Crusoe

Crusoe is a vertically integrated AI cloud that describes itself as “the energy-first AI factory company.” It sources energy, builds and operates hyperscale AI data centers, and rents that capacity as a GPU cloud for training and inference . The problem it targets is the bottleneck behind every large model project: not chips alone, but the power and physical buildings to run them. Crusoe controls the whole stack, from the turbine to the GPU, so it can add capacity without waiting on a landlord or a utility.

The company started in 2018 with patented Digital Flare Mitigation technology, converting wasted natural gas into electricity for computing. In 2025 it divested its bitcoin mining business to focus on AI infrastructure, and it now runs a diversified energy portfolio that includes geothermal and hydro power, gas turbines, and second-life EV batteries.

Where Crusoe sits in the stack

Crusoe operates lower in the stack than a typical model API provider. It owns the energy and buildings, then layers cloud services on top.

How to access it and typical use

Crusoe sells two things depending on how much you want to manage yourself.

• Crusoe Cloud gives you raw GPU infrastructure. You rent clusters of NVIDIA or AMD accelerators and run your own training or serving stack on top, with Managed Kubernetes, Managed Slurm, and AutoClusters to schedule the work. This suits teams training or fine-tuning large models who want dense, high-end GPU capacity without buying hardware.
• Crusoe Managed Inference is a platform service. You call an endpoint and Crusoe serves an open model for you, drawing from its Intelligence Foundry marketplace, which lists models such as Llama, DeepSeek, GLM, Kimi, and Nemotron. This suits teams that want production inference without operating clusters.

Access starts through the Crusoe website: request access, then provision resources through the Crusoe Cloud console and the Command Center operations view. Because Crusoe builds its own sites, capacity is often contracted in advance for larger commitments rather than clicked into existence like a hyperscaler instance.

How Crusoe compares

Crusoe belongs to the “neocloud” category: specialist GPU clouds that compete with hyperscalers on price and availability of scarce accelerators. Its differentiator is owning the energy and buildings, not renting them.

For the neighboring options, see CoreWeave , Nebius , Lambda Cloud , and the serving-focused Together AI . If you want a fully managed model API instead of infrastructure, Amazon Bedrock sits one layer higher again.

When not to use it

• You want a plug-and-play model API and nothing else. A managed API like Bedrock or a first-party lab endpoint is faster to adopt than provisioning cloud capacity.
• You need instant, self-serve elastic scaling for small jobs. Neoclouds shine at reserved, dense GPU capacity, less so at spiky micro-workloads that a hyperscaler spot instance handles cheaply.
• You are locked into one hyperscaler’s managed services. If your data, IAM, and pipelines live inside AWS, Azure, or Google Cloud, moving GPU workloads to a separate cloud adds integration and egress work.
• You need a specific region Crusoe does not serve. Because Crusoe builds physical sites, availability follows its footprint rather than a global default.

Further reading

• What is inference? : the serving step Crusoe Managed Inference is built to accelerate.
• CoreWeave : a large NVIDIA-focused neocloud competitor.
• Nebius : a neocloud with a broader managed AI platform.
• Lambda Cloud : GPU cloud focused on fast access for AI teams.
• From zero to production : how a model workload moves from prototype to a served endpoint.
• Crusoe company page : founding story, mission, and energy approach.
• Crusoe Cloud : GPU cloud services and supported accelerators.

Sources

• Crusoe homepage
• About Crusoe
• Crusoe: contracted AI infrastructure capacity approaches 5 gigawatts
• Crusoe announces new manufacturing facility to produce modular AI factories
• Crusoe to become first cloud operator in space through partnership with Starcloud