Live GPU-hour prices, tracked daily
GPU rental prices · Taipei Ledger
Per-GPU-hour rental prices in US dollars across 10 GPU models (B300 288 GB, B200 180 GB, H200 141 GB, H100 SXM and PCIe/NVL, A100 80GB, MI300X, L40S, RTX PRO 6000, RTX 5090, RTX 4090) from AWS on-demand, Azure pay-as-you-go, RunPod Secure Cloud, RunPod Community, and the Vast.ai marketplace (median
Per-GPU tables of data transfer charges (download and upload medians and maximums in US$ per TB, including offers charging US$30+/TB) and extra disk pricing per GB-month, alongside each day's offer counts.
Daily price history since 02 November 2026 for each GPU model, plotted per provider (AWS, Azure, RunPod Secure, RunPod Community, Vast.ai median).
Explains that prices are normalized per GPU-hour (e.g. AWS p5.48xlarge with 8 × H100 divided by 8), that list prices come from AWS US East (N. Virginia) and Azure East US on-demand Linux instances, that committed-use and reserved prices are excluded, and that Vast.ai figures are the median and lowes
Maps each GPU to the specific AWS instance type (e.g. p6-b300.48xlarge, p5en.48xlarge, g6e.xlarge) and Azure VM (e.g. Standard_ND96isr_H100_v5, Standard_NC40ads_H100_v5) behind the listed price.
A list price that moves more than 50% in one day is held at its previous value until verified; sources not updated today are marked with an asterisk.
Prices are scraped from provider list pricing: AWS and Azure on-demand Linux rates (US regions), RunPod Secure Cloud and Community Cloud, and the Vast.ai marketplace, then divided by the number of GPUs in each machine to get a per-GPU-hour figure.
For Vast.ai, the median and lowest per-GPU price of rentable on-demand offers that day are reported, with the number of offers noted because small sample sizes can cause swings.
Tables and charts are refreshed daily (page generated 2026-11-11 15:17, prices dated 11 November 2026), with 10-day price history charts dating back to 02 November 2026.
The GPU price tracker itself is presented as free to use on the site. The parent Taipei Ledger runs a free weekly newsletter, 'The Week in Taiwan Companies', and invites workplace readers to join a team-plan waiting list, but no paid plans or prices for the tracker are stated on the page.
It publishes what one GPU-hour costs to rent, in US dollars, broken down by GPU model (B300, B200, H200, H100, A100, MI300X, L40S, RTX PRO 6000, RTX 5090, RTX 4090) and provider (AWS, Azure, RunPod Secure/Community, Vast.ai marketplace), updated daily.
Clouds rent whole machines, so each machine's hourly price is divided by the number of GPUs in it — for example, the AWS p5.48xlarge with 8 × H100 is divided by 8.
AWS on-demand (US East, N. Virginia), Azure pay-as-you-go (East US), RunPod Secure Cloud (its own data centres) and Community Cloud (third-party hosts), plus the Vast.ai marketplace where many independent hosts worldwide list offers.
No — figures are compute-only. The hourly price includes a small default disk, but more disk is charged per GB-month, and each Vast.ai host sets its own per-GB data transfer charge. AWS and Azure also charge for data sent out to the internet.
No. Committed-use and reserved prices are usually lower but are explicitly not shown; only on-demand/pay-as-you-go list prices are tracked.
Prices are updated daily (the page shown is dated 11 November 2026), with daily price history charts since 02 November 2026. Sources not updated that day are marked, and a list price moving more than 50% in a day is held at its previous value until checked.
The tracker is free to read on the site. Taipei Ledger also offers a free weekly newsletter, 'The Week in Taiwan Companies', with a team-plan waiting list for workplace use.