GPUPriceBook
Compare cloud GPU rental prices per hour — H100, H200, B200, A100 and more.
In June 2026 the cheapest on-demand NVIDIA H100 (SXM) rental is about $1.93/hr on Vast.ai, and the cheapest A100 80GB is around $0.74/hr on Vast.ai. Newer H200 rents from $3.44/hr and Blackwell B200 from $3.50/hr. Specialist "neoclouds" (RunPod, Vast.ai, Lambda, Hyperstack, Vultr) are typically 2-4x cheaper than AWS, GCP and Azure for the same GPU. GPUPriceBook compares per-hour prices across 14 providers and 13 GPU models, on-demand vs spot. Use the cost calculator to price your job.
Source: Provider pricing pages. Data as of June 2026.
Cheapest GPU rental by model (June 2026)
| GPU | VRAM | Cheapest on-demand | Provider | Cheapest spot/community |
|---|---|---|---|---|
| AMD Instinct MI300X | 192GB HBM3 | $1.85/hr | Vultr Cloud GPU | — |
| NVIDIA B200 | 192GB HBM3e | $3.50/hr | Vultr Cloud GPU | $2.69/hr (AWS) |
| NVIDIA H100 PCIe | 80GB HBM2e | $1.90/hr | Hyperstack | $1.40/hr (Vast.ai) |
| NVIDIA H100 SXM | 80GB HBM3 | $1.93/hr | Vast.ai | $0.95/hr (Vast.ai) |
| NVIDIA H200 | 141GB HBM3e | $3.44/hr | Paperspace (DigitalOcean) | $2.38/hr (AWS) |
| NVIDIA A100 40GB | 40GB HBM2 | $0.58/hr | Vast.ai | $0.75/hr (Azure) |
| NVIDIA A100 80GB | 80GB HBM2e | $0.74/hr | Vast.ai | $0.35/hr (Vast.ai) |
| NVIDIA L4 | 24GB GDDR6 | $0.33/hr | Vast.ai | $0.33/hr (GCP) |
| NVIDIA L40S | 48GB GDDR6 | $0.86/hr | RunPod | $0.74/hr (Nebius) |
| NVIDIA RTX 6000 Ada | 48GB GDDR6 | $0.50/hr | Lambda | $0.95/hr (Nebius) |
| NVIDIA RTX A6000 | 48GB GDDR6 | $0.49/hr | RunPod | $0.40/hr (Hyperstack) |
| NVIDIA RTX 4090 | 24GB GDDR6X | $0.52/hr | Vast.ai | $0.20/hr (Vast.ai) |
| NVIDIA V100 | 16GB / 32GB HBM2 | $0.17/hr | Verda (formerly DataCrunch) | $0.57/hr (Azure) |
Source: Provider pricing pages. Data as of June 2026.
Snapshot June 2026 — cloud GPU prices change weekly; verify on the provider's pricing page before relying on a figure.
Popular GPU models
80GB HBM3 · from $1.93/hr on Vast.ai
NVIDIA H200141GB HBM3e · from $3.44/hr on Paperspace (DigitalOcean)
NVIDIA B200192GB HBM3e · from $3.50/hr on Vultr Cloud GPU
NVIDIA A100 80GB80GB HBM2e · from $0.74/hr on Vast.ai
NVIDIA L40S48GB GDDR6 · from $0.86/hr on RunPod
NVIDIA L424GB GDDR6 · from $0.33/hr on Vast.ai
What you can look up
- Per-GPU price tables — a page for every GPU model (H100, H200, B200, A100, L40S, L4, A6000, RTX 4090, V100, MI300X) listing every provider's per-hour price, cheapest first, on-demand vs spot.
- Per-provider lineups — a page for every provider with its full GPU catalog, prices, free credits and notes.
- Rankings — cheapest H100, cheapest A100, cheapest for inference, best spot prices and cheapest 8xH100 node.
- Head-to-head comparisons — like RunPod vs Vast.ai, RunPod vs Lambda and AWS vs RunPod.
- Cost calculator — the calculator prices a job across every provider and highlights the cheapest.
Specialist & marketplace clouds
Hyperscalers
Per-GPU figures derived from 8-GPU instance prices.
Guides
The cheapest H100 GPU rentals in June 2026, ranked — from ~$1.90/hr PCIe on Hyperstack and ~$2.30/hr SXM on Vultr to community/spot prices under $2, plus why AWS costs 3x more.
2026-06-19 How much does it cost to train an LLM in 2026?A practical, worked estimate of LLM training cost on rented GPUs — from a $200 fine-tune to a multi-million-dollar frontier run — using the 6ND FLOPs rule and June 2026 H100 prices.
2026-06-05 RunPod vs Vast.ai vs Lambda: which GPU cloud is cheapest in 2026?A head-to-head of the three most popular independent GPU clouds — RunPod, Vast.ai and Lambda — on price, GPU catalog, billing, reliability and free credits, with a June 2026 price table.
2026-05-22 On-demand vs spot GPUs: when to use which (and how much you save)Spot, community and interruptible GPUs are 30-65% cheaper than on-demand — but they can be reclaimed. A practical guide to when each makes sense, with a June 2026 savings table.
2026-04-15 Renting vs buying GPUs in 2026: when does owning an H100 pay off?A break-even analysis of renting cloud GPUs versus buying your own — with the hours-per-year crossover for an H100, plus the hidden costs of ownership most people forget.
2026-03-10 Cheapest GPUs for Stable Diffusion and AI inference in 2026You don't need an H100 to run Stable Diffusion or serve a model. The cheapest GPUs for inference in June 2026 — RTX 4090, L4, L40S and A6000 — with prices and what each is good for.
2026-02-18Where the data comes from
Every figure is a dated snapshot taken from each provider's own published pricing page — RunPod, Lambda, Vast.ai, CoreWeave, Together AI, Paperspace/DigitalOcean, Vultr, Hyperstack, Verda (formerly DataCrunch), Crusoe, Nebius, AWS, Google Cloud and Azure. Cloud GPU pricing fell sharply through 2025-2026 and still changes weekly, so each page shows the June 2026 snapshot and links to the vendor's page. Hyperscaler per-GPU prices are derived from their 8-GPU instance prices (instance price / 8) — see our methodology. Always verify the current price with the provider before renting.
Last updated: 2026-06-21