era. As traditional data center hubs like Loudoun County reach their thermal and electrical limits, the industry is migrating toward regions with abundant land and favorable regulatory environments for independent power production. West Virginia’s willingness to certify AI-specific microgrids provides a blueprint for other states looking to capture the billions in capital expenditure currently flowing out of Silicon Valley. For Nscale, a company valued at $14.6 billion just a week ago, the Microsoft deal validates its "energy-first" business model, where the ability to secure power is as valuable as the silicon itself.
The long-term framework of the agreement, which combines a multi-year compute services contract with a data center lease, signals a change in how Microsoft manages its infrastructure. Rather than owning every link in the chain, the Redmond-based giant is increasingly relying on strategic partners like Nscale to handle the "heavy lifting" of power procurement and facility management. This allows Microsoft to maintain a leaner balance sheet while still controlling the cutting-edge hardware necessary to compete with Google and Amazon. As the first Vera Rubin clusters come online in 2027, the Monarch campus will likely become the epicenter of global AI training, proving that in the race for artificial intelligence, the ultimate winner may be the one who controls the most electricity.
Explore more exclusive insights at nextfin.ai.
