GPUBeacon
Language

Price index

Cloud GPU pricing chart

Published on-demand price per GPU hour for every model we track. Cheapest in-stock row, median across providers and a 720-hour month, from the GPUBeacon catalog sample.

Verified 2026-09-05

GPU price per hour, cheapest first

On-demand rows only. Spot, waitlist and quotes are labeled on each GPU page, not mixed in here.

GPUVRAMCheapest / hrMedian / hr720 h monthProviders
RTX 4090 price per hourCheapest on Vast.ai24 GB$0.29$0.42$2094
RTX 5090 price per hourCheapest on Vast.ai32 GB$0.34$0.52$2454
L4 price per hourCheapest on Google Cloud24 GB$0.39$0.43$2815
A10 price per hourCheapest on RunPod24 GB$0.45$0.75$3244
RTX PRO 6000 price per hourCheapest on Packet.ai96 GB$0.66$0.79$4755
L40S price per hourCheapest on RunPod48 GB$0.79$0.89$5695
A100 price per hourCheapest on Vast.ai80 GB$1.19$1.79$8577
H100 price per hourCheapest on Vast.ai80 GB$1.73$2.99$1,24615
MI300X price per hourCheapest on Nebius192 GB$1.99$2.29$1,4333
H200 price per hourCheapest on Koyeb141 GB$3.00$3.64$2,16010
B200 price per hourCheapest on Packet.ai180 GB$3.75$4.33$2,7006
B300 price per hourCheapest on Verda288 GB$7.50$8.08$5,4004

Browse every GPUGPU hour cost calculatorRent a GPUHow we rank an hour

Same GPU, different clouds

H100 price per hour by provider type

The on-demand H100 runs from $1.73 on Vast.ai to $4.98 on Azure: a 2.9x spread for the same badge.

Provider typeProvidersCheapest H100 / hrMedian H100 / hr
Marketplace1$1.73$1.73
Managed GPU cloud5$2.19$2.69
Specialist GPU cloud5$2.49$3.10
Inference + GPU cloud1$3.49$3.49
Hyperscaler3$4.10$4.85

How to read this GPU pricing chart

The cheapest cell is a floor, not a promise. It is the lowest published on-demand hour in our sample, and it can be a marketplace host, a PCIe card or a region your data cannot enter. The median is closer to what a managed cloud will charge you for the same badge.

Always compare the same SKU: H100 SXM, PCIe and NVL are different products with the same name. Read H100 SXM vs PCIe vs NVL before you paste a card, and on-demand vs spot before you trust a discount.

The monthly column multiplies the cheapest hour by 720. Real bills include storage, egress and idle time. Use the hour cost calculator to price the job you actually plan to run, and compare two cards with the GPU duels.

Is this a price history?

No. This is a dated sample, not a 90-day index. We would rather show you one verified number than a trend line we cannot stand behind. If a row looks wrong, confirm the live rate with the provider and tell us.

FAQ

GPU pricing questions

What is the average H100 price per hour in the cloud?+

In the GPUBeacon catalog sample, the median on-demand H100 is $2.99 per GPU per hour and the cheapest in-stock row is $1.73 on Vast.ai. One H100 for a 720-hour month at the cheapest hour is about $1,246.

Why does the same GPU have such different cloud prices?+

The H100 on-demand hour runs from $1.73 on Vast.ai to $4.98 on Azure in this sample, a 2.9x spread. Marketplaces sell leftover capacity, managed clouds sell a platform, specialist clouds sell clusters and support, and hyperscalers sell the account you already have. Form factor (SXM vs PCIe), interconnect, region and stock explain the rest.

How much does a cloud GPU cost per hour?+

It depends on the model. From $0.29 for RTX 4090 to $7.50 for B300 at the cheapest on-demand row in this sample. For reference, A100 starts at $1.19 and B200 at $3.75.

Does GPUBeacon show GPU price history?+

Not yet. This chart is a dated catalog sample, not a 90-day price index. It shows published on-demand rates the day we verified them. Confirm the live rate with the provider before you provision.

Are spot prices included in the chart?+

No. The chart uses on-demand rows only. Spot, interruptible and waitlist rows are labeled on each GPU page and never ranked as the cheaper hour.