About Crusoe

Crusoe’s origin is unusual enough to be worth explaining. Oil extraction produces associated natural gas that is often flared, burned off, because there is no pipeline to carry it. Crusoe built modular data centers that sit at the wellhead and burn that gas to generate electricity instead, converting waste into compute.

The company has since expanded well past that model into purpose-built AI data centers, including a facility in Iceland, but the energy-first thinking persists and underpins its cost position.

Today it is a serious GPU cloud competing for large training contracts, with NVIDIA reference architecture clusters and the InfiniBand fabric that goes with them.

Pricing

Published on-demand rates put L40S around $1.45/hr and A100 80GB around $1.85/hr. Large H100 deployments are contracted rather than clicked, with pricing that depends on term and scale.

This is not a pay-a-few-dollars-and-experiment provider. The self-service tier exists, but the business is oriented toward sustained workloads.

Service Details

Instances are Linux VMs with a CLI, a Terraform provider, and an API. Beyond raw compute there is managed Kubernetes, object storage with an S3-compatible interface, and a managed inference product for serving open models without building the stack yourself.

Multi-node training runs over NVIDIA Quantum InfiniBand on the H100 tiers, with the topology configured for collective operations rather than left to you.

If your interest is sustainability reporting as well as cost, the energy story here is more substantive than most, and unusually well documented.

Available in: United States, Iceland

Learn More

For more information about Crusoe, visit their website at https://crusoe.ai/

Plan	Price/mo	CPU	RAM	Notes
L40S 48GB	—	16 vCPU	120GB	Around $1.45/hr
A100 80GB	—	30 vCPU	240GB	Around $1.85/hr
8x H100 SXM node	—	224 vCPU	2TB	Contract pricing, InfiniBand fabric