GPU price comparison: the cheapest cloud GPU rental, verified daily
The cheapest live $/hr for every rentable GPU across 34 providers, hyperscalers and AI cloud specialists alike. Every price links to its source and shows when it was checked. Nothing stale is ever shown as live.
Showing 32 GPU models across all tiers and providers. Prices shown per GPU-hour.
| GPU | VRAM | On-demand $/hr | Spot/community from | Where | Tier | Verified | |
|---|---|---|---|---|---|---|---|
| H100 | 80-94 GB | $1.99 | $0.94 | Voltage Park on-demand+31 more | datacenter | 2026-08-12 | Rent → |
| H200 | 141-143 GB | $2.60 | $3.31 | GMI Cloud on-demand+21 more | datacenter | 2026-08-12 | Rent → |
| A100 | 40-80 GB | $0.89 | $0.68 | Jarvislabs on-demand+20 more | datacenter | 2026-08-12 | Rent → |
| B200 | 180 GB | $3.49 | $3.20 | Prime Intellect on-demand+17 more | datacenter | 2026-08-12 | Rent → |
| L40S | 48 GB | $0.88 | $0.79 | Massed Compute on-demand+15 more | datacenter | 2026-08-12 | Rent → |
| B300 | 288 GB | $4.99 | $5.81 | Prime Intellect on-demand+11 more | datacenter | 2026-08-12 | Rent → |
| L4 | 24 GB | $0.44 | $0.10 | Jarvislabs on-demand+9 more | datacenter | 2026-08-12 | Rent → |
| L40 | 48 GB | $0.79 | $0.69 | Thunder Compute on-demand+5 more | datacenter | 2026-08-12 | Rent → |
| Tesla V100 | 16 GB | $0.17 | $0.19 | DataCrunch on-demand+5 more | datacenter | 2026-08-12 | Rent → |
| A10 | 24 GB | $1.00 | – | OVHcloud on-demand+4 more | datacenter | 2026-08-12 | Rent → |
| Tesla T4 | 16 GB | $0.35 | $0.18 | Google Cloud on-demand+4 more | datacenter | 2026-08-12 | Rent → |
| MI300X | 192 GB | $2.39 | – | RunPod secure cloud+3 more | datacenter | 2026-08-12 | Rent → |
| GB300 | 288 GB | $8.62 | – | DataCrunch on-demand+1 more | datacenter | 2026-08-12 | Rent → |
| MI325X | 256 GB | $3.80 | $2.00 | DigitalOcean on-demand+1 more | datacenter | 2026-08-12 | Rent → |
| GH200 | 96 GB | $2.29 | – | Lambda on-demand+1 more | datacenter | 2026-08-12 | Rent → |
| A40 | 48 GB | $0.44 | – | RunPod secure cloud+1 more | datacenter | 2026-08-12 | Rent → |
| MI355X | 288 GB | spot only | $2.59 | Vultr spot / interruptible 8× min | datacenter | 2026-08-12 | Rent → |
| MI350X | 288 GB | $6.16 | – | DigitalOcean on-demand | datacenter | 2026-08-12 | Rent → |
| GB200 | 186 GB | $8.00 | – | GMI Cloud on-demand | datacenter | 2026-08-12 | Rent → |
| A30 | 24 GB | $0.35 | – | Massed Compute on-demand | datacenter | 2026-08-12 | Rent → |
| RTX PRO 6000 | 96 GB | $1.10 | $0.44 | Google Cloud on-demand+13 more | workstation | 2026-08-12 | Rent → |
| RTX A6000 | 48 GB | $0.35 | $0.33 | Thunder Compute on-demand+7 more | workstation | 2026-08-12 | Rent → |
| RTX 6000 Ada | 48 GB | $0.75 | $0.74 | TensorDock on-demand+4 more | workstation | 2026-08-12 | Rent → |
| RTX A4000 | 16 GB | $0.10 | $0.17 | TensorDock on-demand+2 more | workstation | 2026-08-12 | Rent → |
| RTX PRO 4500 | 32 GB | $0.72 | – | RunPod secure cloud+1 more | workstation | 2026-08-12 | Rent → |
| RTX A5000 | 24 GB | $0.27 | $0.16 | RunPod secure cloud+1 more | workstation | 2026-08-12 | Rent → |
| Quadro RTX 6000 | 24 GB | $0.69 | – | Lambda on-demand | workstation | 2026-08-12 | Rent → |
| Quadro RTX 5000 | 16 GB | $0.60 | – | OVHcloud on-demand | workstation | 2026-08-12 | Rent → |
| RTX 4090 | 24 GB | $0.35 | $0.20 | TensorDock on-demand+3 more | consumer | 2026-08-12 | Rent → |
| RTX 5090 | 32 GB | $0.86 | $0.29 | Spheron on-demand+2 more | consumer | 2026-08-12 | Rent → |
| RTX 3090 | 24 GB | $0.20 | $0.12 | TensorDock on-demand+2 more | consumer | 2026-08-12 | Rent → |
| RTX 5080 | 16 GB | spot only | $0.22 | SaladCloud community cloud+1 more | consumer | 2026-08-12 | Rent → |
Tables rank by price, lowest first. Commercial relationships never affect ordering.How we collect and rank prices.
On-demand = fixed-price capacity (secure datacenter tiers included). Spot/community = interruptible marketplace capacity, cheaper but not guaranteed (serverless per-GPU pricing from fal, Modal, Replicate and Fireworks lives on each GPU page and the serverless comparison). Multi-GPU node prices are normalized to per-GPU rates where providers only sell nodes. Monthly view assumes 730 hours of continuous use. The $/GB VRAM view divides each price by that offer's VRAM (a value-per-memory lens, not a rate you can buy). Click a GPU for every provider, source link and history.
Browse by what you actually need
By VRAM
Built different, on purpose
Sourced, not scraped-and-forgotten
Every price carries the source link and the exact time we verified it. A price we could not re-verify is flagged, never silently reused.
History nobody can backfill
Every daily snapshot is kept forever in a public ledger. You can see whether today's deal is actually a deal.
Math with the formula shown
The rent-vs-buy calculator prints its formula. The dataset is downloadable. Check our work.