Datacenter Buildouts

Permian Basin sites

The Permian Basin (West Texas, southeast New Mexico) is the historical core. Crusoe has many smaller datacenter modules at oil well-pads here, harvesting flare gas to power compute.

These sites tend to be:

• Smaller individually (single-digit-MW).
• Highly distributed (many sites rather than a few large ones).
• Operationally complex due to remote locations and harsh environments.
• Originally built for crypto mining, now repurposed or supplemented for AI compute.

The collective capacity here is meaningful but the individual-site scale is small relative to modern AI training needs. These sites work for smaller training jobs and inference; the largest frontier-model training needs the bigger consolidated sites.

Wyoming and Iceland

Wyoming sites take advantage of cheap energy in less-populated areas with favorable energy economics. Iceland sites tap into the abundant geothermal and hydroelectric power on the island, plus the favorable cooling climate.

These sites are typically larger than the Permian wellpads and better-suited to consolidated datacenter operations. The Iceland sites in particular benefit from natural cooling that reduces total power-usage-effectiveness (PUE) overhead.

Abilene, Texas

The Abilene buildout is the Stargate-era project — Crusoe has been associated with very large datacenter development in Abilene tied to OpenAI / Oracle / SoftBank's Stargate joint venture. The scale of these sites is hundreds of MW per site and would represent some of the largest AI-dedicated datacenter capacity in the country.

The Abilene work signals Crusoe's evolution from "many small flare-gas sites" to "few very large consolidated AI compute campuses." The energy thesis still applies but at much larger absolute scale.

Datacenter design

Crusoe's datacenter design has evolved across these site classes:

• Modular containers at well-pads. Original design — shipping-container-style modules with self-contained power, cooling, and racks.
• Hardened mid-scale buildings. For sites in less-extreme environments. Tens of MW per site, conventional datacenter form factor.
• Large-scale purpose-built campuses. Multi-hundred-MW facilities purpose-built for AI compute, including liquid cooling for Blackwell-generation hardware.

The evolution mirrors the company's transition from crypto-mining (which tolerated harsh environments because the hardware was cheap) to AI compute (which demands enterprise-grade reliability because the hardware is expensive).

Scaling capacity

The pace of capacity additions has been aggressive through 2024-2026:

• Multi-hundred-MW announced buildouts in various stages of construction.
• Capacity coming online matches signed customer commitments.
• The buildout schedule pushes operational complexity — new sites in new locations every quarter.

The scaling is execution-intensive. Building datacenters on the timeline AI customers want requires permitting, construction, power infrastructure, networking, and operational hiring — all in parallel.

Strategic partnerships

Crusoe's buildouts often involve strategic partners:

• Energy producers providing fuel and site access.
• Real estate developers handling some of the physical construction.
• Oracle and others as platform integration partners on customer-facing deployments.
• NVIDIA on hardware delivery and operational best practices.

The partnership-heavy approach reflects Crusoe's preference to deploy capital efficiently rather than vertically integrate every component of the build.

Takeaway

Crusoe's datacenter footprint spans well-pad-scale modules to multi-hundred-MW campuses. The newer, larger sites are where the strategic action is — they support the frontier-AI training relationships that drive the company's narrative. The next chapter examines the most important of those relationships.