How GPU Tracker calculates the numbers you see

We aggregate public cloud GPU pricing and listing data using the open-source gpuhunt library plus custom provider-specific scraping and normalization code.

The dataset is published as a public JSON file at /api-docs and served from the Vercel CDN.

Prices and inventory are refreshed on a scheduled pipeline every 6 hours. Provider APIs and marketplace listings may change faster than that, so treat the data as a high-quality comparison snapshot, not a guaranteed quote.

Always confirm the final provider invoice and checkout page before provisioning.

True Monthly Cost is an estimate designed to answer: “What will this workload likely cost over a month?” It starts with compute spend and layers in assumptions such as storage, egress, minimum billing increments, and billed-when-stopped behaviors when applicable.

TMC is intentionally conservative compared with raw $/hr comparisons. The goal is decision quality, not optimistic marketing math.

For exact assumptions, use the in-product TMC panel and adjust workload hours, storage, and egress to match your project.

Availability and interruption risk summarize listing quality and runtime reliability signals, especially for spot / marketplace capacity. These are indicators, not SLAs.

A low price can still be a poor choice if your job is long-running, stateful, or expensive to restart. GPU Tracker highlights this tradeoff so you can compare price against survival probability.