# Crusoe Cloud

Canonical: https://inetgeek.com/gpu/crusoe/

GPU cloud selling NVIDIA and AMD instances by the GPU-hour on demand, with spot and reserved capacity through sales, no egress fees, and token-priced managed inference.

## Who it suits

Multi-GPU training at published per-GPU rates on NVIDIA and AMD: H100 $3.90, H200 $4.29 and MI300X $3.45 per GPU-hour, L40S at $1.50, with free egress, a 99.5% per-VM uptime SLA, and SOC 2 Type II and ISO 27001 across the platform.

Consider something else if You need a single H100, newer chips at a list price or health data: the H100 comes only as an 8-GPU VM, B200, GB200 and MI355X are contact-sales, spot and reserved prices are not published, and the Terms of Service bar HIPAA health data.

## iScore

55 / 100 (80% interval 49–60), rank 10 of 13 in gpu — interval overlaps the next rank, so this is an ordering, not a finding.

inetGeek's own reading, computed from sourced facts against category peers by disclosed rules. Not a provider claim. Intervals are 80%; a provider with few documented facts is pulled toward the category middle rather than scored on what little is known.
Method: https://inetgeek.com/iscore/. Full output with every criterion's basis: https://inetgeek.com/scores.json. Every gpu peer ranked: https://inetgeek.com/gpu/crusoe/alternatives/

- Price: 40 (31–49), weight 50%, 7 of 10 criteria scored; scored for peers, not here — undocumented or stated without a number or a yes/limited/no: budget_input_price, flagship_input_price, flagship_output_price
- Capacity and limits: 53 (39–66), weight 24%, 1 of 3 criteria scored; scored for peers, not here — undocumented or stated without a number or a yes/limited/no: context_window, regions
- Capabilities: 100 (94–100), weight 3%, 0 of 1 criteria scored; scored for peers, not here — undocumented or stated without a number or a yes/limited/no: prompt_caching
- Trust and compliance: 71 (65–77), weight 13%, 8 of 12 criteria scored; scored for peers, not here — undocumented or stated without a number or a yes/limited/no: dedicated_infrastructure, pci_dss, rbac, support_included
- Evidence: 95 (95–95), weight 11%, 4 of 4 criteria scored

## What the documentation says

### Pricing

- Free tier: No — The only free offer the docs publish is a $5 credit for the managed AI services, not an ongoing free plan. ([source](https://docs.crusoecloud.com/), read 2026-09-27)
- Kind of free offer: Free credit — Credit is for Serverless Inference and Fine-Tuning; no free GPU-instance credit is published. ([source](https://docs.crusoecloud.com/), read 2026-09-27)
- What the free plan includes: $5 in credits to try Serverless Inference and Serverless Fine-Tuning. No expiry is stated, and no free allowance is published for GPU or CPU instances. ([source](https://docs.crusoecloud.com/), read 2026-09-27)
- Credit card to start: Yes ([source](https://docs.crusoecloud.com/create-an-account), read 2026-09-27)
- Entry paid plan: No plan tiers or setup fees: pay-as-you-go by the hour. The cheapest GPU is the L40S 48GB at $1.50 per GPU-hour; CPU instances start at $0.04 per vCPU-hour (general-purpose). ([source](https://www.crusoe.ai/cloud/pricing), read 2026-09-27)
- Pricing model: On-demand GPU and CPU instances billed by the hour with no minimum commitment, plus spot and reserved capacity through sales. The same pricing page also sells token-priced Serverless Inference and Fine-Tuning, and hourly dedicated inference endpoints (Self-Serve Deployments). ([source](https://www.crusoe.ai/cloud/pricing), read 2026-09-27)
- Egress over allowance: Free. Crusoe Cloud does not currently charge for ingress or egress, within a VPC or to and from the public internet. ([source](https://www.crusoe.ai/cloud/pricing), read 2026-09-27)

### Infrastructure

- Regions: Five data-center locations listed: Iceland (eu-iceland1-a), Houston (us-southcentral1-a), Virginia (us-east1-a), Sparks, Nevada (us-west1-a) and Norway (eu-norway1-a, marked as coming online soon). Not every GPU type is offered in every location. ([source](https://docs.crusoecloud.com/reference/locations), read 2026-09-27)
- Uptime SLA: 99.5% monthly uptime per GPU-enabled VM, with service credits of 10%, 25% or 100% of that VM's monthly bill below it. Persistent disks carry the same 99.5%. ([source](https://legal.crusoe.ai/#service-level-agreements), read 2026-09-27)

### Storage

- Storage price: $0.06 per GiB per month for Object Storage. Persistent disks (block storage, including each VM's OS disk) are $0.08 per GiB per month and shared disks $0.07; local ephemeral NVMe is included in the VM price. ([source](https://www.crusoe.ai/cloud/pricing), read 2026-09-27)

### GPU

- H100 SXM, per GPU-hour: $3.90 per GPU-hour on demand for NVIDIA H100 80GB HGX. Crusoe's VM docs list the H100 only as an 8-GPU VM (h100-80gb-sxm-ib.8x), so the smallest H100 rental is eight GPUs. H200 141GB is $4.29 per GPU-hour; spot prices are not printed (Contact sales). ([source](https://www.crusoe.ai/cloud/pricing), read 2026-09-27)
- Cheapest GPU, per hour: $1.50 per GPU-hour on demand for an NVIDIA L40S 48GB, which the VM docs list in 1-, 2-, 4-, 8- and 10-GPU sizes. The A100 80GB PCIe is $2.00 and the A100 80GB SXM $2.30. ([source](https://www.crusoe.ai/cloud/pricing), read 2026-09-27)
- Largest GPU offered: 288GB per GPU on AMD MI355X, which is Contact sales with no on-demand price. The largest GPU with a published on-demand price is AMD MI300X at 192GB, $3.45 per GPU-hour; NVIDIA B200 180GB and GB200 186GB are also Contact sales. ([source](https://www.crusoe.ai/cloud/pricing), read 2026-09-27)
- Billing granularity: Per hour (or per second), as the pricing FAQ puts it, without saying when each applies; the usage docs track VM usage in instance hours. ([source](https://www.crusoe.ai/cloud/pricing), read 2026-09-27)
- Reserved pricing: Quote-only. Reserved instance agreements are sold through Crusoe Cloud sales at discounted rates; no percentage or reserved rate is published. ([source](https://www.crusoe.ai/cloud/pricing), read 2026-09-27)

### Compliance

- SOC 2: SOC 2 Type II, renewed in 2026 for a May 1, 2025 to April 30, 2026 examination covering both the IaaS cloud and Managed AI Services; the report is available on request in the Trust Center. ([source](https://trust.crusoe.ai/), read 2026-09-27)
- ISO 27001: ISO 27001:2022 certified, awarded by Schellman, covering the whole Cloud Platform (IaaS, PaaS and SaaS) and Managed Inference; ISO/IEC 42001 was certified alongside it. ([source](https://trust.crusoe.ai/), read 2026-09-27)
- HIPAA: Terms of Service bar customers from using the Services to transmit, store or process health information subject to HIPAA. This is a contractual prohibition, not a statement about a HIPAA audit or BAA. The Trust Center separately lists a gated 'Crusoe Cloud: HIPAA Implementation Guide'. ([source](https://legal.crusoe.ai/#terms-of-service), read 2026-09-27)
- GDPR / data residency: No data-location commitment: the DPA lets Customer Data be processed in any country where Crusoe or its subprocessors have facilities, with SCCs for restricted EU transfers. EEA locations (Iceland, Norway) exist as region choices. ([source](https://legal.crusoe.ai/#data-processing), read 2026-09-27)
- Encryption at rest: Yes for disks: ephemeral NVMe uses self-encrypting drives (AES-XTS) that are cryptographically erased when the VM is destroyed; persistent and shared disks are encrypted at rest with keys pooled across the storage cluster. Object Storage does not support server-side encryption. ([source](https://docs.crusoecloud.com/storage/disks/overview), read 2026-09-27)
- Customer-managed keys: Only for Serverless Fine-Tuning: CMEK uses a KMS key you own in your AWS account, which Crusoe assumes through a role you control. Not offered for VMs or disks. ([source](https://docs.crusoecloud.com/identity-and-security/customer-managed-keys), read 2026-09-27)
- Role-based access control: Role-based access at organization level (org-admin, org-editor, org-reader) and project level (project-editor, project-reader); the highest held permission applies. ([source](https://docs.crusoecloud.com/identity-and-security/managing-organization-users), read 2026-09-27)
- Private networking / VPC: Each project gets a default VPC network with zonal subnets in every region it can use; custom networks and subnets are supported. Private-IP traffic is limited to instances in the same region. ([source](https://docs.crusoecloud.com/networking/vpc-networks/overview), read 2026-09-27)