Crusoe Cloud AI features and pricing (October 2026)

Crusoe Cloud is a GPU cloud for AI training and inference that runs in data centers Crusoe builds and powers itself. It serves AI labs, platform teams and enterprises that need GPU clusters, managed Kubernetes or Slurm, or managed inference APIs.

Crusoe Cloud offers NVIDIA GB200 NVL72, B200, H200, H100, A100 and L40S as well as AMD MI300X and MI355X, on demand, as spot capacity or reserved. Crusoe Managed Kubernetes, Crusoe Managed Slurm and AutoClusters handle cluster operations. Managed Inference serves open models from 2B to 753B parameters as serverless APIs, self-serve deployments or tailored deployments with SLAs, and Serverless Fine-Tuning adapts models without a cluster.

Founded in 2018, Crusoe describes itself as vertically integrated: it sources energy, builds data centers, including sites in Texas, Iceland and Norway, and manufactures modular data centers. The contracting entity is Crusoe Technologies LLC, and a European headquarters opened in Dublin in 2025.

• Runs large training clustersCustomers provision multi-node GPU clusters on Crusoe Managed Kubernetes or Crusoe Managed Slurm. AutoClusters are described as fault tolerant, which targets long training runs where a failed node would otherwise stall a job and require manual recovery by the platform team.
• Serves open models with a shared cacheManaged Inference uses MemoryAlloy, a cluster-native memory fabric that persists across nodes and routes requests so that repeated prefill work is avoided. The vendor states 9.9x faster time to first token than vLLM on Llama 3.3 70B in its own four-node benchmark.
• Fine-tunes models without a clusterServerless Fine-Tuning adapts open models with LoRA-based workflows, and a tuned model can go to a self-serve inference deployment in one click. Teams skip provisioning and tearing down GPU clusters for short tuning jobs.
• Offers pay-per-token model APIsServerless Inference exposes curated open models from labs such as DeepSeek, Google, Meta, OpenAI, Alibaba and NVIDIA through API keys, priced per million tokens. It suits experiments and smaller workloads before a team moves to a dedicated deployment.

• Crusoe builds its own data centers and sources their power, including wind, solar, hydropower and geothermal energy, according to the vendor.
• On-demand GPU prices are public, from $1.50 per GPU hour for an L40S to $4.29 for an H200, as of September 2026.
• AMD MI300X and MI355X accelerators are offered alongside NVIDIA GB200, B200, H200 and H100 GPUs.
• ISO/IEC 27001 and ISO/IEC 42001 certificates cover the cloud platform and Managed Inference; a SOC 2 Type II report is available via the trust center.

The contracting party is the US company Crusoe Technologies LLC, and its terms name courts in Denver, Colorado. Crusoe runs its own data centers in Texas.

Prices for GB200, B200 and MI355X as well as spot and reserved capacity come from sales, so request quotes for the full term. The public pages do not list regions; confirm where instances and storage will run, especially for EU personal data, and review the Data Processing and Security Terms on the legal site. If you plan to use spot capacity, check the interruption rules.

No reviews yet. Be the first to share how your team uses it.

Every review shows its verification level. We read each one before it goes live. How we check reviews