Live H100, H200, B200, A100 & RTX 5090 prices across 10+ providers. Stop paying AWS/Azure 3x markup.
H100 spot from $0.99/hr · updated daily
Live pricing — updated daily.
All prices USD per GPU-hour · auto-updated daily · affiliate disclosure
| Provider | H100 SXM | H100 spot | H200 | A100 | RTX 5090 |
|---|---|---|---|---|---|
| Loading live prices… | |||||
Prices collected today. Subject to change. Spot prices fluctuate.
See how much you'd save vs AWS/Azure.
Select options to see your savings
Best fits for Indian teams, EU compliance, and budget.
Don't overpay for compute you don't need.
The standard for training large LLMs (70B+ parameters) and high-throughput production inference. Unmatched memory bandwidth.
The sweet spot for fine-tuning 13B-34B models. Significantly cheaper than the H100, while still offering massive 80GB VRAM capacity.
Incredible value for small model inference (7B-8B), prototyping, and image generation (Stable Diffusion). The best performance-per-dollar.
NVIDIA confirmed H100 rental rates increased in 2026. A100 also rose ~15%. Old hardware appreciating — unprecedented.
Order today, delivery late 2027. Blackwell B200: 16-26 weeks. Reserved pools: sold out.
Startups burn $30K-100K/month on GPUs. Saving 60% = 12 months vs 30 months runway.
EU/UK/India startups can't always use US-hosted GPUs. We track who has domestic capacity.
Tell us what you need — we'll match you with available capacity at the lowest price. Free.
No spam. 2-3 providers with pricing within 24 hours.
Real data, no fluff.
Yes. H100 lead times are 36-52 weeks. B200 is allocation-only. HBM3e memory is sold out for 2026. NVIDIA confirmed H100 rental prices rose 20%.
Spot H100 on Vast.ai or Spheron at $0.99/hr. On-demand from $2.50/hr. AWS is $8.50/hr — you save 70-88%.
Yes. E2E Networks (Indian listed company) offers H100 at ₹209/hr with GST invoicing. Spheron also supports INR billing.
7B: RTX 5090 is enough for inference. 13B fine-tune: 1x H100. 70B training: 8x H100 cluster. Most startups overpay by renting H100s for models that fit on cheaper GPUs.
Under ~40% utilization, cloud wins. Most startups run 20-40% utilization. Renting at $0.99-2.50/hr beats ₹18L+ upfront for 8 H100s.
Spot instances are unused GPU capacity offered at a 60-80% discount, but they can be interrupted at any time. Use Spot for fault-tolerant workloads like model training (with checkpointing) or batch processing. Use On-Demand for live production APIs and web services where uptime is critical.
Yes. E2E Networks has H100s in Mumbai. Yotta and NxtGen also have domestic clusters. For DPDP Act compliance, E2E is best.