AWS alternatives
Start with the GPU and configuration you need, then compare the providers below with AWS. The shared-model list comes from current published offers. A lower hourly rate can require a larger node or different storage and transfer charges.
TL;DR: which alternative fits
| Platform | Best for | Why |
|---|---|---|
| RunPod | Compare H100, H200, A100, L40S, L4 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | Secure Cloud (datacenter, reliable) and Community Cloud (peer hosts, cheaper). Per-second billing, serverless option. |
| OVHcloud | Compare H100, H200, A100, L40S, L4 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | European public cloud with GPU instances. |
| Spheron | Compare H100, H200, A100, L40S firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | Decentralized GPU network with fixed pricing. |
| Jarvislabs | Compare H100, H200, A100, L4 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | Simple per-hour GPU instances for ML. |
| Massed Compute | Compare H100, H200, A100, L40S firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | Dedicated GPU cloud, US datacenters. |
| DataCrunch | Compare H100, H200, A100, L40S firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | European GPU cloud, per-minute billing, spot options. |
| Crusoe | Compare H100, H200, A100, L40S firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | Climate-aligned AI cloud, datacenter GPUs. |
| CoreWeave | Compare H100, H200, A100, L40S firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | Large-scale AI cloud; instance-based GPU pricing. |
| Google Cloud | Compare H100, H200, A100, L4 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | Big-cloud attached GPUs (A3/A2/G2 families). Prices from the public Cloud Billing Catalog API, us-central1. |
| Hyperstack | Compare H100, H200, A100 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | European GPU cloud with on-demand and reserved pricing. |
| Scaleway | Compare H100, L40S, L4 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | European cloud, EU datacenters; bills in EUR, shown here converted at the ECB daily reference rate (see methodology). |
| Nebius | Compare H100, H200, L40S firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | AI-focused cloud with datacenter GPUs in EU/US. |
| DigitalOcean | Compare H100, H200, L40S firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | GPU Droplets incl. AMD MI300X/MI325X; per-hour billing, mainstream cloud UX. |
| Together AI | Compare H100, H200 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | AI cloud with GPU cluster pricing. |
| GMI Cloud | Compare H100, H200 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | NVIDIA reference platform partner; aggressive on-demand H200/B200 rates. |
| TensorDock | Compare H100, A100 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | GPU marketplace with host-listed machines; GPU rate excludes vCPU/RAM/storage, billed a la carte per second. |
| Prime Intellect | Compare H100, H200 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | Compute marketplace aggregating multiple clouds; on-demand and spot. |
| Thunder Compute | Compare H100, A100 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | Budget GPU VMs, per-minute billing; among the cheapest A100/H100 rates tracked. |
| Lambda | Compare H100, A100 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | AI-focused cloud with on-demand instances and 1-click clusters. Per-minute billing. |
| Microsoft Azure | Compare H100, A100 firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | Big-cloud GPU VMs (NC/ND series). Prices from the public Retail Prices API, eastus, Linux. |
| Vultr | Compare A100, L40S firm rates. A shared GPU name does not establish the same node size, region or included CPU and memory. | Global cloud with GPU instances. |
"Best for" is this page's editorial judgment per platform; "Why" is each platform's one-line profile. Live prices below are the proof layer.
Alternatives with live on-demand prices
| Provider | H100 | H200 | A100 | L40S | L4 | |
|---|---|---|---|---|---|---|
| RunPod | $2.89 | $3.79 | $1.59 | $1.09 | $0.49 | Visit → |
| OVHcloud | $2.99 | $6.20 | $3.07 | $1.80 | $1.00 | Visit → |
| Spheron | $2.65 | $4.79 | $1.43 | $2.11 | – | Visit → |
| Jarvislabs | $2.69 | $3.99 | $0.89 | – | $0.44 | Visit → |
| Massed Compute | $2.73 | $3.62 | $1.35 | $0.97 | – | Visit → |
| DataCrunch | $3.70 | $4.83 | $1.31 | $1.56 | – | Visit → |
| Crusoe | $3.90 | $4.29 | $2.00 | $1.50 | – | Visit → |
| CoreWeave | $6.16 | $6.30 | $2.70 | $2.25 | – | Visit → |
| Google Cloud | $9.80 | $9.31 | $2.93 | – | $0.560 | Visit → |
| Hyperstack | $2.50 | $3.99 | $1.35 | – | – | Visit → |
| Scaleway | $3.22 | – | – | $1.65 | $0.884 | Visit → |
| Nebius | $4.50 | $5.40 | – | $1.35 | – | Visit → |
| DigitalOcean | $4.41 | $4.47 | – | $1.57 | – | Visit → |
| Together AI | $3.99 | $5.99 | – | – | – | Visit → |
| GMI Cloud | $2.00 | $2.60 | – | – | – | Visit → |
| TensorDock | $2.25 | – | $1.50 | – | – | Visit → |
| Prime Intellect | $2.43 | $3.14 | – | – | – | Visit → |
| Thunder Compute | $3.20 | – | $1.09 | – | – | Visit → |
| Lambda | $3.29 | – | $1.99 | – | – | Visit → |
| Microsoft Azure | $6.98 | – | $3.40 | – | – | Visit → |
| Vultr | – | – | $2.40 | $1.67 | – | Visit → |
Cheapest verified on-demand rate per provider from today's snapshot; a "last verified" badge marks a provider we could not re-verify today (its rows show the last good price). "–" means the provider does not list that GPU (or we do not track it yet).
If you need X, choose Y
| If you need | Choose | Today from |
|---|---|---|
| The cheapest H100 on-demand today | GMI Cloud | $2.00/hr |
| The cheapest H200 on-demand today | GMI Cloud | $2.60/hr |
| The cheapest A100 on-demand today | Jarvislabs | $0.89/hr |
| The cheapest L40S on-demand today | Massed Compute | $0.97/hr |
| The cheapest L4 on-demand today | Jarvislabs | $0.44/hr |
Routing computed from today's verified on-demand rates among the alternatives on this page only (the same numbers as the live table above); it changes as prices change.
FAQ
How should I compare an alternative with AWS?
Match the GPU variant, GPU count and required region first. Compare the total node rate, then add the storage, transfer and billing rules for your workload. Unknown fees are not zero.
Full profile: AWS review and pricing.