Crusoe Cloud vs Nebius

A cloud infrastructure provider offering GPU compute, managed Kubernetes, and serverless services for large-scale AI workloads. Users can run open-source models, fine-tune proprietary models, or serve inference endpoints with published pricing and flexible commitment options.

Hybrid usage-based model: on-demand GPU compute ($1.50–$4.29/GPU-hr), spot and reserved options, serverless inference and fine-tuning (per-token pricing), storage ($0.06–$0.10/GiB/month). No stated free tier.

• GPU Instances — On-DemandFrom $1.50/GPU-hr to $4.29/GPU-hr on-demand; spot and reserved options available
  • ✓Latest high-performance GPUs
  • ✓Unthrottled compute
  • ✓NVIDIA GB200 NVL72
  • ✓AMD MI355X support
• CPU Instances$0.04–$0.09/vCPU-hr
  • ✓Data processing
  • ✓Model checkpointing
  • ✓Cluster orchestration
• Storage$0.06–$0.10/GiB/month
  • ✓Persistent disks
  • ✓Shared disks
  • ✓Container registry
  • ✓Object storage
• Serverless Fine-Tuning$0.40–$10.00 per 1M input tokens
  • ✓One-click deployment
  • ✓No cluster provisioning
  • ✓No surprise bills
  • ✓Full model portability
• Serverless InferenceInput/output token pricing varies by model; from $0.05 to $5.00+ per 1M tokens
  • ✓DeepSeek V3, V4 Pro, V4 Flash
  • ✓Gemma, GLM, GPT-OSS, Kimi K2.6
  • ✓Llama 3.3, Nemotron 3 family
  • ✓Qwen 3
  • ✓Up to 9.9x faster time to first token
  • +1 more
• Self-Serve Deployments$5.50/GPU-hr; volume and monthly rates available
  • ✓Deploy in minutes
  • ✓Open or fine-tuned models
  • ✓No infrastructure overhead
• Tailored DeploymentsContact sales
  • ✓Direct team support
  • ✓Highest optimization
  • ✓Benchmarked endpoints
  • ✓Bring your own model
• Provisioned ThroughputContact sales
  • ✓Guaranteed throughput
  • ✓Discounts for longer commitments
• Managed KubernetesPer-cluster-hour pricing (specific rates not listed)
  • ✓Full cluster management
  • ✓GPU and CPU orchestration
  • ✓Autoscaling
  • ✓Fault tolerance via AutoClusters

Not documented yet.