GPU Economy

GPU Economy reads the public price pages and APIs of 60+ GPU clouds every hour — hyperscalers (AWS, Azure, Google Cloud, Oracle), neoclouds (Lambda, CoreWeave, Nebius, Crusoe, RunPod and others) and marketplaces (Vast.ai, Clore.ai) — and normalizes every listed rate to US dollars per GPU per hour. This server gives an AI agent those prices directly, instead of scraping pages.

Tools (all read-only; no key, no account):

Prices

• list_gpus — every tracked GPU model (H100, H200, B200, B300, A100, L40S, RTX 4090/5090, MI300X and more) with VRAM, how many providers list it and the median on-demand rate.
• gpu_prices — current rates for one GPU across providers, cheapest first; on-demand by default, or spot, community, serverless and reserved; optionally US or EU regions only.
• provider_prices — every current rate at one provider.
• provider_info — one provider: its type, regions, how its prices are read, and its cheapest on-demand rate on each GPU.

Choosing and comparing

• find_gpus — GPU models that meet a memory and price requirement (e.g. 80 GB+, under $3/hour), cheapest first.
• compare_gpus — two GPU models side by side: VRAM, median and cheapest rate, price per GB, and every provider that lists both.
• compare_providers — two providers on every GPU both list, and which is cheaper on each.

Cost and planning

• estimate_cost — what N GPUs cost for a number of hours at listed rates, at the cheapest providers and at the median, with the arithmetic.
• commitment_savings — spot and 1- or 3-year committed rates against the same provider's on-demand rate, with the typical discount per term.
• buy_vs_rent — a card's retail price against its rental rates: months of rental that equal the purchase.
• token_cost — modelled cost per million output tokens from NVIDIA's published throughput; labelled as modelled, cards not comparable on it.

Trends

• price_index — the Cloud GPU Price Index: the median on-demand H100 SXM rate, weekly, hourly and monthly.
• price_history — weekly