Crusoe Spark Liquid Platform Boosts AI Compute Density

The modular factory system supports 3.5MW of compute for next-gen NVIDIA hardware to shorten deployment timelines.

Updated on Oct. 21, 2026 in Artificial Intelligence

Modular industrial data center unit with metallic cooling pipes situated on a concrete pad in an outdoor yard. — Crusoe has launched its Spark Liquid platform, a modular AI factory unit capable of supporting 3.5MW of compute to accelerate data center deployment. AI Illustration. Upload story photo >

Live Poll

Would you prefer your business adopt modular AI infrastructure to reduce setup costs and time?

Crusoe has introduced Spark Liquid, its sixth-generation modular AI factory platform capable of supporting 3.5MW of compute per unit. The system is designed to accelerate data center deployment and is compatible with NVIDIA GB300 and Vera Rubin hardware.

Why it matters

By shifting from traditional, multi-year facility construction to factory-built modular units, this platform aims to decrease the time required to bring large-scale compute capacity online. This shift supports localized AI inference requirements for organizations needing lower latency.

The Spark Liquid platform delivers 3.5MW of compute per unit, representing a 5x increase in compute density over Crusoe's previous generations. It incorporates a closed-loop direct liquid cooling architecture that functions without water consumption.

The players

Crusoe

A developer of modular AI infrastructure that builds and operates factory-shipped data center systems.

Ollama

A software platform serving over nine million developers and 85% of Fortune 500 enterprises, utilizing Crusoe infrastructure.

NVIDIA

A designer of graphics processing units and AI acceleration hardware, specifically the GB300 and Vera Rubin architectures.

The details

Crusoe Spark Liquid units are manufactured in a factory setting before being shipped for deployment, effectively shortening construction timelines from years to months. The architecture is engineered specifically to accommodate NVIDIA GB300 NVL72 and Vera Rubin NVL72 hardware. The system utilizes non-evaporative cooling, a method that manages heat through a sealed circulating liquid loop, eliminating the need for traditional water-intensive cooling infrastructure.

October 21, 2026: Crusoe introduced the Spark Liquid platform.
2027: Production of Spark Liquid units begins at the Brighton factory.

The Tech Race

Crusoe is positioning its sixth-generation modular hardware to meet the stringent power and cooling demands of the latest NVIDIA GB300 and Vera Rubin platforms. This development follows a broader industry trend of moving away from bespoke data center builds toward standardized, factory-built compute modules.

The platform enables large-scale enterprises and software providers to deploy AI compute closer to localized demand. It will be available for production at the Brighton factory starting in 2027.

The takeaway

This shift toward factory-built data centers highlights the growing necessity for rapid, high-density infrastructure as AI demand scales. Watch for the 2027 production ramp at the Brighton facility as an indicator of how quickly these modular units can move from assembly lines to live deployment.