Replicate is a service for running AI models through an API, billed by run time or by input and output. The pricing page lists A100 80GB at $5.04/hr; it mentions no free credits, only a Try for free sign-in button.
Google Cloud GPUs
Google Cloud GPUs are GPU-attached Compute Engine VMs for AI training and inference. New customers get $300 in free credits; A100 80GB (a2-ultragpu-1g) is listed at about $5.07 per hour on demand.
Description of Google Cloud GPUs
Google Cloud GPUs are GPU-attached virtual machines on Compute Engine, offered in accelerator-optimized machine families that the pricing pages group as A2, A3, A4, A4X, G2 and G4. They are aimed at teams running AI training, inference and other GPU workloads on Google Cloud infrastructure.
On the accelerator-optimized pricing page, on-demand hourly prices are quoted per machine type including attached GPUs, vCPUs, memory and bundled Local SSD where applicable. Examples read today are a2-ultragpu-1g (one A100 80GB) at about $5.07 per hour, g2-standard-4 (one L4) at about $0.71 per hour, and a3-highgpu-8g (eight H100 80GB) at about $88.49 per hour. The regional selector means prices vary by region, and the region for these figures was not confirmed. GPU machine types are not eligible for sustained use discounts or flexible committed use discounts.
Google states vCPUs, GPUs and memory carry a one-minute minimum charge and are billed per second afterward, and that idle running instances are billed. Spot VMs are listed at up to 91% off on-demand prices. New customers get $300 in free credits, and all customers can use 25+ products free up to monthly limits. Google's free-trial page says no charge applies until the account is upgraded; it does not state the credit validity period or card requirement in the text read.
Features of Google Cloud GPUs
- Attaches GPUs to Compute Engine VMs across A2, A3, A4, A4X, G2 and G4 accelerator-optimized families.
- Bills GPUs per second after a one-minute minimum, with Spot VMs listed at up to 91% off on-demand prices.
- Offers new customers $300 in free credits to run, test and deploy workloads.
Frequently asked questions
Does Google Cloud offer free credits for GPUs?
Google states new customers get $300 in free credits. The page does not state the credit validity period or card requirement in the text read, so check the free trial terms.
What does an A100 80GB cost on Google Cloud?
The pricing page lists a2-ultragpu-1g at about $5.07 per hour on demand, including vCPUs, memory and Local SSD. The region was not confirmed; prices vary by region.
How is GPU usage billed?
Per Google, GPUs have a one-minute minimum charge, then per-second billing. Idle running instances are billed. Spot VMs are listed at up to 91% off.
Website: https://cloud.google.com/compute/docs/gpus
Company: Google Cloud
Key facts
GPU VMs on Google Cloud for teams already on its infrastructure; usage-billed per second with region-dependent prices. A100 80GB about $5.07/hr on demand; $300 credit for new customers.
| Pricing model | Usage-based |
|---|---|
| Free tier | Yes · $300 free credits for new customers; terms not confirmed |
| Starting price | Not verified |
| Data residency | Not verified |
| Compliance (as stated by vendor) | Not verified |
| Integrations | Not verified |
| Public API | Not verified |
| Self-hostable | Not verified |
| Official site | Not verified |
| Can a general AI assistant replace it? | Specialist GPU clouds such as RunPod, Lambda, Vast.ai or Modal are alternatives to compare on hourly rates. |
| Last verified | 2026-11-17 by editor |
Alternatives & Similar Tools
Lambda is a GPU cloud for AI training and inference, offering on-demand Instances and 1-Click Clusters. The pricing page lists A100 80GB at $3.19 per GPU hour; no free credits are mentioned.
Modal is a serverless cloud for running AI and data code on GPUs, billed per second. The Starter plan is $0 plus compute and includes $30 of free credits per month for up to 3 seats.
Google Colab is a hosted Jupyter Notebook service needing no setup, for learners and ML practitioners. It includes free access to GPUs and TPUs, though Google states GPU access is heavily restricted.
Kaggle Notebooks is a cloud environment for running machine learning code on Kaggle, used for competitions and learning. Kaggle's documentation lists a weekly GPU and TPU quota; exact hours were not readable today.
RunPod is a GPU cloud for AI training and inference, billed per second with no commitment. Pricing pages read today mention no free credits; the A100 PCIe is listed from $1.19/hr on Community Cloud.
Update history
- 2026-11-17 · ai alternative note updated (editor)
- 2026-11-17 · editor verdict updated (editor)
- 2026-11-17 · pricing page url updated (editor)
- 2026-11-17 · free tier updated (editor)
- 2026-11-17 · pricing model updated (editor)
Beyond the tool
Need the outcome rather than the software?