GPU Cloud Pricing 2026
As of 2026-09-19. Official on-demand per GPU-hour: RunPod H100 PCIe $2.89 1st, RunPod H100 NVL $3.19 2nd, Lambda H100 PCIe $3.29 3rd.
As of 19 September 2026. The main table uses each vendor’s official on-demand list price in US dollars per GPU-hour, H100 and B200 only, ranked from low to high. RunPod H100 PCIe is 1st at $2.89, RunPod H100 NVL 2nd at $3.19, Lambda H100 PCIe 3rd at $3.29. Cloud share: cloud providers. Token inference: LLM API pricing. Capex: big-tech capex.
Official on-demand per GPU-hour
A 1-GPU SKU is used when sold. Lambda and CoreWeave SXM rows divide an 8-GPU instance by 8. AWS is us-east-1 Linux Shared. Azure is eastus Linux Consumption. RunPod’s pricing page is dated 13 September 2026.
| # | SKU | GPU | $ / GPU-hr | Window |
|---|---|---|---|---|
| 1 | RunPod H100 PCIe | H100 PCIe | $2.89 | Pods list; page dated 13 Sep 2026 |
| 2 | RunPod H100 NVL | H100 NVL | $3.19 | Pods list |
| 3 | Lambda H100 PCIe | H100 PCIe | $3.29 | 1-GPU self-serve |
| 4 | RunPod H100 SXM | H100 SXM | $3.49 | Pods list |
| 5 | Lambda H100 SXM | H100 SXM | $3.99 | 8-GPU SKU per GPU |
| 6 | CoreWeave HGX H100 | H100 SXM | $6.16 | 8-GPU instance $49.24 |
| 7 | Lambda B200 SXM6 | B200 | $6.69 | 8-GPU SKU per GPU |
| 8 | RunPod B200 | B200 | $6.79 | Pods list |
| 9 | AWS P5 H100 | H100 SXM | $6.88 | us-east-1 Linux; p5.48xlarge $55.04 / 8 |
| 10 | Azure H100 NVL | H100 NVL | $6.98 | eastus Linux; NC40ads 1 GPU |
| 11 | CoreWeave HGX B200 | B200 | $8.60 | 8-GPU instance $68.80 |
| 12 | GCP A3 High H100 | H100 SXM | $11.06 | a3-highgpu-8g $88.49 / 8 |
| 13 | Azure H100 SXM | H100 SXM | $12.29 | eastus Linux; ND96isr $98.32 / 8 |
| 14 | AWS P6 B200 | B200 | $14.24 | us-east-1 Linux; p6-b200.48xlarge $113.9328 / 8 |
How this list is ranked
As of 2026-09-19. The main table uses each vendor’s official on-demand list price in US dollars per GPU-hour, H100 and B200 only. A 1-GPU SKU is used when sold; otherwise the 8-GPU instance price is divided by 8. Ranks follow $/GPU-hour, low to high.
| Source | Snapshot | What it prints | Weight |
|---|---|---|---|
| Runpod GPU Cloud Pricing | Page dated 13 Sep 2026; compiled 2026-09-19 | Pods: H100 PCIe $2.89; H100 NVL $3.19; H100 SXM $3.49; B200 $6.79 | Main |
| Lambda AI cloud pricing | 2026-09-19 | Self-serve: H100 PCIe 1-GPU $3.29; H100 SXM 8-GPU $3.99/GPU; B200 SXM6 8-GPU $6.69/GPU | Main |
| CoreWeave Cloud Pricing | 2026-09-19 | North America on-demand: HGX H100 8-GPU $49.24; HGX B200 8-GPU $68.80. Per-GPU $6.16 / $8.60 | Main |
| AWS Price List us-east-1 (version 20260918212757, published 2026-09-18) | 2026-09-18 | Linux Shared OnDemand: p5.48xlarge $55.04; p5.4xlarge $6.88; p6-b200.48xlarge $113.9328 | Main |
| Azure Linux VM pricing + Retail Prices API | 2026-09-19 | eastus Linux Consumption: NC40ads H100 v5 $6.98; ND96isr H100 v5 $98.32 | Main |
| GCP Accelerator-optimized VM pricing | 2026-09-19 | Price column: a3-highgpu-8g $88.490000119. A4 B200 is N/A in that column | Main |
| Amazon EC2 Capacity Blocks for ML pricing | 2026-09-19 | Reservation: p5.4xlarge US $5.191; p6-b200.48xlarge US $12.355. Next update listed as October 2026 | Aux (reserved) |
| Alibaba Cloud EGS billing | 2026-09-19 | Billing rules; no public H100 / B200 USD column | Aux (pool) |
| Volcengine GPU billing | 2026-09-19 | Docs nav; no public USD column | Aux (pool) |
Reservations and committed use
AWS Capacity Blocks are reservations, not on-demand. Lambda 1-Click Clusters start at 16 GPUs for two weeks. GCP A4 B200 is N/A in the on-demand column; this table copies DWS Flex-start only.
| SKU | $ / GPU-hr | Window |
|---|---|---|
| AWS Capacity Blocks p5.4xlarge | $5.191 | US reservation. Next update listed as October 2026 |
| Lambda 1-Click H100 ×16 | $6.16 | 2 weeks–1 year |
| GCP 1-year CUD a3-highgpu-8g | $7.67 | $61.383674231 / 8 |
| GCP DWS Flex-start a4-highgpu-8g | $8.06 | $64.44 / 8; on-demand column is N/A |
| Lambda 1-Click B200 ×16 | $9.86 | 2 weeks–1 year |
| AWS Capacity Blocks p6-b200.48xlarge | $12.355 | US reservation |
H200 and nearby SKUs
H200 stays off the main table. RunPod H200 lists $4.59. CoreWeave HGX H200 is $50.44 for eight GPUs. AWS p5en.48xlarge Linux on-demand is $63.296. GCP a3-ultragpu-8g is $84.806908493. Capability: LLM rankings.
互动体验：拖拽排位《GPU Cloud Pricing 2026》
GPU Cloud Pricing 2026
想打造自己的个性化天梯图并开放给读者嵌入吗？