01BGPU cloud · providers· quotes observed 2026-10-14 to 2026-10-20
GPU prices by cloud provider
Each provider's on-demand list price for each chip, side by side. The matrix holds that one billing model; prices under any other billing model follow in tables of their own, never mixed into it.
On-demand list price by provider and chip
On-demand42 prices · US$ per accelerator-hour (per chip-hour for TPU and Trainium) · observed 2026-10-14 to 2026-10-20| Provider | B300$/GPU·hr | B200$/GPU·hr | GB200 NVL72$/GPU·hr | H200$/GPU·hr | H100$/GPU·hr | A100 80GB$/GPU·hr | MI355X$/GPU·hr | MI300X$/GPU·hr | L40S$/GPU·hr | V100$/GPU·hr | TPU v6e$/chip·hr | TPU v5e$/chip·hr | TPU v5p$/chip·hr |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AWSHyperscaler | $17.80 | $14.24 | $7.91 | $6.88 | $3.43 | $1.86 | $3.06 | ||||||
| AzureHyperscaler | $10.60 | $12.29 | |||||||||||
| Google CloudHyperscaler | $11.06 | $5.07 | $2.70lowest | $1.20lowest | $4.20lowest | ||||||||
| HyperstackSpecialist cloud | $7.40lowest | $6.00lowest | $3.99lowest | $3.20lowest | $1.35lowest | ||||||||
| LambdaSpecialist cloud | $6.69 | $3.99 | $2.79 | $0.79lowest | |||||||||
| NebiusSpecialist cloud | $9.50 | $8.50 | $5.40 | $4.50 | $1.55lowest | ||||||||
| Oracle OCIHyperscaler | $15.00 | $14.00 | $16.00lowest | $10.00 | $10.00 | $4.00 | $8.60lowest | $6.00lowest | $3.50 | $2.95 | |||
| Together AISpecialist cloud | $9.99 | $8.19 | $5.99 | $3.99 |
A blank cell means the provider has no on-demand list price for that chip in the current snapshot. “Lowest” marks the chip's cheapest on-demand list price — the floor its chip page leads with. Each chip page lists every quote with its source.
Capacity blocks
Capacity Blocks1 price · not comparable with on-demand list prices| Chip | Provider | US$ per hour | Observed | Source |
|---|---|---|---|---|
| Trainium2 | AWStrn2.48xlarge; 16 × Trainium2 chips | $2.24per chip-hour | AWS (trn2) published pricing |
Serverless, billed per second
Serverless / per-second8 prices · not comparable with on-demand list prices| Chip | Provider | US$ per hour | Observed | Source |
|---|---|---|---|---|
| B300 | ModalB300 GPU task; one GPU; billed per second | $7.10per GPU-hour | Modal published pricing | |
| B200 | ModalB200 GPU task; one GPU; billed per second | $6.25per GPU-hour | Modal published pricing | |
| H200 | ModalH200 SXM GPU task; one GPU; billed per second | $4.54per GPU-hour | Modal published pricing | |
| H100 | ModalH100 SXM GPU task; one GPU; billed per second | $3.95per GPU-hour | Modal published pricing | |
| A100 80GB | ModalA100 80GB GPU task; one GPU; billed per second | $2.50per GPU-hour | Modal published pricing | |
| L40S | ModalL40S GPU task; one GPU; billed per second | $1.95per GPU-hour | Modal published pricing | |
| RTX 5090 | BeamRTX 5090 PCIe 32 GB serverless GPU task; one GPU; rate published per second, billed by the millisecond | $1.09per GPU-hour | Beam published pricing | |
| RTX 4090 | BeamRTX 4090 PCIe 24 GB serverless GPU task; one GPU; rate published per second, billed by the millisecond | $0.69per GPU-hour | Beam published pricing |
Attribution for these prices: “AI Compute Tracker (aicomputetracker.com), CC BY 4.0; third-party terms apply to data from Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, GPU cloud pricing pages, Beam”.