Compute
GPU cloud (on-demand)
Dedicated GPU instances rented by the hour.
What it is
GPU clouds rent you virtual machines or containers with dedicated GPUs (H100, H200, B200, A100, L40S, RTX 4090…) billed by the hour. You run anything: training, fine-tuning, inference servers, rendering.
How pricing works
Billed per GPU per hour, sometimes per second. Multi-GPU nodes are priced per node. Storage, egress bandwidth and idle time cost extra; commitments (1–3 years) give large discounts.
We compare everything in one unit: USD per GPU-hour.
Typical uses
- Training and fine-tuning models
- Self-hosting open-weight LLMs
- Batch inference
- Rendering and simulation
What to compare
- GPU model and VRAM
- Interconnect (NVLink, InfiniBand) for multi-GPU
- Availability and regions
- Billing granularity
Watch out for
- The cheapest listed GPU is often unavailable; check capacity.
- Egress and storage fees can exceed GPU cost for data-heavy jobs.
- Compare like with like: H100 SXM ≠ H100 PCIe.
All 20 APIs, cheapest first
Free to browseTick 2 to 4 APIs to compare them side by side.
| API | $ / GPU-hr | GPU | VRAM | Interconnect | Quality | Price data | Compare |
|---|---|---|---|---|---|---|---|
| DeepInfra dedicated A100 80GB DeepInfra | $0.89 | A100 80GB | 80 GB | — | – | Verified Oct 6, 2026 | |
| Nebius L40S (Intel CPU) Nebius | $1.55 | L40S | 48 GB | none | – | Verified Oct 6, 2026 | |
| DigitalOcean GPU Droplet L40S DigitalOcean (Paperspace) | $1.57 | L40S | 48 GB | — | – | Verified Oct 6, 2026 | |
| RunPod A100 80GB (Secure Cloud) RunPod | $1.59 | A100 80GB | 80 GB | none | 78 | Verified Oct 4, 2026 | |
| DeepInfra dedicated H100 DeepInfra | $2.20 | H100 SXM | 80 GB | — | – | Verified Oct 6, 2026 | |
| Crusoe A100 80GB SXM Crusoe Cloud | $2.30 | A100 80GB | 80 GB | nvlink | – | Verified Oct 6, 2026 | |
| HF Endpoints A100 (x1) Hugging Face Inference Endpoints | $2.50 | A100 80GB | 80 GB | — | – | Verified Oct 6, 2026 | |
| DigitalOcean GPU Droplet MI300X DigitalOcean (Paperspace) | $2.59 | MI300X | 192 GB | — | – | Verified Oct 6, 2026 | |
| Crusoe AMD MI300X Crusoe Cloud | $3.45 | MI300X | 192 GB | — | – | Verified Oct 6, 2026 | |
| DeepInfra dedicated B200 DeepInfra | $3.69 | B200 | 180 GB | — | – | Verified Oct 6, 2026 | |
| Nebius HGX H100 Nebius | $3.85 | H100 SXM | 80 GB | infiniband | – | Verified Oct 6, 2026 | |
| Crusoe H100 (HGX) Crusoe Cloud | $3.90 | H100 SXM | 80 GB | infiniband | – | Verified Oct 6, 2026 | |
| Lambda H100 SXM Lambda | $4.29 | H100 SXM | 80 GB | nvlink | 85 | Verified Oct 4, 2026 | |
| Crusoe H200 (HGX) Crusoe Cloud | $4.29 | H200 | 141 GB | infiniband | – | Verified Oct 6, 2026 | |
| DigitalOcean GPU Droplet H100 DigitalOcean (Paperspace) | $4.41 | H100 SXM | 80 GB | — | – | Verified Oct 6, 2026 | |
| Nebius HGX H200 Nebius | $4.50 | H200 | 141 GB | infiniband | – | Verified Oct 6, 2026 | |
| CoreWeave H200 CoreWeave | $6.31 | H200 | 141 GB | infiniband | 92 | Verified Oct 4, 2026 | |
| EC2 p5 (H100 SXM ×8), per GPU Amazon Web Services | $6.88 | H100 SXM | 80 GB | infiniband | 90 | Unverified sample | |
| Lambda B200 Lambda | $6.99 | B200 | 180 GB | nvlink | 95 | Verified Oct 4, 2026 | |
| Nebius HGX B200 Nebius | $7.15 | B200 | 180 GB | infiniband | – | Verified Oct 6, 2026 |