Crusoe's Abilene site $59M/MW
- Crusoe’s Abilene, Texas AI campus has become the clearest real-world price tag for hyperscale AI infrastructure — roughly $15 billion for 1.2 gigawatts. - That works out to about $12.5 million per megawatt for the full campus, with the first phase already live on Oracle Cloud. - The bigger story is speed and power access — not just chips — because ERCOT-approved interconnection is now a strategic asset.
Data centers are turning into power projects with servers attached. That’s the real lesson from Crusoe’s Abilene site. The campus in west Texas is no longer just a flashy “Stargate” headline — it’s a concrete example of what AI buildouts actually cost, how fast they can move, and what the bottleneck really is. The surprise is that the headline math floating around online looks off. The public numbers point to something closer to $12.5 million per megawatt for the full site, not $59 million. (datacenterdynamics.com) ### What is Abilene, exactly? Abilene is Crusoe’s flagship AI data center campus at Lancium’s Clean Campus in Texas. Crusoe, Blue Owl Capital, and Primary Digital Infrastructure are funding and building it. The site was designed for Oracle Cloud Infrastructure workloads tied to OpenAI’s Stargate effort, and Crusoe says the full campus reaches 1.2 gigawatts across eight buildings. (datacenterdynamics.com) ### Where does the dollar figure come from? The cleanest public figure is the financing round Crusoe announced in May 2025 — an additional $11.6 billion that brought total committed capital for the Abilene joint venture to about $15 billion. Put that against(datacenterdynamics.com)and at $59 million per megawatt, so that higher number is probably mixing different scopes — maybe IT load versus total campus power, or a narrower construction slice. That’s an inference, but it fits the disclosed project numbers much better. (datacenterdynamics.com) ### Why does megawatt math matter so much? Because AI data centers are now constrained by electricity first, buildings second, chips third. A gigawatt-scale campus is not a normal warehouse full of racks. It needs substations, transmission access, cooling, b(datacenterdynamics.com)es and infrastructure investors do — not the way software people usually do. (lancium.com) ### Why is Abilene getting so much attention? Because it actually moved. Construction on the first phase started in June 2024. Crusoe later said the first phase was live on September 30, 2025, with the initial two buildings spanning 980,000 square feet and supporting more than 200 megawatts of IT capacity. In this market, a site that gets power lined up and buildings energized(lancium.com)can execute, not just pitch. (crusoe.ai) ### So is the bottleneck labor or power? Both, but power is the harder gate. Labor matters — Forbes described about 2,000 workers on site in early March 2025 — yet labor can usually be bought if the economics work. Grid access is different. Lancium says the Abilene campus has a fully approved 1.2-gigawatt ERCOT inte(crusoe.ai)is the scarce asset. It’s like having a runway slot at a packed airport — the building only matters if you can actually land. (forbes.com) ### What changed after the original Stargate hype? The tenant picture got messier. In March 2026, reports said Microsoft agreed to rent a large adjacent Crusoe-developed Abilene project after Oracle and OpenAI stepped away from talks on that separate capacity. That does (forbes.com)can stay valuable even when the named customer changes. (bloomberg.com) ### What does this mean for AI infrastructure? Basically, the market is discovering that “AI capex” really means “power-secured industrial construction.” The winners won’t just be model labs or chip vendors. They’ll also be developers that can line up land, substations, permits, financing, and crews fast enough to turn a power entitlement into working compute. Abilene matters because it makes that shift visible. (lancium.com) ### Bottom line The most useful number from Abilene is not a viral $59 million-per-megawatt claim. It’s that a real 1.2-gigawatt AI campus now carries about $15 billion of committed capital — and the rarest thing on site may be the grid connection, not the GPUs. (datacenterdynamics.com)as/))