Crusoe

Scale AI workloads on high-performance GPUs powered by renewable energy, featuring breakthrough speed for large language model training and managed inference.

Crusoe screenshot

About Crusoe

Crusoe provides vertically integrated AI infrastructure designed to handle the heavy computational demands of modern artificial intelligence. By combining renewable energy sources with purpose-built data centers, the platform offers a sustainable alternative to traditional cloud providers. The company operates its own "AI factories," managing everything from energy generation to the virtualization layer, which allows for optimized performance and reduced environmental impact. This holistic approach ensures that innovation in intelligence is directly aligned with the future of sustainable energy production. The core of the offering is Crusoe Cloud, which grants access to the latest high-performance GPUs from both NVIDIA and AMD, including H100, H200, and MI300X instances. Users can choose between on-demand, spot, and reserved pricing models to balance cost and availability. Beyond raw compute, Crusoe offers Managed Inference services through its Foundry, allowing developers to deploy top open-source models like Llama 3.3, DeepSeek-V3, and Qwen via a high-speed inference engine with proprietary MemoryAlloy technology. This setup minimizes the complexities of managing underlying hardware while providing breakthrough speed for production-grade AI. This platform is specifically tailored for AI founders, machine learning researchers, and enterprise engineering teams who need to scale large-scale model training and real-time inference. It is particularly effective for those looking to avoid GPU supply constraints while maintaining a commitment to sustainability. Its infrastructure supports a wide range of use cases, from early-stage research and development to serving millions of active users in real-time production environments. The platform is designed to be highly collaborative, with a team that actively incorporates user feedback into the product roadmap. What distinguishes Crusoe from other cloud providers is its energy-first philosophy and extreme vertical integration. Originally known for converting wasted natural gas into compute power, the company now focuses on 100% renewable or low-carbon energy sources like wind and solar. This integration not only provides a cleaner environmental footprint but also enables Crusoe to build and scale infrastructure at a pace that matches the rapid growth of the AI industry. Their massive gigawatt-scale data center campuses demonstrate a unique capability to provide the foundational physical layers required for the next generation of AI development.

Crusoe pros & cons

Pros

  • Provides access to top-tier hardware like NVIDIA GB200 NVL72 and AMD MI355X.
  • Vertically integrated model ensures that AI workloads are powered by renewable energy.
  • Spot pricing offers significant cost savings of over 50% compared to on-demand rates.
  • Supports a 99.98% cluster uptime for mission-critical GPU workloads.
  • Offers a highly collaborative roadmap that incorporates customer feedback and requests.

Cons

  • Pricing for high-end models like the GB200 and B200 is not public and requires sales contact.
  • Spot instances are interruptible, which may not be suitable for time-sensitive production environments.
  • Minimum contract lengths for reserved capacity are not explicitly listed on the pricing page.

Crusoe use cases

  • AI founders can scale GPU capacity 5x within hours to serve millions of real-time users efficiently.
  • Machine learning researchers can utilize spot instances to train large language models at significantly reduced costs.
  • Enterprise engineering teams can meet sustainability mandates by running high-performance compute on renewable-powered infrastructure.
  • App developers can integrate top LLMs like DeepSeek or Llama via a simple pay-as-you-go API without managing infrastructure.

Crusoe features

  • spot and reserved pricing models
  • memoryalloy proprietary inference engine
  • persistent and shared disk storage
  • managed kubernetes clusters
  • renewable-powered data centers
  • managed inference foundry
  • amd mi300x and mi355x gpu instances
  • nvidia h100 and h200 gpu instances

Crusoe pricing

Is Crusoe free? No, Crusoe doesn't offer a free plan; pricing starts at $1.60 / gpu-hr.

NVIDIA H100 HGX On-demand

$3.90 / gpu-hr

  • Unthrottled compute
  • High-performance GPU
  • On-demand agility
  • 80GB GPU memory
  • Reliable support

NVIDIA H200 HGX On-demand

$4.29 / gpu-hr

  • Latest NVIDIA hardware
  • 141GB GPU memory
  • Maximum agility
  • Optimized for training
  • High throughput

NVIDIA H100 HGX Spot

$1.60 / gpu-hr

  • Interruptible compute
  • Up to 60% cost savings
  • Ideal for batch jobs
  • 80GB GPU memory
  • Scalable resources

AMD MI300X On-demand

$3.45 / gpu-hr

  • High-performance AMD GPU
  • 192GB GPU memory
  • Cost-effective scaling
  • Unthrottled performance
  • Latest hardware

Crusoe FAQs

What is the difference between on-demand and spot pricing?

On-demand pricing provides guaranteed availability for immediate AI development and mission-critical production workloads. Spot pricing offers significantly lower rates for interruptible compute, making it ideal for non-time-critical tasks like batch processing or model checkpointing.

What models are available through Crusoe Managed Inference?

The platform supports a variety of top-performing open-source models including Llama 3.3 70B, DeepSeek-V3, and Qwen3. Customers can also bring their own fine-tuned models to the Foundry for specialized performance and enhanced LLM results.

What storage options does Crusoe Cloud offer?

Crusoe provides reliable, low-latency storage options including persistent disks and shared disks. Persistent disks are priced at $0.08 per GiB/month, while shared disks are slightly lower at $0.07 per GiB/month.

Does Crusoe offer managed services for orchestration?

Yes, Crusoe offers Managed Kubernetes to simplify the deployment and scaling of AI applications across GPU and CPU resources. This service is priced at $0.10 per cluster hour.

How is Managed Inference billed?

Managed Inference is billed on a pay-as-you-go basis per one million tokens. Pricing is divided into input tokens, output tokens, and cached tokens, with specific rates depending on the model chosen, such as $0.25/$0.75 for Llama 3.3 70B.

Ratings & reviews

No reviews yet. Be the first to share how Crusoe worked for you.