With $2.77 billion raised and 4.9 GW of AI infrastructure contracted, Crusoe Cloud is a GPU cloud that runs on renewable and stranded energy. Instances start at $0.90/hr (A40) up to $4.29/hr (H200), with zero network egress fees. ML and AI teams get H100 or H200 clusters with Managed Kubernetes, load balancers, and a managed inference service powered by MemoryAlloy KV caching.
Founded in 2018 and backed by $2.77 billion in funding, Crusoe Cloud is a GPU compute provider that powers AI workloads with renewable and stranded energy. It offers H100 instances at $3.90/hr and H200 at $4.29/hr with no ingress or egress fees, across 4 cloud regions in the US and Norway. Crusoe Managed Kubernetes and a proprietary inference engine with MemoryAlloy KV caching are included on the same platform.
Frequently Asked Questions
What is Crusoe Cloud and what does it offer?
Crusoe Cloud is a GPU compute cloud operated by Crusoe Energy Systems, a company founded in 2018 in Denver, Colorado by Chase Lochmiller and Cully Cavness. It provides on-demand, spot, and reserved GPU instances powered by renewable and stranded energy sources, including captured flare gas from oil fields and hydroelectric power. The platform offers NVIDIA H100, H200, A100, L40S, A40, and B200 GPU instances across 4 regions in the US and Norway. Beyond raw compute, Crusoe Cloud includes Crusoe Managed Kubernetes (CMK), L4 passthrough load balancers, VPC networking with InfiniBand for GPU-to-GPU communication, and a managed inference service with proprietary MemoryAlloy technology. As of June 2026, Crusoe has contracted 4.9 gigawatts of AI infrastructure capacity and has raised $2.77 billion in funding. The company charges nothing for network ingress or egress, making it one of the few GPU clouds with genuinely zero network fees.
How much does Crusoe Cloud cost?
Crusoe Cloud prices GPU instances on a per-GPU, per-hour basis with three billing modes: on-demand, spot, and reserved. On-demand pricing starts at $0.90/hr for a single A40 GPU (48GB VRAM), $1.45/hr for an A100 PCIe, $3.90/hr for an H100 SXM5 (80GB), and $4.29/hr for an H200 (141GB HBM3e). Spot instances are available at discounts up to 90% off on-demand hyperscaler rates; L40S spot pricing has been observed at $0.50/hr. Reserved instance contracts at 6-month to 3-year terms unlock further reductions from on-demand prices. B200, GB200 NVL72, and AMD MI355X GPU capacity requires a direct sales conversation for pricing and is not available for self-serve sign-up. Storage is billed per GiB per month, and there is no free tier or trial credit. Network egress and ingress are both free.
Where are Crusoe Cloud data centers located?
Crusoe Cloud operates 4 cloud regions: two in the United States (including its flagship campus in Abilene, Texas, where the first buildings of a 1.2 GW facility are operational) and one in Norway (powered by 100% hydroelectric energy through a partnership with Polar). The fourth region covers the US East. Beyond cloud compute, Crusoe operates energy-first compute sites in Canada (Alberta) that run on captured flare gas. As of June 2026, Crusoe has contracted 4.9 gigawatts of AI infrastructure capacity globally, with a development pipeline exceeding 40 GW. The company paused work on a planned 1.8 GW campus in Wyoming in June 2026. Crusoe also announced a partnership with Starcloud to deploy cloud GPU capacity on a satellite launching in late 2026, targeting availability in early 2027.
Does Crusoe Cloud offer GPU servers?
Yes, GPU compute is Crusoe Cloud's primary product. The platform offers NVIDIA A40 (48GB), A100 PCIe (80GB), L40S (48GB), H100 SXM5 (80GB), H200 SXM (141GB), and B200 instances, along with AMD MI355X and NVIDIA GB200 NVL72 for enterprise customers via sales. On-demand pricing starts at $0.90/hr for A40 and reaches $4.29/hr for H200, with spot pricing available at up to 90% off. GPU clusters use InfiniBand networking for high-bandwidth GPU-to-GPU communication, and NVIDIA RDMA-aware topology placement ensures peak performance across multi-node training runs. Crusoe was one of the first cloud providers to offer B200 VMs, which became available in early 2026. The Crusoe Managed Inference service adds proprietary optimization with MemoryAlloy KV caching on top of the raw GPU capacity.
What are the best alternatives to Crusoe Cloud?
The most direct alternatives to Crusoe Cloud for GPU compute are CoreWeave, Lambda Labs, and RunPod. CoreWeave focuses on Kubernetes-native GPU infrastructure and has a larger US region footprint with higher-volume GPU inventory, making it better suited for very large-scale training runs; however, it charges for egress where Crusoe does not. Lambda Labs offers simple on-demand GPU access with one-click cluster deployment, appealing to research teams that prefer minimal infrastructure management over a managed Kubernetes platform. RunPod and Vast.ai offer community GPU marketplaces with lower entry prices but less consistency in hardware availability and SLA guarantees. Hetzner is an option for CPU-heavy or budget-sensitive workloads in Europe, though it has very limited GPU offerings. Pick Crusoe if you need zero egress fees, energy sustainability credentials, and managed inference with MemoryAlloy; pick CoreWeave for higher-volume GPU inventory at enterprise scale.
Who is Crusoe Cloud best for?
Crusoe Cloud is best for ML engineers and AI platform teams running large distributed training jobs on H100 or H200 clusters who want to avoid hyperscaler egress fees on data-intensive workloads. AI teams deploying high-throughput inference for open-weight models like Llama 3 or Mistral are a strong fit, particularly those using the Crusoe Managed Inference service with MemoryAlloy for prefix-heavy requests. Research teams and AI startups with fault-tolerant training pipelines benefit from spot instance pricing at up to 90% off, with L40S spot capacity observed at $0.50/hr. Organizations with sustainability mandates benefit from Crusoe's renewable and stranded energy model, documented annually in the company's Impact Report. Crusoe is not a good fit for teams needing FedRAMP, HIPAA, or PCI-DSS certified compute, for developers needing a general-purpose cloud with CDN or object storage, or for teams requiring GPU capacity in Asia Pacific or South America.
Does Crusoe Cloud support Kubernetes?
Yes, Crusoe Cloud offers Crusoe Managed Kubernetes (CMK), a fully managed Kubernetes service that was in production use by a significant portion of Crusoe's customer base as of January 2026. CMK uses Cilium as the container network interface (CNI) and supports Kubernetes v1.32. The Crusoe Load Balancer Controller integrates with CMK to automatically provision L4 passthrough load balancers for Kubernetes Service objects of type LoadBalancer, with each load balancer receiving a static public IPv4 address. For distributed GPU training, CMK supports NVIDIA Run:ai and Slurm on Kubernetes for scheduling and orchestration across multi-node GPU clusters. Compared to AWS EKS or GKE, CMK is tighter in scope but integrates directly with Crusoe's InfiniBand GPU networking, which gives a performance advantage for GPU-dense distributed training. There is no separate control plane fee; you pay only for the underlying GPU or CPU instance hours.
How does Crusoe Cloud compare to CoreWeave in 2026?
Crusoe Cloud and CoreWeave are both GPU-first cloud providers targeting AI training and inference, but they differ in scale, region footprint, and pricing structure. CoreWeave has surpassed $5 billion in annual revenue and offers a larger US region portfolio and higher raw GPU inventory volume, making it the stronger choice for very large enterprise training clusters needing geographic redundancy. Crusoe offers H100 SXM5 at $3.90/hr and H200 at $4.29/hr on-demand with zero egress fees, while CoreWeave charges for outbound data transfer, which adds materially to total cost on large training or inference deployments. Crusoe's MemoryAlloy-powered Managed Inference has no direct equivalent in CoreWeave's product catalog as of June 2026. Crusoe has 4 cloud regions in the US and Norway versus CoreWeave's broader US footprint, which matters for latency-sensitive or data-residency use cases. Pick Crusoe for zero egress fees, renewable energy credentials, and a managed inference layer; pick CoreWeave for higher GPU inventory headroom or wider US region options.
Top Alternatives
- Hetzner: Pick Hetzner if you need budget-friendly VPS or CPU servers in Europe; pick Crusoe if you need H100 or H200 GPU clusters for AI training or inference.