Crusoe

Crusoe operates vertically integrated AI data centers and a GPU cloud anchored to energy sources—wind, solar, geothermal, and captured flare gas—rather than conventional grid power.

Reviewed by 7wData
API Available

On this page

Publisher review

Crusoe operates vertically integrated AI data centers and a GPU cloud anchored to energy sources—wind, solar, geothermal, and captured flare gas—rather than conventional grid power. Founded in 2018, the company pivoted from bitcoin mining in 2025 to focus exclusively on large-scale AI infrastructure. The platform includes Crusoe Cloud (hourly GPU rental), Crusoe Managed Inference (LLM deployment with MemoryAlloy latency optimization), and Spark modular data centers for on-site deployment.

Crusoe's primary customers are well-funded AI labs and foundation model builders: Figure, Cursor, Cognition, Fireworks, and others. The company reported 17x year-over-year growth in added contract value and 150% ARR growth in 2025, with 99.98% uptime across production clusters. H100s run at $3.90/hour and H200s at $4.29/hour; mid-tier A100s at $1.45–$1.95/hour.

On-demand, spot, and reserved pricing are available with multi-year discounts. The trade-off is clear: Crusoe is infrastructure, not a platform. There are no pre-built frameworks, no managed notebooks, no orchestration UI.

Customers must field specialized AI engineering teams and own their deployment pipeline. Internal employee reviews on Glassdoor cite organizational scaling challenges, and some reviewers report 8–15% performance overhead on multi-node training versus NVIDIA H100 baselines. Image generation runs 10–15% slower than comparable offerings. However, for committed production workloads and teams with strong sustainability mandates, the combination of cost competitiveness on entry-level GPUs, renewable alignment, and white-glove support justify the narrower ecosystem.

Get the AI & data signal, daily.

335k+ subscribers read this every morning. One email, both newsletters. Unsubscribe anytime.

How it works

  1. GPU Cloud with API Control

    REST API (v1alpha5) for programmatic instance creation, block storage, VPC management, and firewall rules; Go client SDK included; hourly or per-minute billing with no setup fees.

  2. Managed Inference Service

    LLM deployment with MemoryAlloy technology for ultra-low time-to-first-token; production SLAs and 24/7 enterprise support included.

  3. Modular Spark Data Centers

    Portable containerized compute facilities designed for on-site edge deployment; 150% expansion deployed at Redwood Materials (Snyder, Texas) in March 2026.

  4. Multi-Tier Pricing & Reservation

    On-demand (per-hour), spot (30–70% discount for interruptible jobs), and reserved capacity with custom long-term contracts; no data transfer charges.

  5. Renewable & Low-Carbon Power

    Infrastructure powered by wind, solar, geothermal, hydropower, and captured natural gas; battery storage partnerships with Form Energy and Redwood Materials; carbon-negative claims.

  6. GPU Portfolio

    Access to NVIDIA H200, H100, A100, L40S, GB200 (via sales), and AMD MI300X, MI355X; InfiniBand RDMA networking.

  7. Production-Grade Reliability

    99.98% cluster uptime; 24/7 support team; 100% customer satisfaction score as of October 2025.

Strengths and trade-offs

Strengths

  • Cost-competitive entry-level GPU pricing ($1.45–$1.95/hour for A100s); 5–10% cheaper than RunPod on standard instances.
  • Vertically integrated power strategy reduces operational fragility; no grid-dependent bottlenecks; renewable alignment appealing to ESG-conscious enterprises.
  • Stable, high-touch production support with 99.98% uptime; customers report reliable multi-month training runs.

Trade-offs

  • H100s at $3.90/hour are 45% more expensive than RunPod ($2.69/hour), pricing out cost-sensitive research labs.
  • Infrastructure-only offering: no managed notebooks, frameworks, or orchestration; requires in-house engineering teams and months of development.
  • Performance gaps on multi-node training (8–15% overhead) and image generation (10–15% slower); benchmarking essential before committing production workloads.
  • Limited GPU inventory; H200 and specialized accelerators face availability constraints.

Pricing context

Crusoe Cloud operates on hourly GPU rental with no setup fees. On-demand: H100 ($3.90/hr), H200 ($4.29/hr), AMD MI300X ($3.45/hr), A100 SXM ($1.95/hr), A100 PCIe ($1.45–$1.65/hr). Spot pricing offers 30–70% discounts for interruptible workloads.

Reserved capacity and multi-year contracts unlock "deepest possible discounts" (exact figures by negotiation). No egress charges. Managed Inference pricing available per-token on request. Model: consumption-based (hourly) plus reserved-capacity contracts; commercial, no free tier.

Getting started with Crusoe

  1. Sign up for Crusoe Cloud

    Navigate to the Crusoe Cloud sign-up page and create an account. Provide your email, set a password, and complete the verification process. No payment method is required initially, but you will need to add one to launch instances.

  2. Generate API credentials

    Log into the Crusoe Cloud dashboard and navigate to the API keys section. Generate a new API key and secret. Store these securely, as you will use them to authenticate all programmatic interactions with the platform.

  3. Launch a GPU instance

    Use the Crusoe Cloud REST API or the Go client SDK to create a new instance. Specify the GPU type (e.g., A100, H100), region, and instance size. Attach block storage and configure VPC and firewall rules as needed for your workload.

  4. Connect via SSH

    Once the instance is running, retrieve its public IP address from the dashboard or API response. Use an SSH client with your private key to connect to the instance. Verify connectivity and check that the GPU drivers are installed and recognized.

  5. Deploy a training job

    Upload your training script and dataset to the instance using SCP or a shared storage volume. Run your training job using your preferred framework (e.g., PyTorch, TensorFlow). Monitor resource utilization with nvidia-smi and adjust instance size or spot pricing as needed.

Frequently Asked Questions

What is Crusoe and what does it offer for AI workloads?

Crusoe operates vertically integrated AI data centers and a GPU cloud powered by renewable energy. It offers hourly GPU rental through Crusoe Cloud, managed inference for LLM deployment, and modular Spark data centers for on-site use, targeting well-funded AI labs.

How much does Crusoe charge for GPU rental per hour?

Crusoe's on-demand GPU pricing is $3.90 per hour for H100s, $4.29 for H200s, $3.45 for AMD MI300X, and $1.45 to $1.95 for A100 variants. Spot pricing offers 30–70% discounts for interruptible workloads, with no setup fees or egress charges.

What makes Crusoe's energy model different from other cloud providers?

Crusoe powers its data centers with wind, solar, geothermal, hydropower, and captured flare gas, avoiding conventional grid dependency. This vertical integration reduces operational fragility and appeals to ESG-conscious enterprises, with battery storage partnerships for reliability.

What are the performance trade-offs of using Crusoe for AI training?

Crusoe shows 8–15% performance overhead on multi-node training versus NVIDIA H100 baselines and 10–15% slower image generation. However, it offers 99.98% uptime and reliable multi-month runs, making benchmarking essential before committing production workloads.

Who are Crusoe's primary customers and what type of support do they get?

Crusoe's customers include well-funded AI labs like Figure, Cursor, Cognition, and Fireworks. They receive white-glove, 24/7 enterprise support with production SLAs, but must field specialized engineering teams as Crusoe provides infrastructure only, not managed notebooks or orchestration.

How does Crusoe compare to competitors like RunPod in pricing and features?

Crusoe's A100s are 5–10% cheaper than RunPod, but H100s at $3.90/hour are 45% more expensive than RunPod's $2.69. Crusoe lacks managed notebooks and frameworks, requiring in-house engineering, while offering renewable energy and 99.98% uptime.

Alternatives

How Crusoe compares

Direct head-to-head against 3 competitors. Picked by 7wData.

This tool

Crusoe

Pricing
Crusoe Cloud operates on hourly GPU rental with no setup fees. On-demand: H100 ($3.90/hr), H200 ($4.29/hr), AMD MI300X ($3.45/hr), A100 SXM ($1.95/hr), A100 PCIe ($1.45–$1.65/hr). Spot pricing offers 30–70% discounts for interruptible workloads. Reserved capacity and multi-year contracts unlock "deepest possible discounts" (exact figures by negotiation). No egress charges. Managed Inference pricing available per-token on request. Model: consumption-based (hourly) plus reserved-capacity contracts; commercial, no free tier.
Target
Crusoe operates vertically integrated AI data centers and a GPU cloud anchored to energy sources—wind, solar, geothermal, and captured flare gas—rather than conventional grid power.
Strength
Cost-competitive entry-level GPU pricing ($1.45–$1.95/hour for A100s); 5–10% cheaper than RunPod on standard instances.
Watch for
H100s at $3.90/hour are 45% more expensive than RunPod ($2.69/hour), pricing out cost-sensitive research labs.

AWS SageMaker

Pricing
Pay-as-you-go per instance hour; no upfront cost. Example: ml.p3.2xlarge ~$3.06/hr.
Target
Data scientists and ML engineers needing a full MLOps platform on AWS.
Deployment
Cloud, AWS regions
Strength
Deep integration with AWS ecosystem (S3, Redshift, EMR).
Watch for
Complex pricing; costs can escalate with managed training and storage.

Lambda Labs

Pricing
On-demand GPU: A100 80GB ~$1.10/hr; reserved instances available.
Target
Researchers and AI teams needing on-demand GPU clusters for training.
Deployment
Cloud, bare metal
Strength
Simple, transparent pricing with no hidden fees for networking.
Watch for
Limited geographic regions and fewer GPU types than Crusoe.

RunPod

Pricing
On-demand GPU: RTX 3090 ~$0.22/hr; A100 ~$1.09/hr.
Target
Developers needing fast, affordable GPU deployment for inference and training.
Deployment
Cloud
Strength
Instant provisioning with pre-built templates for popular frameworks.
Watch for
Less enterprise support and no long-term reserved pricing options.

User reviews

No user reviews yet. Be the first to write one.

Sources

Reporting on this tool draws on these publicly available sources.

  1. www.crusoe.ai — Company positioning as 'AI factory,' product overview, customer list (Figure, Cursor, Cognition, Fireworks, HeyGen, Modal, Luma), energy model, energy-first approach and mission.
  2. www.crusoe.ai — Founded 2018, Denver headquarters, Chase Lochmiller (CEO), Cully Cavness (COO), Series funding history ($350M Series C in 2022, $600M Series D in 2024, $1B+ Series E in 2025), Stargate 1.2 GW facility in Abilene Texas, 1.8 GW Wyoming partnership, bitcoin mining divestment to NYDIG in 2025.
  3. www.crusoe.ai — GPU pricing tiers (H200 $4.29/hr, H100 $3.90/hr, A100 variants $1.45–$1.95/hr, AMD MI300X $3.45/hr), on-demand/spot/reserved models, no setup fees, per-minute billing, no data-transfer charges.
  4. docs.crusoecloud.com — REST API v1alpha5, block storage, VPC management, firewall rules, API authentication via X-Crusoe-Timestamp and Authorization headers.
  5. deploybase.ai — H100 cost comparison vs RunPod ($3.90 vs $2.69), performance gaps (8–15% multi-node overhead, 10–15% slower image generation), batch-processing optimization, limited GPU inventory, newer auto-scaling/reserved features.
  6. www.eesel.ai — Use case (well-funded technical teams building custom models), limitations (no out-of-the-box solutions, requires expensive AI engineers, months to value), Glassdoor employee feedback (chaotic environment, organizational scaling challenges).
  7. medium.com — Energy-first positioning, renewable power integration, vertical integration strategy.
  8. www.crunchbase.com — Funding rounds, founding year, CEO/leadership.
  9. computestacker.com — 99.98% uptime claim, 24/7 enterprise support, 100% customer satisfaction score, pricing comparison matrix.
  10. www.spheron.network — Pricing comparison, sustainability differentiation, on-demand vs reserved options.