hundredthrun/gpu-cloud-prices

GPU Cloud Prices (H100, H200, B200 and more)

Public on-demand, spot and listed GPU rental prices from 25 cloud providers (H100, H200, B200, B300, GB200, GB300, A100, L40S, MI300X, MI355X), normalized to USD per GPU-hour. Every row links to the provider page it came from and records the date it was read.

• Live tables, calculators and provider comparisons: https://hundredthrun-compute-desk.pages.dev
• Open data docs on the desk: https://hundredthrun-compute-desk.pages.dev/open-data
• Source repo: https://github.com/hundredthrun/gpu-cloud-prices
• JSON API: https://hundredthrun-compute-desk.pages.dev/api/v1/prices.json
• Kaggle mirror: https://www.kaggle.com/datasets/hundredthrun/gpu-cloud-rental-prices
• Browse the cheapest price per GPU (Hugging Face Space): https://huggingface.co/spaces/hundredthrun/gpu-cloud-prices
• Starter notebook (pandas, charts, price changes): https://github.com/hundredthrun/gpu-cloud-prices/blob/main/notebooks/gpu-cloud-prices-starter.ipynb

Files

File	What it holds
gpu_prices.csv	One row per provider, GPU and price type, with source_url, date_checked, fetch_method and confidence.
gpu_price_history.csv	Min, median and max price per GPU per daily snapshot (starts 2026-11-04).
gpu_price_rows_daily.csv	Every provider row as it stood on each snapshot date, so you can track individual price changes.
api_prices.csv	Sourced LLM API prices (USD per 1M input and output tokens) for open models, used by the desk's self-host vs API calculator.
lead_times.csv	Availability and lead-time notes as stated by providers.
models.csv	Open model parameter counts used by the desk's GPU memory calculator, with sources.

Load a config with:

from datasets import load_dataset
ds = load_dataset("hundredthrun/gpu-cloud-prices", "gpu_prices")
print(ds["train"][0])

Columns (gpu_prices.csv)

• provider, gpu, variant, gpus_per_unit
• price_usd_per_gpu_hr, price_type (on_demand / spot / preemptible / reserved), term, region, billing_unit
• price_as_listed, derivation, availability_note, lead_time_class
• source_url, date_checked, fetch_method, confidence, notes

Method

Prices are read from each provider's public pricing page or official price API. Multi-GPU node prices are divided by GPU count; the original wording is kept in price_as_listed. Confidence is high for API or itemized prices, medium for "starting at" prices, low otherwise. Unpublished values stay blank. Nothing is estimated.

Limitations

Public list prices only. Negotiated, reserved and committed-use prices are often lower. Check date_checked and the source before you buy.

Updates

Prices are re-checked daily. Dataset versions are pushed on the Tuesday refresh (and on demand). Price history charts: https://hundredthrun-compute-desk.pages.dev/history/

Card last rebuilt: 2026-11-09.

License

CC BY 4.0. Credit "Hundredth Run compute desk" with a link to https://hundredthrun-compute-desk.pages.dev