Cloud GPU price comparison: AWS vs Google Cloud
Every figure below is normalized per GPU: an 8×A100 node and a 1×A100 machine are compared per GPU, so the numbers are apples-to-apples. These are list prices, no negotiated discounts. Prices as of 2026-08-31.
| GPU | AWS $/GPU-hr | GCP $/GPU-hr | Idle 24/7 /mo | Up to /mo saved (est.) |
|---|---|---|---|---|
| NVIDIA B200 180 GB | $14.24/GPU-hr p6-b200.48xlarge · us-east-2 spot $2.56/GPU-hr (est.) | $16.11/GPU-hr a4-highgpu-8g · us-central1 spot $4.95/GPU-hr (est.) | $83,171 p6-b200.48xlarge | $58,418 |
| NVIDIA H200 141 GB | $7.91/GPU-hr p5en.48xlarge · us-east-2 spot $1.90/GPU-hr (est.) | $10.44/GPU-hr a3-ultragpu-8g · us-west1 spot $6.03/GPU-hr (est.) | $46,206 p5en.48xlarge | $32,454 |
| NVIDIA H100 80 GB | $6.88/GPU-hr p5.48xlarge · us-east-2 spot $1.79/GPU-hr (est.) | $10.98/GPU-hr a3-highgpu-8g · us-central1 spot $6.59/GPU-hr (est.) | $40,179 p5.48xlarge | $28,221 |
| NVIDIA A100 40 GB | $2.74/GPU-hr p4d.24xlarge · us-east-1 spot $1.21/GPU-hr (est.) | $3.67/GPU-hr a2-highgpu-1g · us-central1 spot $2.12/GPU-hr (est.) | $16,029 p4d.24xlarge | $11,259 |
| NVIDIA L40S 48 GB | $1.86/GPU-hr g6e.xlarge · us-east-2 spot $0.67/GPU-hr (est.) | — | $1,359 g6e.xlarge | $954 |
| NVIDIA A10G 24 GB | $1.01/GPU-hr g5.xlarge · us-east-2 spot $0.42/GPU-hr (est.) | — | $734 g5.xlarge | $516 |
| NVIDIA L4 24 GB | $0.80/GPU-hr g6.xlarge · us-east-2 spot $0.27/GPU-hr (est.) | $0.71/GPU-hr g2-standard-4 · us-central1 spot $0.42/GPU-hr (est.) | $516 g2-standard-4 | $362 |
| NVIDIA T4 16 GB | $0.53/GPU-hr g4dn.xlarge · us-east-2 spot $0.17/GPU-hr (est.) | — | $384 g4dn.xlarge | $270 |
| NVIDIA V100 16 GB | $3.06/GPU-hr p3.2xlarge · us-east-1 spot $0.43/GPU-hr (est.) | — | $2,234 p3.2xlarge | $1,569 |
Monthly figures are for the whole instance with the lowest $/GPU-hour in the row (named beside the figure), not per GPU — a multi-GPU box can win on $/GPU-hour while costing far more per month than a smaller machine carrying the same GPU. The est. savings assume business hours only (50 hours a week); that share is the same for every row — the rate cancels out — so the dollar column is what differs. A dash means none of the instance types we track on that cloud carries the GPU: GCP's T4 and V100 are sold as GPUs attached to N1 machines and are not tracked.
Spot figures are indicative — AWS values are derived from AWS's published typical savings vs on-demand for the region; GCP values are the listed Spot price. Spot capacity can be reclaimed by the provider.
Stop paying for idle cloud servers
Idlefy runs a free audit of your fleet, then flips dev servers to stopped-by-default: lease one when you need it, and
when the lease ends it shuts itself off — stop, not terminate, persistent disks intact. Timers are set by your team, never
guessed, and Idlefy only manages instances tagged idlefy=enabled — on AWS that boundary is enforced by the IAM policy itself.
Managing GPU instances? See /solutions/gpu/. Browse all AWS and Google Cloud instance costs.
Prices are on-demand list price and exclude storage, network, data transfer, and any negotiated or volume discounts. Actual bills vary by usage and account-level pricing. Stopping preserves EBS and persistent-disk volumes; data on local instance-store or local-SSD drives — which several of the GPU types above carry — does not survive a stop. Prices as of 2026-08-31.