Comparison
Cloud GPU for AI: who offers the best perf/price?
RunPod, Lambda, Vast.ai — three platforms, three pricing models. Real April 2026 prices, on-demand, single GPU.
Data updated 2026-04-15
Hourly rates
Cost per GPU (€/h, on-demand)
| GPU | VRAM | RunPod | Lambda | Vast.ai | Best price |
|---|---|---|---|---|---|
| NVIDIA H100 SXM | 80 GB HBM3 | 3.89 € | 2.49 € | 2.85 € | Lambda |
| NVIDIA A100 SXM | 80 GB HBM2e | 1.64 € | 1.29 € | 1.10 € | Vast.ai |
| NVIDIA RTX 4090 | 24 GB GDDR6X | 0.69 € | 0.50 € | 0.35 € | Vast.ai |
Performance / Euro
FP16 TFLOPS per euro per hour
The higher the ratio, the more you get for your money.
| GPU | FP16 TFLOPS | RunPod (TFLOPS/€) | Lambda (TFLOPS/€) | Vast.ai (TFLOPS/€) |
|---|---|---|---|---|
| NVIDIA H100 SXM | 989 | 254.2 | 397.2 | 347.0 |
| NVIDIA A100 SXM | 312 | 190.2 | 241.9 | 283.6 |
| NVIDIA RTX 4090 | 165 | 239.1 | 330.0 | 471.4 |
The providers
Profile of each platform
Our take
Which provider to choose?
For heavy training
Lambda — lowest H100 prices on the market, dedicated ML infrastructure, optimised inter-GPU bandwidth.
For serverless inference
RunPod — auto-scaling, serverless API, per-second billing. Ideal for production endpoints.
For a tight budget
Vast.ai — marketplace with spot instances, the lowest prices on RTX 4090 and A100. Variable reliability.
Prices change weekly. This page is updated monthly. Listed prices are indicative — always verify on the provider's site before provisioning.
Need help choosing?
We size your GPU infrastructure
Training, fine-tuning, inference — we recommend the optimal configuration for your use case and budget.
Discuss your GPU project