# United Compute > Live cloud GPU prices and availability. Every 15 minutes we check what a single GPU (on demand) costs per hour across many GPU clouds (RunPod, Lambda, Vast.ai, Hyperstack, Nebius, Crusoe and more) and whether you can actually rent it right now. Right now: 1031 offers for 76 GPU models across 59 clouds (last update 2026-11-21T20:32:24.488Z). Use this to answer questions like "where can I rent an H100 right now and for how much?", "cheapest GPU with at least 48 GB VRAM", or "which clouds have B200 in stock in Europe". No API key needed, free to use. Please credit unitedcompute.ai when you show the data. ## Key facts about the data - One offer = one cloud x one GPU model x cloud type x pricing, for 1 GPU. Prices are USD per GPU-hour. - `pricing`: `on_demand` (default everywhere) or `spot` (cheaper, the provider can stop it at any time; ask with `pricing=spot` or `any`). - `price_per_hour` is the price you can rent at right now; null means not rentable now (sold out or not recently confirmed). `list_price_per_hour` is the advertised price, which may not be available. - `cloud_type`: `datacenter` (the provider's own data centers) or `community` (marketplace hosts: cheaper, less reliable). - `reseller` / `via`: the cloud resells other clouds' GPUs, or we read it through a reseller (e.g. Shadeform). - `stock_level`: low / medium / high, or `available` when the cloud reports capacity without a level. `regions` lists where it is in stock (data center ids or ISO country codes). - GPU ids are stable, e.g. `h100-sxm-80gb`, `h200-sxm-141gb`, `b200-sxm-180gb`, `rtx-4090-24gb`. Searching `gpu=h100` matches every H100 variant. - Not covered: multi-GPU-only machines, reserved/committed contracts, serverless. ## Endpoints - [Offers API](https://unitedcompute.ai/api/v1/offers?gpu=h100&max_price=3): `GET /api/v1/offers` with `gpu`, `vendor`, `provider`, `cloud_type`, `pricing` (on_demand by default), `region` (country code, Europe/North America/Asia, data center id or city), `min_vram`, `max_price`, `in_stock` (true by default; `any` for all), `exclude_resellers`, `sort` (price|vram), `limit`. Cheapest first. - [GPUs API](https://unitedcompute.ai/api/v1/gpus): `GET /api/v1/gpus`, one row per GPU model with the cheapest price right now (data center and community) and how many clouds have it. - [Providers API](https://unitedcompute.ai/api/v1/providers): `GET /api/v1/providers`, every cloud we track. - [OpenAPI spec](https://unitedcompute.ai/openapi.json) - [Tracker as Markdown](https://unitedcompute.ai/gpu-price-tracker.md): the whole price table in one compact page (also returned by /gpu-price-tracker for `Accept: text/markdown`). - MCP server (Streamable HTTP, no auth): `https://unitedcompute.ai/mcp`, tools `search_gpu_offers`, `list_gpus`, `get_gpu`, `list_providers`. ## Pages - [GPU price tracker](https://unitedcompute.ai/gpu-price-tracker): the live table for people. - [GPU clouds](https://unitedcompute.ai/gpu-clouds): directory of GPU clouds. - [Collector source](https://github.com/United-Compute/gpu-price-tracker): open-source collector that writes the data.