GPU Cloud Providers Directory
Directory of major GPU cloud providers with pricing snapshots and feature comparison. Last reviewed on 2026-04-30.
☁️ Major Cloud Providers
AWS (Amazon Web Services)
CLOUD
- P5 instances with H100 GPUs
- P4d instances with A100 GPUs
- Spot instances available
- Global availability (25+ regions)
- SageMaker integration
H100 (p5.48xlarge):
$98.32/hr
A100 (p4d.24xlarge):
$32.77/hr
Google Cloud Platform
CLOUD
- A3 VMs with H100 GPUs
- A2 VMs with A100 GPUs
- TPU v6e/v7 available
- Preemptible instances
- Vertex AI integration
A100 80GB:
$2.77/hr
TPU v6e:
$3.22/hr
Microsoft Azure
CLOUD
- ND H100 v5-series VMs
- ND A100 v4-series VMs
- Spot VMs available
- Azure ML integration
- Reserved instance discounts
A100 80GB:
$3.67/hr
H100 80GB:
$3.99/hr
🚀 Specialized GPU Clouds
Lambda Labs
SPECIALIZED
- H100 clusters available
- No long-term commitments
- Pre-installed ML frameworks
- Persistent storage included
- API access for automation
H100 80GB:
$2.14/hr
A100 40GB:
$2.40/hr
RunPod
MARKETPLACE
- Community cloud marketplace
- Serverless GPU endpoints
- A100, B200, MI325X, RTX 4090
- Spot pricing options
- Docker container support
A100 80GB:
$1.89/hr
RTX 4090:
$0.69/hr
Vast.ai
MARKETPLACE
- Peer-to-peer GPU marketplace
- Wide GPU selection across hosts
- Verified and unverified tiers
- Interruptible bidding
- Docker-based deployment
A100 80GB:
$1.79/hr
RTX 4090:
$0.74/hr
Paperspace
MANAGED
- Gradient notebooks
- Managed ML platform
- A100, H100, RTX GPUs
- Autoscaling clusters
- Free tier available
A100 80GB:
$2.99/hr
RTX A6000:
$1.89/hr
CoreWeave
- Kubernetes-native cloud
- H100, H200, B200 at scale
- InfiniBand networking
- Reserved capacity
- Enterprise SLAs
H100 80GB:
$2.85/hr
H200 141GB:
$4.25/hr
FluidStack
MARKETPLACE
- Aggregator across data centers
- Among the lowest A100 rates
- API-first platform
- Instant provisioning
- H100 and MI300X capacity
A100 80GB:
$1.65/hr
RTX 4090:
$0.65/hr
📊 Quick Comparison
| Provider | H100 | A100 | Consumer GPUs | Spot/Preemptible | API Access | Min Commitment |
|---|---|---|---|---|---|---|
| AWS | ✓ | ✓ | ✗ | ✓ | ✓ | Per second |
| Google Cloud | ✓ | ✓ | ✗ | ✓ | ✓ | Per second |
| Azure | ✓ | ✓ | ✗ | ✓ | ✓ | Per minute |
| Lambda Labs | ✓ | ✓ | ✗ | ✗ | ✓ | Per hour |
| RunPod | ✗ | ✓ | ✓ | ✓ | ✓ | Per hour |
| Vast.ai | ✓ | ✓ | ✓ | N/A | ✓ | Per hour |
| Paperspace | ✗ | ✓ | ✓ | ✗ | ✓ | Per hour |
| CoreWeave | ✓ | ✓ | ✗ | ✗ | ✓ | Monthly |
🎯 Provider selection notes
- For enterprise compliance: the major hyperscalers (AWS, Azure, Google Cloud) lead on SLAs, audit certifications, and integration with adjacent managed services.
- For unmanaged on-demand at lowest list price: marketplace providers such as Vast.ai and FluidStack typically offer the lowest hourly rates.
- For mature spot / preemptible markets: AWS and Google Cloud have the deepest spot capacity; see the spot instance guide.
- For H100 / H200 / B200 access without long commits: Lambda Labs, CoreWeave, and Crusoe regularly list availability on shorter terms.
- For TPU access: Google Cloud is the only public source — see the TPU explainer.
- For lower-carbon regions: Genesis Cloud (Iceland) and Nebius (Finland) operate in cleaner-grid locations; see the carbon calculator.
Always confirm current pricing and availability on the provider's own page — this directory is a research starting point, not a quote.
📚 Reading next
- On-demand vs reserved vs spot — choose the discount structure that fits the workload shape before locking in a provider.
- Hidden cloud GPU costs — storage, egress, networking, and idle time line items that the GPU rate excludes.
- Spot instance strategy — operational details for running training on preemptible capacity.