GPU EconomyCloud GPU prices, read every hour
updated
H100 index$3.5426 providersWidestV10015.9×Mover, 7dB300−1.4%Coverage709 quotes63 providersReads, 24 h17,43726 runs

Google Cloud vs Hugging Face Inference Endpoints

As of , Hugging Face Inference Endpoints lists the lower on-demand rate on 3 of the 5 GPUs both offer on-demand; Google Cloud wins on 2. On the priciest shared card both list on-demand, NVIDIA B200: Google Cloud $16.11 (us-west1) vs Hugging Face Inference Endpoints $9.25 (aws) per GPU-hour on-demand — Hugging Face Inference Endpoints is $6.86 (43%) cheaper.

Cloud GPU rental prices of Google Cloud and Hugging Face Inference Endpoints, compared per GPU-hour on the GPUs both list. Where a provider has several tiers or regions, its cheapest on-demand rate is used and labelled. Community capacity (RunPod Community Cloud) and marketplace floors are shown only where a provider has nothing else, and never decide which is cheaper.

On-demand, per GPU-hour

Each price links to the pricing page it was read from and says when it was read.
GPUGoogle CloudHugging Face Inference EndpointsDifferenceCheaper
NVIDIA B200$16.11
us-west1 · GPU only
captured
Deploy ↗
$9.25
aws
captured
Deploy ↗
$6.86 (43%)Hugging Face Inference Endpoints
NVIDIA H200$9.31
us-west1 · GPU only
captured
Deploy ↗
$5.00
aws
captured
Deploy ↗
$4.31 (46%)Hugging Face Inference Endpoints
NVIDIA A100 80GB$3.93
us-central1 · GPU only
captured
Deploy ↗
$2.50
aws
captured
Deploy ↗
$1.43 (36%)Hugging Face Inference Endpoints
NVIDIA L4$0.56
us-east4 · GPU only
captured
Deploy ↗
$0.80
aws
captured
Deploy ↗
$0.24 (30%)Google Cloud
NVIDIA T4$0.35
us-east5 · GPU only
captured
Deploy ↗
$0.50
aws
captured
Deploy ↗
$0.15 (30%)Google Cloud

Spot / interruptible, per GPU-hour

GPUGoogle CloudHugging Face Inference Endpoints
NVIDIA B200$1.45
captured
—
NVIDIA H200$5.57
captured
—
NVIDIA A100 80GB$0.40
captured
—

GPU by GPU

Not compared: GPUs only one of the two lists, reserved/committed tiers, and anything either provider prices only on request. See each provider's page for its full list.

Data updated: