# GPU Cloud Providers Compared Independently

> Twelve GPU cloud providers compared on dated prices, billing step, node size, storage and egress: marketplaces, neoclouds, serverless and hyperscalers.

- Canonical URL: https://computesuppliers.com/gpu-cloud-providers/
- Source: [Providers' own pricing pages, read 26 October 2026](https://lambda.ai/pricing)
- Checked: 2026-10-26
- Structured data: https://computesuppliers.com/data/gpu-cloud-providers.json

## Four kinds of GPU cloud

| Kind | Examples | Good for | Watch out for |
|---|---|---|---|
| Neocloud (GPU-first cloud) | Lambda, Hyperstack, CoreWeave, Together AI, Jarvislabs | training and inference at lower hourly rates | capacity can sell out; some rent only whole nodes |
| Developer cloud | RunPod, DigitalOcean, Vultr | quick single-GPU work, templates | billing minimums, storage rules |
| Serverless | Modal, RunPod Serverless | bursty inference you pay for by the second | CPU and memory billed on top; cold starts |
| Hyperscaler | AWS, Azure, Google Cloud, Oracle | integration with an existing cloud, compliance | much higher on-demand GPU rates; 8-GPU VMs |
| Marketplace | Vast.ai | lowest prices from independent hosts | reliability varies by host; prices set by hosts |

## Provider by provider (26 October 2026)

| Provider | H100 on-demand | Billing step | Egress | Notes |
|---|---|---|---|---|
| Hyperstack | $2.50 PCIe, $3.20 SXM | per minute | free | prepaid balance ([source](https://www.hyperstack.cloud/gpu-pricing)) |
| RunPod | $2.89 PCIe, $3.99 SXM | per second | not stated | idle volumes $0.20/GB-month ([source](https://www.runpod.io/pricing)) |
| Jarvislabs | $3.49 | per minute | free in region | spot from $1.54 ([source](https://jarvislabs.ai/pricing)) |
| Modal | $3.95 | per second | $0.04/GiB after 1 TiB | serverless; CPU and RAM extra ([source](https://modal.com/pricing)) |
| Lambda | $3.99 | not stated | free | ([source](https://lambda.ai/pricing)) |
| Together AI | $3.99 | hourly | not stated | spot $1.99 ([source](https://www.together.ai/pricing)) |
| DigitalOcean | $4.41 | per second, 5-minute minimum | large allowance included | billed while powered off ([source](https://www.digitalocean.com/pricing/gpu-droplets)) |
| CoreWeave | $6.16 per GPU | not stated | free | 8-GPU nodes only ([source](https://www.coreweave.com/pricing)) |
| Azure | $12.29 per GPU | per second | paid | ND H100 v5 is an 8-GPU VM ([source](https://prices.azure.com/api/retail/prices)) |

Vultr lists H100 only on long contracts, and Vast.ai prices are set live by hosts, so neither has an H100 on-demand price in our table. Google Cloud, AWS and Oracle are being checked by hand.

## How to choose

1. Size the job first with the [VRAM calculator](https://computesuppliers.com/llm-vram-calculator/).
2. Price the whole job with the [cost calculator](https://computesuppliers.com/gpu-cost-calculator/).
3. Check capacity and region on the provider's page before you start.

---
Compute Suppliers · Contact: contact@computesuppliers.com