FAQ
How is this different from GetDeploying / ComputePrices / other comparison sites?
Most GPU price sites show today's prices only. GetDeploying has a genuinely good weekly index — but it's weekly, its API costs $299/mo and excludes historical data, and rankings share a page with sponsor placements. Price of Compute is daily, the full history is in the free API, affiliate links never touch rankings or the Index, and there's a weekly editorial newsletter. We're a reference, not a listings business.
Are these prices actually attainable?
They're the prices providers listed at scrape time — capacity at that price isn't guaranteed, especially on marketplaces. Every number carries a timestamp, and we aggregate medians across many observations so one phantom listing can't become "the price." An availability signal is on the roadmap.
Why is the same GPU 3–4× more expensive on one provider than another?
Different reliability tiers, interconnects, regions, and business models. A community-cloud 4090 and an enterprise-SLA 4090 are almost different products — which is why we label pricing types and never blend them.
Do you include egress, storage, and other fees?
No — these numbers are GPU-hour compute prices only. Total cost of a job varies with your data patterns. Treat the Index as the market's pulse, not your invoice.
Spot vs. on-demand — which do you show?
Both, always separately, labeled. Default views are on-demand. Mixing them would make every number meaningless.
Where does the money come from?
Two places, both disclosed: (1) some provider links are affiliate links — always labeled, never affecting prices, rankings, or the Index; (2) ads/sponsorships may appear, clearly marked as such, with no say in content or methodology. The data, charts, and API stay free.
Why should I trust your numbers?
Don’t trust — verify. The methodology is public, every observation is timestamped, raw source fragments are stored for auditability, and the API gives you the same data we use. If you find an error, tell us and we’ll fix it publicly.
Provider X is missing.
Request it. Each provider is an independent adapter; popular requests get built first. (AMD SKUs and TPU tracking are on the roadmap — most-requested items from every prior launch in this space.)
How often does data update?
Scrapes run every hour per provider (marketplace sources every 3 hours); daily aggregates finalize at 00:00 UTC. Every page shows its last-updated timestamp — if something looks stale, it's labeled, not hidden.
Can I use your charts/data in an article, paper, or app?
Yes — that’s the point. Cite “Price of Compute” with a link. The API and CSV downloads are free under the limits on the API page.