Plans were unveiled on Oct. 15 for Project Horizon, a two-gigawatt artificial intelligence campus in West Texas.
The project was detailed by Jason Warner and Eiso Kant, Co-CEOs and Co-Founders of Poolside.
“We’ve partnered with CoreWeave under an agreement that secures both our immediate and long-term compute needs,” they wrote in a post.
In the near term, CoreWeave will provide a frontier-scale cluster of NVIDIA GB300 NVL72 systems -- more than 40,000 GPUs coming online beginning in December 2025 – “to power our research and model training,” they noted.
Over the longer term, CoreWeave will serve as anchor tenant for the first 250 MW phase of the Horizon campus under a 15-year lease, with 500 MW reserved for future expansion – “establishing the foundation for true vertical integration.”
"Project Horizon is our answer to the infrastructure and power bottlenecks facing the industry, and limiting every lab competing for the frontier: a ground-up data-center development designed to give us long-term control over power, cost, and scale through vertical integration," Warner and Kant wrote. "We’ve secured a 2 GW behind-the-meter AI campus on 568 acres of development-ready land. The campus will be developed in eight phases of 250 MW each, ensuring scalable, modular growth aligned with advances in compute demand and silicon efficiency."
Situated in West Texas, the site sits adjacent to a major natural gas hub, along with established water and processing infrastructure, they said.
The site is also connected by dual long-haul fiber routes, enabling ultra-high-bandwidth, low-latency connectivity to major network hubs.
"Designed as a next-generation, behind-the-meter AI campus, Horizon offers multi-gigawatt expansion potential, built with redundancy and flexibility at every layer of the power stack. We’ve secured generation through aero-derivative turbines equipped with SCR systems to meet strict emissions standards while enabling a modular, accelerated build; grid-interconnect redundancy to ensure continuity of supply; and battery storage to enhance reliability and load balancing," Warner and Kant said.
They noted that traditional data-center construction takes years and demands thousands of people on site. "Horizon will be built differently. While modular data centers aren’t new, the hybrid modular model we’re adopting is. We’re drawing on lessons from past hyperscale builds and incorporating them from the start, designing for parallelization rather than sequential construction. We’re building in a way that maximizes production across both the modular fabrication line and traditional construction front—minimizing on-site crews while maximizing total output."
Major components such as electrical, cooling systems and compute are manufactured off-site and delivered fully integrated, accelerating deployment without sacrificing scale.
Warner and Kant said that what makes Horizon unique "isn’t just 2 MW data-hall modules -- it’s that the entire electrical and cooling architecture is engineered for incremental load."
Each system is designed to energize and operate independently, allowing new capacity to come online the moment a modular data hall is placed and connected. This lets training and inference begin immediately, while additional capacity continues to roll out in parallel. The result is continuous, compounding deployment instead of the stop-and-wait cycle that slows most projects, the executives noted.
"West Texas offers the conditions to make this possible. At the center of the Permian Basin—one of the world’s most energy-rich regions—Horizon has direct access to abundant, low-cost natural gas and the physical capacity to scale to multi-gigawatt levels. Air cooling avoids the water-consumption challenges of desert regions, and the modular build minimizes on-site labor requirements, even as we temporarily house specialized crews during construction," Warner and Kant said.
This approach "reflects how AI is reshaping infrastructure. Where traditional data centers optimized for latency and mixed workloads, AI training optimizes for scale and speed, as time to capacity has become as critical as total capacity. Every design choice at Horizon serves that goal: factory-built components reduce risk, early supplier and energy partnerships remove bottlenecks, and the modular build keeps GPUs coming online month after month."