Petaflare

Google Cloud GPU pricing per GPU on Compute Engine, and GPUs you pay in crypto

Google Cloud sells GPUs inside whole virtual machines, at five different kinds of price, all paid by card or invoice; our GPUs are paid in crypto only. Divided per GPU, here is what each card costs, how to get one, and how it compares with renting the same card by the hour.

Crypto-only payment, no card, no KYCUpdated 19 November 2026Prices checked 29 October 2026

Google Cloud GPU pricing on demand, divided per GPU, ran from $0.71 an hour for an NVIDIA L4 to $11.06 to $11.68 for an H100 in the Iowa region, read on 29 October 2026. An A100 80GB cost $5.07 and an H200 $10.60. Spot prices were roughly 40% to 50% lower, Flex-start lower still for jobs that can wait for capacity, and the B200 had no on-demand price at all. Every price includes the VM's CPUs, memory and local SSD, and GPUs cannot be used on the $300 free trial.

Google Cloud GPU prices per GPU-hour

Iowa (us-central1), Linux, read on Google Cloud's accelerator-optimized pricing page on 29 October 2026. Per-GPU price = machine price ÷ number of GPUs.

Source: Google Cloud accelerator-optimized VM pricing. The page loads its prices with JavaScript and could not be re-read on 19 November 2026, so these are the 29 October figures; prices vary by region.
Machine typeGPUsOn demandSpotFlex-start3-year commitment
a3-highgpu-8g8 × H100 SXM$11.06$6.59$4.79$4.86
a3-megagpu-8g8 × H100 SXM$11.68$5.70$5.04$5.08
a3-ultragpu-8g8 × H200$10.60$6.36$5.30$4.65
a4-highgpu-8g8 × B200Not offered$4.95$8.06—
a2-ultragpu-1g1 × A100 80GB$5.07$3.01$2.40$3.50
a2-highgpu-1g1 × A100 40GB$3.67$2.09—$1.29
g2-standard-41 × L4$0.71$0.42—$0.40

Source: Google Cloud accelerator-optimized VM pricing. The page loads its prices with JavaScript and could not be re-read on 19 November 2026, so these are the 29 October figures; prices vary by region.

How Google Cloud bills GPUs

On current hardware, GPUs come built into accelerator-optimized machine types: A4 for the B200, A3 for the H100 and H200, A2 for the A100 and G2 for the L4. You pay for the whole machine, including its vCPUs, memory and local SSD, which is why a single H100 there costs several times what a GPU cloud charges for the card alone. Older N1 machines take attachable GPUs such as the T4, billed separately from the machine.

Compute Engine bills per second after a one-minute minimum, and prices change from one region to the next. Disks, network egress to the internet and other services such as Cloud Storage are billed on top.

Five ways to pay for a GPU on Google Cloud

  • On demand. The list price, no commitment. One H100 on an a3-highgpu-8g works out to $11.06 an hour.
  • Spot. Spare capacity at a large discount that Google can reclaim at any time, so jobs must save checkpoints: $6.59 per H100 on the same machine.
  • Flex-start. Part of the Dynamic Workload Scheduler: you ask for capacity and your VMs start when it is available, at a discount. $4.79 per H100, $2.40 for an A100 80GB.
  • Calendar mode. Capacity reserved for fixed future dates. The 8-GPU B200 machine was $90.22 an hour this way, against $64.44 with Flex-start.
  • Committed use. One- or three-year commitments: a 3-year commitment brings an H100 down to $4.86 and an H200 to $4.65, paid whether you use them or not.

Before you launch: quota and the free trial

Two things stop many first attempts. First, the free trial: Google Cloud's $300 of credit for 90 days does not cover GPUs, and its documentation lists adding GPUs to VM instances among the things a free trial account cannot do. You must upgrade to a paid billing account. Second, quota: GPU quota is set per region and per GPU model, and large machines usually need a quota increase request before they will start.

If you only want to experiment, free notebook services are simpler: see Colab and Kaggle free GPUs and their limits.

Create a GPU VM with gcloud

With the gcloud CLI installed and a paid billing account, these commands list the GPUs available in a zone, check your GPU quota and create the smallest G2 machine with one L4. GPU VMs need the TERMINATE maintenance policy, because they cannot be live-migrated.

# GPUs available in a zone
gcloud compute accelerator-types list --filter="zone:us-central1-a"

# Your GPU quotas in the region
gcloud compute regions describe us-central1 | grep -B1 -A1 GPUS

# One L4 on the smallest G2 machine, Ubuntu, 200 GB disk
gcloud compute instances create gpu-vm \
  --zone=us-central1-a --machine-type=g2-standard-4 \
  --maintenance-policy=TERMINATE \
  --image-family=ubuntu-2204-lts --image-project=ubuntu-os-cloud \
  --boot-disk-size=200GB
The Ubuntu image has no NVIDIA driver; install it after the VM starts, or use one of Google's deep learning images instead.

Google Cloud or a GPU cloud paid in Bitcoin or USDT

The same cards, per GPU-hour: Google Cloud on 29 October 2026, Petaflare on 19 November 2026.

Google Cloud prices include large CPUs, memory and local SSD; ours are per GPU with the machine included, and none of our machines can be reclaimed while you have hours. Every card we rent is on the price list.
GPUGoogle Cloud on demandGoogle Cloud spotPetaflare, fixed
H100 80GB$11.06$6.59$2.39
H200 141GB$10.60$6.36$3.29
B200 180GBNot offered$4.95$4.89
A100 80GB$5.07$3.01$0.99

Google Cloud prices include large CPUs, memory and local SSD; ours are per GPU with the machine included, and none of our machines can be reclaimed while you have hours. Every card we rent is on the price list.

When Google Cloud is worth its price

Google Cloud makes sense when your data and services already live there: training next to BigQuery or Cloud Storage avoids moving terabytes, and identity, networking and compliance are already set up. For a GPU on its own, a GPU cloud costs a fraction: an A100 rental here is $0.99 an hour against $5.07 on demand on Google Cloud. The same gap exists at the other big clouds; see what EC2 GPU instances cost and Azure GPU pricing, or every provider in the provider comparison.

Google Cloud GPU questions

Does GCP provide GPUs?

Yes. Compute Engine offers NVIDIA B200, H200, H100, A100 and L4 GPUs in its A4, A3, A2 and G2 machine types, and older cards such as the T4 on N1 machines.

Can I use GPUs on the Google Cloud free trial?

No. A free trial billing account cannot add GPUs to VM instances; you have to upgrade to a paid account and usually request GPU quota.

What is the cheapest GPU on Google Cloud?

Of the machines in our table, the L4 on a g2-standard-4: $0.71 an hour on demand and $0.42 as spot in Iowa on 29 October 2026. Older T4 GPUs attached to N1 machines are another low-cost option.

Does Cloud Run support GPUs?

Yes, Cloud Run can attach NVIDIA L4 GPUs to container services, billed while instances run. It suits inference that scales with traffic; check Cloud Run's own pricing page for current rates.

Sources