Key facts
Price examples are static snapshots checked 2026-07-12.
Rows mix different markets, so the note column explains what is and is not included.
The cheapest row can still be the wrong row if host trust, region, storage, or interruption risk does not fit.
Snapshot checked 2026-07-12
Sortable public GPU-hour examples
Open the source page before buying. Rows differ in included CPU, RAM, storage, support, and trust assumptions.
| Source | ||||||
|---|---|---|---|---|---|---|
| Runpod | H100 PCIe | 80 GB | Pods community cloud | $2.89/hr | Per GPU-hour, per-second metering shown on pricing page | Source |
| Runpod | H100 SXM | 80 GB | Pods community cloud | $2.99/hr | Per GPU-hour, per-second metering shown on pricing page | Source |
| Lambda | H100 PCIe | 80 GB | 1x self-serve instance | $3.29/hr | Per GPU-hour, plus applicable sales tax | Source |
| Lambda | H100 SXM | 80 GB | 8x self-serve instance | $3.99/hr | Per GPU-hour, plus applicable sales tax | Source |
| CoreWeave Classic | H100 PCIe | 80 GB | Classic GPU component | $4.25/hr | GPU component per hour; CPU, RAM, and root disk are separate | Source |
| Runpod | A100 PCIe | 80 GB | Pods community cloud | $1.39/hr | Per GPU-hour, per-second metering shown on pricing page | Source |
| Runpod | A100 SXM | 80 GB | Pods community cloud | $1.49/hr | Per GPU-hour, per-second metering shown on pricing page | Source |
| Lambda | A100 SXM | 80 GB | 8x self-serve instance | $2.79/hr | Per GPU-hour, plus applicable sales tax | Source |
| CoreWeave Classic | A100 PCIe | 80 GB | Classic GPU component | $2.21/hr | GPU component per hour; CPU, RAM, and root disk are separate | Source |
| Runpod | L40S | 48 GB | Pods community cloud | $0.99/hr | Per GPU-hour, per-second metering shown on pricing page | Source |
| CoreWeave | L40S | 48 GB | 8-GPU on-demand node | $2.25/hr per GPU | Derived from published $18.00/hr 8-GPU node | Source |
| Runpod | RTX 4090 | 24 GB | Pods community cloud | $0.69/hr | Per GPU-hour, per-second metering shown on pricing page | Source |
Static snapshot rows are intentionally conservative. Marketplace rows can move minute to minute; enterprise rows can require private quotes.
How to interpret the result
Treat this page as a planning worksheet. Static examples and transparent formulas are useful for narrowing options, but real procurement still needs source quotes, workload benchmarks, storage and egress estimates, and failure-mode checks.
For buyer workflow context, read Buyer Guide, How Compute Is Priced, and GPU Cloud Price Comparison.
FAQ
Why are hyperscaler GPU prices not the whole table?
Hyperscaler GPU pricing is usually attached to exact regions, instance shapes, operating systems, commitments, quota, and calculators. This tool focuses on public per-GPU examples that can be compared without pretending they are full procurement quotes.
Can I sort by actual total workload cost?
Use this page to choose candidate hourly rates, then use the GPU cost estimator or inference-cost calculator to convert those rates into workload cost.
Why does one provider have multiple H100 rows?
H100 PCIe, H100 SXM, H100 NVL, single-GPU instances, and multi-GPU nodes can have different prices and performance characteristics.