Crusoe Begins Phase Two of $15 Billion AI Data Center Buildout in Texas

Crusoe, the vertically integrated AI infrastructure company, has begun construction on the second phase of its AI data center campus in Abilene, Texas, part of a $15 billion joint venture with Blue Owl Capital and Primary Digital Infrastructure. When complete, the site will support 1.2 gigawatts of compute capacity and host one of the densest concentrations of NVIDIA’s latest AI chips in the world.

“The foundation of artificial intelligence starts with energy and data centers,” said Crusoe CEO and co-founder Chase Lochmiller. “Crusoe is bringing these AI factories to life at an unmatched speed and scale that enables our customers to usher in a new era of AI-driven abundance.”

Construction of the first two buildings began in June 2024 and is expected to be completed in the first half of 2025. Phase two, which includes six additional buildings, started in March and will add roughly 1 gigawatt of capacity. All eight buildings are scheduled to be operational by mid-2026. Each structure will house up to 50,000 NVIDIA GB200 NVL72 units, connected via a unified RDMA fabric for large-scale AI model training and inference.

Crusoe is responsible for designing, building, and operating the entire site. “We developed the data center itself, and we’re in charge of building it, and we will own it in a joint venture partnership,” said CEO Chase Lochmiller in a Bloomberg interview. “These large-scale data centers with chips… can manufacture intelligence.”

The Abilene project is located at the Lancium Clean Campus and is already employing more than 3,000 people daily on-site. That number is expected to rise to nearly 5,000 during the peak of the second phase. According to estimates from the Development Corporation of Abilene (DCOA), the first phase alone is projected to generate approximately $1 billion in economic value over 20 years. The full campus is expected to significantly multiply that impact.

Lochmiller emphasized that data center design is undergoing a structural shift driven by new compute requirements. “Twenty years ago, a standard data center rack might have been two kilowatts, maybe four,” he said. “Today, for the GB200 NVL72 configuration, we’re budgeting 130 kilowatts per rack, all with direct-to-chip water cooling.”

“Those are 600 kilowatts per rack,” he said. “You’re looking at more than a 100x increase in overall power density. That fundamentally shifts what you’re building.”

He added that future hardware such as Nvidia’s Vera Rubin Ultra will further increase rack-level power requirements to over 600 kilowatts more than 100 times the density of traditional racks requiring fundamental changes in infrastructure. The campus will use closed-loop, direct-to-chip cooling systems designed to eliminate evaporative water loss, while wind energy from West Texas was a major factor in site selection. “Energy is the core of this,” Lochmiller said. “We’ve never seen data center infrastructure built at this scale before.”

Beyond density, Lochmiller noted that the way AI compute is structured has changed. “These data centers are no longer a bunch of individual computers all plugged in, operating independent workflows. You really should be thinking about the data center as the computer,” he said. “All of these [GPUs] are interconnected on the same high-performance RDMA fabric so that they can really be thinking together and sharing information.”

The scale of construction has also introduced challenges in labor and materials. “Labor is honestly the biggest expense here,” he said. “There’s a shortage of construction workers, a shortage of electricians that can really make these facilities happen at scale.”

Despite these constraints, demand continues to grow. “We are seeing incredible demand from across a diverse set of customers that want to build scaled infrastructure to support their own AI computing needs,” Lochmiller said. “From our perspective, this is the largest infrastructure investment in human history, and it’s not slowing.”

Blue Owl, which is backing the project through its Real Assets platform, described it as one of its most significant investments. “The AI-driven demand for compute power highlights one of the most compelling investment opportunities our Real Assets platform has seen since our inception,” said Marc Zahr, Co-President and Global Head of Real Assets.

“This is a unique opportunity to be at the forefront of a paradigm-shifting infrastructure investment,” added Jared Sheiker, Chief of Staff at Blue Owl. “We are pleased to do so with such exemplary partners.”

Primary Digital Infrastructure, which serves as an advisor and investor in the project, sees the data center as aligned with its mandate. “By partnering with forward-thinking operators like Crusoe and financial lenders like Blue Owl, we are collectively delivering vital capacity with the scale, speed, and sustainability required to fuel AI innovation,” said Bill Stein, Executive Managing Director and CIO at Primary Digital Infrastructure.

Oracle has signed a 15-year lease at the facility, linking the site directly to OpenAI’s compute strategy. Crusoe has not disclosed specific tenant relationships, but the scale of the build and chip supply has made the Abilene campus a key node in the emerging AI infrastructure landscape.

Energy provisioning remains a core constraint in enabling large-scale compute. “We’ve never seen data center infrastructure built at this scale before,” Lochmiller said. “If you look at Northern Virginia, as of the end of last year, a report from JLL shared that it was 4.5 gigawatts of total capacity. What we’re building in Abilene, Texas is 1.2 gigawatts, one facility that’s a quarter of that entire region.”

“And now we’re looking at facilities that are five times bigger,” he added.