GPU cloud prices, November 2026.
Prices checked , updated every 3 days until the month ends
So far in November 2026 the cheapest H100 SXM rents for $2.69/GPU-hr at RunPod, against a typical $4.19/GPU-hr, checked 1 Nov.
The month so far
Typical rate
$2.27/GPU-hr
level with October
Rentals
178
from 10 providers, as many as October
Rates moved
0
0 cheaper · 0 dearer
New providers
0
+0 new rentals from providers already here
Cheapest by GPU
The cheapest on-demand rate for each of the boards the most providers rent out, and the typical rate for that board.
H100 SXM
H100 SXM · Community Cloud
1x H100 SXM · 80 GB · 20 vCPU
RunPod
A typical rate is $4.19/GPU-hr, the middle of 15 rentals. The same cheapest price as in October.
A100 80GB SXM
A100 SXM · Community Cloud
1x A100 80GB SXM · 80 GB · 16 vCPU
RunPod
A typical rate is $2.25/GPU-hr, the middle of 11 rentals. The same cheapest price as in October.
B200
B200 · Community Cloud
1x B200 · 180 GB · 28 vCPU
RunPod
A typical rate is $8.60/GPU-hr, the middle of 19 rentals. The same cheapest price as in October.
L40S
L40S · Community Cloud
1x L40S · 48 GB · 16 vCPU
RunPod
A typical rate is $1.67/GPU-hr, the middle of 12 rentals. The same cheapest price as in October.
H200 SXM
H200 · Community Cloud
1x H200 SXM · 141 GB · 24 vCPU
RunPod
A typical rate is $5.99/GPU-hr, the middle of 9 rentals. The same cheapest price as in October.
B300
B300 · Community Cloud
1x B300 · 288 GB · host size not stated
RunPod
A typical rate is $11.35/GPU-hr, the middle of 12 rentals. The same cheapest price as in October.
How we count
This report is rewritten by every refresh until November 2026 ends, and after that it is never changed. This version was written from the refresh of 1 Nov 2026.
A move is a price that ended the month somewhere other than where it started it, in the provider's own currency, so an exchange rate moving is not a move, and a price that went down and back up again is not one either.
The cheapest of a tier is a price that could lead a ranking on the day: the provider shows it as available and its prices were read in the last three days.
Every rate is the on-demand price for one GPU for one hour, so a node of eight is counted at what one of its GPUs costs, converted to US dollars at the day's ECB rate. Reserved and committed rates are not indexed.
A serverless rate is counted among the rentals and the moves, and is in no typical rate and no cheapest, because it buys a container that runs when called rather than a machine you hold.
The lists name each provider and board once: a provider whose rates follow demand moves every size of a board together, and that is one line. The counts hold every rental.
Prices by month
Where to go next