What it costs to rent an H100, H200, B200 or B300 in October 2026
On-demand GPU prices from Lambda, Runpod and CoreWeave, checked Oct 22, 2026, with the math for what a real training or inference job costs.
GPU cloud price tracker: H100, H200, B200 and B300 rental prices in one CSV
On-demand and spot list prices from Lambda, Runpod and CoreWeave, checked Oct 22, 2026, plus a script that prices a whole training or inference job.
Our rental price article is a snapshot. This is the same data as a file you can sort, filter and update, plus a script that does the job math.
Preview: gpu-prices.csv
provider,offer,gpu,gpu_memory_gb,pricing,billing,instance_gpus,price_usd_per_hour,...
Lambda,8x instance,H100 SXM,80,on-demand,per GPU,8,3.99,...
Runpod,Secure Cloud,H100 SXM,80,on-demand,per GPU,,3.99,...
Runpod,Community Cloud,H100 SXM,80,on-demand,per GPU,,2.69,...
CoreWeave,HGX node,H100 SXM,80,on-demand,per 8-GPU node,8,49.24,...
Runpod,Secure Cloud,B300,288,on-demand,per GPU,,8.99,...
Running python3 cost.py --gpu "H100 SXM" --gpus 8 --hours 10 lists Runpod Community Cloud at $215.20, Lambda and Runpod Secure Cloud at $319.20, and a CoreWeave node at $492.40 (list prices, Oct 22, 2026).
What's inside
gpu-prices.csv: 24 offers covering H100 SXM, H100 PCIe, H200, B200, B300 and RTX 5090, with source and date checkedstorage-prices.csv: Runpod network storage and CoreWeave hot object storage, per GB per monthcost.py: the job-cost calculator. Python 3.8+, no dependenciestests/: tests, including the worked examples from the articleREADME.md: what each column means and how to keep it current
How to use it
- Unzip and run
python3 cost.py --listto see every offer. - Price a job:
python3 cost.py --gpu B200 --gpus 8 --hours 10. - Add
--storage-gb 2000to include a month of storage, or--spotto include interruptible offers. - Before a big run, check the provider's page and update the row and its
checkeddate.
Who it's for
Builders deciding where to fine-tune or serve a model, and anyone budgeting a GPU project.
Related
FAQ
Why do CoreWeave rows cost so much more?
CoreWeave prices whole 8-GPU nodes, so the row is the node price. Divide by 8 to compare per GPU. Nodes also bundle different CPUs, memory and networking than other providers' instances.
Are spot prices included?
Two CoreWeave North America spot prices are in the CSV. The script leaves spot out unless you pass --spot, because spot capacity can be taken back mid-job.
Why isn't there a CoreWeave on-demand B300 price?
On Oct 22, 2026, CoreWeave listed on-demand HGX B300 as 'contact sales', so there was no public price to record.
How do I add another provider?
Add a row with the same columns, including the source URL and the date you checked it. The script picks it up automatically.
Keep reading
Keep reading
Newsletter
The Codema publication. Consent is required. You are not subscribed until you confirm from the email, and nothing is stored if email delivery is not connected.