Crusoe Energy secures $15B financing for 1.2 GW Abilene data center campus in partnership with Blue Owl

Summary

Crusoe Energy has secured $15 billion in financing to build out a 1.2-gigawatt data center campus in Abilene, Texas, in partnership with Blue Owl Capital — a deal that illustrates how the infrastructure race for AI compute has moved decisively away from venture-style capital into institutional credit markets.

The deal structure

The financing is project-level, not a Crusoe equity raise. A dedicated propco LLC owns the campus buildings, each carrying a long-term lease with a single, undisclosed investment-grade customer. JP Morgan leads a construction debt syndicate — just over $7 billion for the expansion — with Bank of America and other lenders participating. Blue Owl, one of the larger private real estate credit platforms, co-owns the entity alongside Crusoe and provided the broader capital structure to reach the $15 billion total. Crusoe earns as both a project developer and a co-owner, collecting rent that exceeds debt service over a 15-year lease term.

Chase Mason, Crusoe's CEO, is direct about what makes the math work: long-term offtake agreements with high-credit-quality tenants are what unlock capital at this scale. Without that anchor, the numbers don't move.

Why Abilene

Crusoe's founding premise has been energy-first site selection rather than defaulting to established corridors like Northern Virginia. Abilene sits in one of the windiest corridors in the United States, where renewable developers built out substantial wind and solar capacity on the back of federal production tax credits — then found themselves with more supply than demand, frequently forced to curtail output or sell power at negative prices.

The production tax credit dynamic is worth understanding. Independent power producers receive a per-kilowatt-hour subsidy regardless of sale price, which creates an incentive to keep selling even when spot prices go negative. Once the 10-year credit window expires, curtailment becomes the only option when marginal demand disappears. A gigawatt-scale AI data center solves that problem permanently. Crusoe projects it will account for roughly 30% of total tax revenue for the city of Abilene.

For context on scale: all the data center capacity ever built in Northern Virginia totals approximately 4.5 gigawatts, accumulated over three decades. The Abilene campus alone will reach 1.2 gigawatts for a single customer, and Crusoe is evaluating other sites at 5 gigawatts.

Supply chain constraints

Mason says the bottlenecks in AI infrastructure builds shift constantly. H100 GPU scarcity was the binding constraint at launch. Now it is energy access and data center capacity. Within that, high-voltage transformers carry long lead times, with significant manufacturing in China — though Mason is not particularly worried about trade-war disruption given supply chain diversity. More acute are medium-voltage transformers, switchgear, and chillers. Crusoe has moved to manufacture electrical switchgear in-house, with factories in Tulsa, Oklahoma and outside Denver, Colorado.

The introduction of Nvidia's GB200 is accelerating a shift to liquid cooling across the industry. The chip's energy density produces heat loads that traditional aluminum heat sinks can no longer handle at scale. Crusoe's Abilene facility uses a closed-loop architecture: water circulates through copper pipes over the chips, picks up heat, passes through an external chiller, and returns to the system. The loop is filled once. Mason pushes back on media narratives suggesting data centers consume water continuously at the rate it flows over individual chips.

Demand and the training-to-inference shift

Mason is skeptical of any near-term demand slowdown narrative. The conversations he is party to show demand accelerating, not plateauing, across both training and inference workloads. His read on the paradigm shift toward reinforcement learning and test-time compute scaling is that it does not reduce demand for large clusters — it extends their useful life. Infrastructure built for a frontier training run continues to earn its keep serving inference and post-training workloads afterward.

Competitive dynamics

Mason is openly bearish on undercapitalized entrants trying to build AI data centers speculatively. He describes a wave of operators who have access to a plot of land, a nearby power line, and ambitions to land hyperscaler contracts — but lack the credit relationships and operational depth to compete at relevant scale. His argument is structural: anything at the hundreds-of-megawatt or gigawatt level requires long-term customer contracts with investment-grade counterparties to unlock institutional financing. A $20 million speculative build, in his framing, is a BB gun at a grizzly bear. The companies with the capital to anchor these deals are, by definition, the companies with the strongest balance sheets in corporate history — and they are directing that spend toward a small number of proven operators.