DATACENTER
B200
Cheapest verified$6.69 / GPU-hour
INDEPENDENT PRICE DESKON-DEMAND · USD / GPU-HOUR● VERIFIED EDITION
What does a GPU cost right now
We open each provider’s pricing page, record the number, and leave the source in plain sight.
The price board.
180 GB VRAM
DATACENTER
H200
Cheapest verified$3.99 / GPU-hour
DATACENTER
H100 SXM
Cheapest verified$3.20 / GPU-hour
DATACENTER
H100 PCIe
Cheapest verified$2.50 / GPU-hour
DATACENTER
A100 80GB
Cheapest verified$1.09 / GPU-hour
DATACENTER
A100 40GB
Cheapest verified$1.99 / GPU-hour
DATACENTER
AMD MI300X
Cheapest verified$2.59 / GPU-hour
DATACENTER
AMD MI325X
Cheapest verified$3.80 / GPU-hour
WORKSTATION
L40S
Cheapest verified$1.09 / GPU-hour
WORKSTATION
L40
Cheapest verified$0.79 / GPU-hour
WORKSTATION
RTX 6000 Ada
Cheapest verified$1.57 / GPU-hour
WORKSTATION
RTX A6000
Cheapest verified$0.35 / GPU-hour
WORKSTATION
A40
Cheapest verified$0.59 / GPU-hour
WORKSTATION
RTX 4000 Ada
Cheapest verified$0.76 / GPU-hour
WORKSTATION
RTX A4000
Cheapest verified$0.15 / GPU-hour
CONSUMER
RTX 5090
Cheapest verified$1.19 / GPU-hour
CONSUMER
RTX 4090
Cheapest verified$0.89 / GPU-hour
PRICES ARE PER GPU, PER HOUR, BEFORE TAXMARKETPLACE LISTINGS VARY LIVE
BULK COMPUTE
OFF THE BOARD
OFF THE BOARD
FOR LONGER RUNS
Need 8–1,000 GPUs for a month or more?
On-demand pricing is the wrong tool for that. Tell us the GPU, count, term and region and we’ll bring back three quotes from vetted providers within one business day. Free to you — the providers pay us.
THE RECEIPT
How we check prices.
Manual, dated verification
Each number on this page was read from the provider’s public pricing page on the date shown. If a card says “not listed”, the provider doesn’t publish that GPU on-demand.
On-demand only
Reserved, spot, and committed-term prices are excluded unless labelled. Marketplaces (Vast.ai) fluctuate by the minute, so we link you to their live board instead of printing a number.
Per GPU, per hour, before tax
Multi-GPU instance prices are divided by GPU count. Storage, egress and taxes are extra everywhere.
Tiers matter
RunPod’s pricing page lists a single pod price per GPU without labelling Secure vs Community tier; Hyperstack lists several H100 variants (we show SXM and the unlabelled “H100” row).
A STARTING POINT, NOT A RANKING
Which provider for what.
| If you… | Start with | Why |
|---|---|---|
| Want a 4090/5090 for a few hours of image or LLM work | RunPod, Vast.ai | Consumer cards listed, pay-as-you-go, large template library |
| Need an A100/H100 at the lowest fixed price | Thunder Compute, Hyperstack | Tied for the lowest verified H100 SXM rate today ($3.20); Thunder has the lowest A100 80GB ($1.09) |
| Want the simplest, most reliable single-box experience | Lambda | Pre-built ML images, 1-click Jupyter; priced at a premium |
| Already run on DigitalOcean / want AMD MI300X | DigitalOcean | Same account and billing as your droplets; MI300X at $2.59 |
| Need B200 / H200 at scale or a 1–12 month term | Get quotes ↗ | Reserved pricing is negotiated and usually sits well below on-demand (DigitalOcean’s published 12-month H100 rate is $3.26 vs $4.41 on-demand; CoreWeave advertises reserved discounts up to 60%) |