CoreWeave Locks Down Power Inputs

Project Horizon shows that CoreWeave is moving from renting GPU capacity to locking down the raw inputs of AI itself, land, fuel, and long term power. Instead of waiting for a utility scale data center to come online, the company is tying itself to a West Texas site with behind the meter natural gas generation, where the first 250MW phase is anchored by CoreWeave and another 500MW is reserved, turning power access into a supply advantage that smaller GPU clouds cannot easily copy.1 2 3 4

• The structure is unusually vertical. Poolside controls the campus development, CoreWeave supplies cloud infrastructure and serves as anchor tenant, and the initial deployment includes more than 40,000 NVIDIA GB300 NVL72 GPUs under a 15 year lease starting with the first phase. That makes Horizon less like a normal colocation deal and more like a purpose built AI factory stack.2 4
• West Texas is the point. The site sits in the Permian Basin with direct access to abundant natural gas, which means compute can be added by bringing generation on site instead of waiting years for scarce grid interconnects. This is the same basic energy logic that helped Crusoe grow, but at a much larger planned scale.3 5 6
• The timing matters because CoreWeave lost its attempted shortcut to owned capacity when Core Scientific terminated their merger on October 30, 2025 after stockholder approval failed. Horizon shows the fallback plan, secure power through organic builds and partnerships rather than acquiring an existing operator outright.7 8 1

The next phase of AI infrastructure competition will be decided less by who can buy GPUs, and more by who can assemble gigawatts fast enough to keep those GPUs fed. Horizon points toward a market where the winners look more like energy developers and industrial builders, with CoreWeave pushing furthest in that direction.1 3 5