Cloud GPU Pricing Comparison
Compare cloud GPU rental costs across 14+ providers. Find the cheapest H100, A100, L40S, and RTX 4090 instances with on-demand and spot pricing. Use the cloud GPU rental guide to compare billing models and total cost before choosing a provider.
Cloud GPU Pricing Summary
| GPU | VRAM | Min On-Demand | Min Spot | Savings | Providers | |
|---|---|---|---|---|---|---|
| B200 | 192GB | $3.75/hr | $6.79/hr | -81% off | 3 | Details → |
| H200 | 141GB | $2.30/hr | $3.72/hr | -62% off | 7 | Details → |
| H100 | 80GB | $1.47/hr | $1.47/hr | - | 13 | Details → |
| H100 PCIe | 80GB | $2.89/hr | $2.89/hr | - | 3 | Details → |
| H100 NVL | 93GB | $1.33/hr | $1.33/hr | - | 2 | Details → |
| MI300X | 192GB | $2.39/hr | $2.39/hr | - | 1 | Details → |
| A100 SXM | 80GB | $1.15/hr | $1.57/hr | -37% off | 9 | Details → |
| A100 PCIE | 40GB | $0.720/hr | $1.15/hr | -59% off | 6 | Details → |
| L40S | 48GB | $0.910/hr | $0.990/hr | -9% off | 4 | Details → |
| L4 | 22GB | $0.032/hr | $0.032/hr | - | 4 | Details → |
| RTX 4090 | 24GB | $0.268/hr | $0.134/hr | 50% off | 3 | Details → |
| A40 | 48GB | $0.440/hr | $0.440/hr | - | 3 | Details → |
| RTX 4080 | 15GB | $0.098/hr | $0.098/hr | - | 2 | Details → |
| RTX 6000 Ada | 48GB | $0.750/hr | $0.438/hr | 42% off | 5 | Details → |
| RTX A6000 | 48GB | $0.450/hr | $0.530/hr | -18% off | 6 | Details → |
| RTX 3090 | 24GB | $0.015/hr Lowest | $0.014/hr | 4% off | 3 | Details → |
| RTX A4000 | 16GB | $0.060/hr | $0.250/hr | -317% off | 4 | Details → |
| Tesla V100 | 32GB | $0.140/hr | $0.992/hr | -609% off | 5 | Details → |
| T4G | 32GB | $0.420/hr | $0.182/hr | 57% off | 1 | Details → |
| T4 | 16GB | $0.350/hr | $0.140/hr | 60% off | 3 | Details → |
| A10G | 192GB | $1.01/hr | $0.501/hr | 50% off | 1 | Details → |
| Gaudi | - | $1.64/hr | $0.638/hr | 61% off | 1 | Details → |
| A100 40GB | 40GB | $3.40/hr | $1.63/hr | 52% off | 2 | Details → |
| H200 141GB | 1128GB | $10.60/hr | - | - | 1 | Details → |
| H100 80GB | 640GB | $12.29/hr | $7.22/hr | 41% off | 1 | Details → |
| V100 | 16GB | $0.025/hr | $0.020/hr | 20% off | 3 | Details → |
| A100 80GB | 80GB | $1.39/hr | $1.39/hr | - | 2 | Details → |
| A10 | 24GB | $0.454/hr | - | - | 3 | Details → |
| L40 | 48GB | $0.820/hr | $0.820/hr | - | 3 | Details → |
| RTX A5000 | 24GB | $0.250/hr | $0.270/hr | -8% off | 5 | Details → |
| P100 | 16GB | $1.46/hr | $0.555/hr | 62% off | 1 | Details → |
| P4 | 8GB | $0.600/hr | $0.240/hr | 60% off | 1 | Details → |
| A100 40GB SXM | 40GB | $1.99/hr | - | - | 1 | Details → |
| RTX 4070 Ti | 12GB | $0.500/hr | $0.500/hr | - | 1 | Details → |
| RTX 3090 Ti | 24GB | $0.460/hr | $0.460/hr | - | 1 | Details → |
| RTX 3080 Ti | 12GB | $0.180/hr | $0.180/hr | - | 1 | Details → |
| RTX 3080 | 10GB | $0.170/hr | $0.170/hr | - | 1 | Details → |
| B200 SXM6 | 180GB | $6.69/hr | - | - | 1 | Details → |
| A6000 | 48GB | $1.09/hr | - | - | 1 | Details → |
| GH200 | 96GB | $2.29/hr | - | - | 1 | Details → |
| Quadro RTX 6000 | 24GB | $0.690/hr | - | - | 1 | Details → |
| A30 | 24GB | $0.410/hr | $0.210/hr | 49% off | 1 | Details → |
| RTX 3070 | 8GB | $0.130/hr | $0.130/hr | - | 1 | Details → |
| RTX 5080 | 16GB | $0.590/hr | $0.590/hr | - | 1 | Details → |
| RTX 5090 | 32GB | $0.990/hr | $0.990/hr | - | 1 | Details → |
| NVIDIA H200 NVL | 143GB | $3.39/hr | - | - | 1 | Details → |
| RTX 2000 Ada | 16GB | $0.240/hr | $0.240/hr | - | 1 | Details → |
| RTX 4000 Ada | 20GB | $0.280/hr | $0.280/hr | - | 1 | Details → |
| RTX 4000 Ada SFF | 20GB | $0.440/hr | $0.440/hr | - | 1 | Details → |
| RTX 5000 Ada | 32GB | $0.830/hr | $0.830/hr | - | 1 | Details → |
| RTX A2000 | 6GB | $0.500/hr | $0.500/hr | - | 1 | Details → |
| RTX A4500 | 20GB | $0.250/hr | $0.250/hr | - | 1 | Details → |
| RTX PRO 6000 MaxQ | 96GB | $0.500/hr | $0.500/hr | - | 1 | Details → |
| RTX PRO 6000 | 96GB | $2.09/hr | $2.09/hr | - | 1 | Details → |
| RTX PRO 6000 WK | 96GB | $1.89/hr | $1.89/hr | - | 1 | Details → |
| V100 FHHL | 16GB | $0.190/hr | $0.100/hr | 47% off | 1 | Details → |
| Tesla V100 | 16GB | $0.190/hr | $0.190/hr | - | 1 | Details → |
| V100 SXM2 | 16GB | $0.230/hr | $0.230/hr | - | 1 | Details → |
| V100 SXM2 32GB | 32GB | $0.490/hr | $0.250/hr | 49% off | 1 | Details → |
| A100 PCIe | 80GB | $0.280/hr | $0.160/hr | 43% off | 1 | Details → |
| A100 SXM 40GB | 40GB | $1.00/hr | $1.00/hr | - | 1 | Details → |
| B300 | 288GB | $7.39/hr | $7.39/hr | - | 1 | Details → |
| H100 | - | $5.54/hr | - | - | 1 | Details → |
| H200 NVL | 143GB | $3.79/hr | $3.79/hr | - | 1 | Details → |
| HGX B200 | - | $8.87/hr | - | - | 1 | Details → |
| NVIDIA RTX PRO 4500 Blackwell Server Edition | 32GB | $0.500/hr | $0.500/hr | - | 1 | Details → |
| PRO 6000 MIG 24GB | 24GB | $0.490/hr | $0.490/hr | - | 1 | Details → |
| PRO 6000 MIG 48GB | 48GB | $1.00/hr | $1.00/hr | - | 1 | Details → |
| RTX PRO 4000 | 24GB | $0.570/hr | $0.570/hr | - | 1 | Details → |
| RTX PRO 4500 | 32GB | $0.720/hr | $0.720/hr | - | 1 | Details → |
| RTX PRO 5000 | 48GB | $0.960/hr | $0.960/hr | - | 1 | Details → |
Provider Types Comparison
Hyperscalers
Enterprise-grade with SLAs, compliance, and global regions. Higher prices but maximum reliability.
GPU Clouds
Specialized for AI/ML workloads. Better GPU availability and competitive pricing.
Marketplaces
Peer-to-peer GPU rental. Lowest prices but variable availability and quality.
Need a project-level GPU rental estimate?
Use the GPU rental cost calculator to model GPU-hours, compare ownership with the GPU server cost guide, open the serverless GPU pricing calculator for bursty inference, or use the GPU pricing index to rank GPUs before choosing a model. If you are new to cloud capacity, read the guide to renting a GPU for AI first.
Buy vs Rent: Break-Even Analysis
When to Rent Cloud GPUs
- Usage less than 1,000 hours/year
- Variable or unpredictable workloads
- Need different GPU types for different tasks
- Want to avoid hardware maintenance
- Need to scale up/down quickly
When to Buy Hardware
- Usage more than 3,000+ hours/year
- Consistent, predictable workloads
- Data privacy/security requirements
- Long-term projects (2+ years)
- Have IT infrastructure to manage hardware
💡 Quick Math
That's about 14 months of 24/7 usage. Factor in electricity, cooling, and maintenance for hardware.
Calculate H100 rental cost → Calculate A100 rental → Calculate H200 rental → Calculate B200 rental → Calculate RTX 5090 rental →Cloud GPU Rental Guide
On-Demand vs Spot
Pay full price, guaranteed availability. Use for production workloads, demos, and time-sensitive tasks.
50-70% cheaper but can be interrupted. Perfect for training jobs with checkpointing, batch processing, and experiments.
Commitment Options
1-3 year commitments for 30-60% savings. Best for predictable, long-term workloads.
Commit to spend amount, not specific instances. More flexibility than reserved.
Cost Optimization Tips
- Use spot instances for training with checkpoints
- Right-size your instances (don't over-provision)
- Use auto-scaling for variable workloads
- Consider GPU clouds for better availability
- Compare multi-GPU vs single-GPU pricing
- Check regional pricing differences
Cloud GPU Pricing FAQ
Cloud GPU pricing is typically per-hour based on the GPU type, instance configuration (CPU, RAM, storage), and region. Prices shown here are per-GPU hourly rates in US regions. Multi-GPU instances may have different per-GPU costs due to interconnect and other factors.
On-demand instances are available immediately and run until you stop them - you pay full price for guaranteed availability. Spot (or preemptible) instances can be 50-70% cheaper but may be interrupted when demand is high. Spot is ideal for fault-tolerant workloads like training with checkpoints.
Prices vary frequently, but GPU-specialized clouds like Lambda Labs, CoreWeave, and RunPod typically offer lower H100 prices than hyperscalers (AWS, Azure, GCP). However, hyperscalers may offer better reserved instance discounts for long-term commitments. Check our real-time comparison above.
Yes, spot instances can be reclaimed with short notice (usually 30 seconds to 2 minutes). To handle this, implement checkpointing in your training code to save progress regularly. Most ML frameworks (PyTorch, TensorFlow) support automatic checkpointing.
It depends on your needs. For enterprise with compliance requirements, choose AWS, Azure, or GCP. For best GPU availability and pricing, consider CoreWeave or Lambda Labs. For budget experimentation, try Vast.ai or RunPod. Consider factors like GPU availability, networking speed, and support quality.
Key strategies: 1) Use spot instances for training with checkpoints, 2) Right-size instances (don't over-provision), 3) Use reserved instances for predictable workloads, 4) Compare providers regularly as prices change, 5) Consider GPU clouds over hyperscalers for better rates, 6) Use auto-scaling to avoid idle resources.