# GPU Economy

> Retail purchase prices and cloud rental prices for datacenter and consumer GPUs,
> collected from public provider pricing pages and normalized to a single
> comparable unit: US dollars per GPU per hour.

Prices are re-scraped on a schedule and stored append-only, so every figure has a
capture timestamp and a source URL. Repeated hourly reads older than 14 days move to an
archive file (each quote's first read, every change and the last read of each day stay). Node-priced offerings (CoreWeave, Lambda SXM,
Paperspace) are divided by GPU count so a per-node quote never gets compared
against a per-GPU one. Listings in another currency (EUR, and since 2026-09-28 any: RUB, JPY, INR, ...) show the
listed price next to the USD figure with the official reference rate and date used (ECB;
the Bank of Russia for RUB and VND). Only USD listings enter the Price Index.
Every page shows the date its data was captured.
Every price in every table is in the HTML the server sends, as a plain table, so a
fetch without a browser gets every figure with its source link and read time. First-party
JavaScript enhances controls and, since 2026-11-03, reports browser confirmation and
explicit filter/comparison actions through our own server. No third-party script is
loaded (/methodology#no-javascript; /methodology#readers).

We read published price lists, not availability: a listed rate can be sold out,
waitlisted or absent from the region you need, and whether a card is free to rent
is not something we measure. Cite a price as the rate a provider published on the
date beside it, never as capacity you can book.

## Data

- [Cloud GPU Price Index](https://gpueconomy.com/price-index): weekly median publicly listed on-demand price of one NVIDIA H100 GPU-hour across major providers. Marketplace floors (Vast.ai) and quotes listed in any currency other than USD are excluded. For GPU Economy's own directly collected series, a week is published only when at least 8 providers contributed; this series starts in the week of 2026-09-07, the first week that met that guard (series[0].quotes and series[0].providers in price-index.json say how many). This 8-provider coverage guard applies to our own weekly and hourly index, not to the separate reconstructed SkyPilot series described below. Community-cloud and serverless list prices are included (today RunPod Community Cloud and Modal: a fixed hourly rate for one H100, not a marketplace floor), unlike the card median on card pages, which leaves community capacity out; without the one community quote each week would read about $0.025 higher (markets_included in price-index.json). The median is taken over one quote per provider and region, so a provider that lists several regions enters once per region; weighting_check in price-index.json recomputes every week as the median of per-provider medians, with the largest provider's quote count and an HHI - a check on that rule, not a second index. A quote count counts rate cards, not GPUs available to book. Every published week also carries its middle half (series[].p25_usd_per_gpu_hour and p75_usd_per_gpu_hour, the last two columns of price-index.csv): the 25th and 75th percentile of the same quotes as its median, added on 2026-09-19 to closed weeks too, computed from their receipts - no median changed. A withheld week is a statement about coverage, not a missing number: price-index.json carries current.status and withheld_reason (too few providers, or providers not confirmed within 48 hours, each with its last read), and a week absent from price-index.csv is listed under withheld[] - cite it as "not published for the week of X: N of 8 providers confirmed", never as the previous week's value. Each week also carries a chain-linked index (base week = 100, geometric mean of provider+region week-over-week ratios) that moves only on price changes, not on panel composition. Each week is also cut by operator class (by_operator_class in price-index.json: hyperscaler = AWS, Azure, Google Cloud, Oracle; specialist = GPU clouds with their own price list) - the same median over the same quotes, not a separate index; marketplaces have no value because their floors never enter it. CC BY 4.0. /price-index#discounts also gives the typical discount of spot and 1-year or 3-year committed rates against the same provider's on-demand rate (median across rate pairs, a term only with 3+ providers); the same figures, with every rate pair and its source link, are market_discounts in price-index.json.
- How much reading is behind the figures: price-index.json carries `observations` - `count` (every listed rate ever read), `since` (the first read), `providers`, `gpu_models`, `hourly_runs` and `what_one_is`; the same sentence is on /price-index. One observation is one read of one listed rate, and the same rate read every hour counts every hour; it is not a transaction and not availability, and no published figure is computed from the count. The number is not repeated in this file on purpose: a count typed into text ages silently, so read it from the JSON or the page, both generated.
- [Reproduce the index](https://gpueconomy.com/price-index/reproduce): every published week has a receipt, /price-index/receipts/{YYYY-Www}.csv - one row per quote that entered the value, with its source URL and read time, plus every quote read for that week and kept out of it (included=false, with the rule in excluded_by: marketplace_floor, eur_listing, not_confirmed_48h) - and a one-line Python command that turns the receipt back into the published median. Each week in price-index.json carries its receipt URL.
- The monthly average of the hourly index over a fixed monthly panel, frozen when the month closes: [price-index-monthly.csv](https://gpueconomy.com/price-index-monthly.csv); the panel of each month under https://gpueconomy.com/price-index/panel/
- The same index at every hourly scrape, each point written once with its quotes: [price-index-hourly.csv](https://gpueconomy.com/price-index-hourly.csv), receipts per week under https://gpueconomy.com/price-index/hourly/ ; how it may be relied on: https://gpueconomy.com/price-index/specification
- [Historical H100 series (SkyPilot catalog)](https://gpueconomy.com/gpu/h100-sxm/history#reconstructed-index): a deliberate, second-hand reconstruction from [skypilot-org/skypilot-catalog](https://github.com/skypilot-org/skypilot-catalog). The series starts in the week of 2025-01-06 and is published separately as [price-index-history.csv](https://gpueconomy.com/price-index-history.csv) (`source=skypilot-catalog`) and `historical_series` in [price-index.json](https://gpueconomy.com/price-index.json). For each ISO week, we take the catalog's state at the end of that week, select the cheapest per-GPU on-demand H100 SXM instance per provider, and publish the median when at least 4 providers contributed. Weeks with 4-7 providers therefore meet this series' own coverage guard. We vouch that the catalog published these prices on that date, not that we checked the provider pages at the time; exact source commits are linked on the history page. It uses a different provider panel and is never merged with our own series or used to fill its withheld weeks.
- [Public dataset](https://gpueconomy.com/data): every file with its fields, update cadence and license (CC BY 4.0) - prices.json (also as prices.csv, same fields), price-index.json/.csv, price-index-history.csv, history.json (per-model weekly series from both sources), changelog.json (also as changelog.csv: every observed price move, newest first - 30 days for every provider, the last 24 hours for the Vast.ai floor, which moves hourly), providers.json, gpus.json.
- MCP server (read-only, no key): `https://gpueconomy.com/mcp`, Streamable HTTP. Tools: list_gpus, gpu_prices, provider_prices, price_index, recent_price_changes, find_gpus (cards meeting a memory and price requirement), compare_gpus, compare_providers, estimate_cost (GPUs x hours at listed rates), price_history, commitment_savings (spot and 1-/3-year rates against the same provider's on-demand), provider_info, token_cost (modelled cost per million output tokens), buy_vs_rent, search and fetch (pages as text, for ChatGPT connectors) - region filter US/EU on the price tools, four prompts, the published files as resources - every price with its read time (scraped_at) and the GPU Economy page that links its source. How to connect: https://gpueconomy.com/mcp
- Slices of the price file, same fields, fewer rows: `/gpu/{gpu_id}/prices.json` (one card across providers) and `/providers/{provider_id}/prices.json` (one provider across cards). Ids match the page URLs.
- Machine-readable: [price-index.json](https://gpueconomy.com/price-index.json) (current value, constituents with source links, weekly series), [price-index.csv](https://gpueconomy.com/price-index.csv) (the series), and [prices.json](https://gpueconomy.com/prices.json) (every current rental quote with provider, GPU, price type, region, the instance or SKU it was priced for, normalized and listed price, and source URL).
- [Monthly market reviews](https://gpueconomy.com/market): /market/{YYYY-MM} - generated from the data for every closed month: base-100 H100 index at month end and month over month, dollar median with its panel, largest fall and rise with provider and week, cheapest listed H100/H200/B200; each figure linked to its source.
- [Embed widgets](https://gpueconomy.com/embed): the same figures for another site's page, as static iframes (no script): a card's five cheapest rates, the H100 Price Index, two GPUs side by side, one provider's cards - each rate linked to its source, CC BY 4.0 with an attribution link.
- [Price changelog](https://gpueconomy.com/changelog): every observed price change, newest first. As data: /changelog.json and /changelog.csv; providers' own rate changes (marketplace floors left out) as an Atom feed: /changelog.xml.

## Pages

- [Methodology](https://gpueconomy.com/methodology): how prices are collected and normalized, what is never done (estimated prices, silent currency conversion) and the one modelled figure, cost per million tokens, and every source with its status. Every rate's risk tag (low / medium / high) is read off its market - whether the capacity can be taken back mid-run - not off the provider; it is the risk field in prices.json (/methodology#risk).
- [Methodology changelog](https://gpueconomy.com/methodology/changelog): every dated change to those rules, with the version the files carry. Every figure we publish travels with `methodology_version` (top level in price-index.json and prices.json, a column in every CSV and in each weekly receipt), so a cited number can be read against the rules that produced it. patch = collection and coverage, minor = new fields or cuts, major = a change of definition, which never happens under the same series name.

- [Home](https://gpueconomy.com/): the NVIDIA H100 SXM at a glance - market median (the Price Index), cheapest listed rate, price spread, hyperscaler premium and 12-month change; every listed on-demand rate with its source and capture time; what a month of rental costs for 1-8 GPUs; weekly history; buy vs rent; and a cards x providers spread matrix. The same view exists per card at /gpu/{model}.
- Consumer cards also carry NVIDIA's published starting price and the retail median as a share of it (e.g. RTX 4090 list $1,599); datacenter cards have no published list price, so no such share is shown.
- [Buy vs rent](https://gpueconomy.com/buy-vs-rent): per GPU, the median retail price across sellers - inside one kind of price: stores' own prices where three or more stores carry the card, marketplace sellers' asking prices otherwise, never averaged together, and never a completed sale - against the cheapest rental rate and the median rate card (each provider once, at its cheapest on-demand rate, community capacity and marketplace floors out), and the months of continuous rental that equal the purchase price (rental fee only; power, hosting and resale not counted). Break-even is also given at 100, 70, 50 and 25% utilisation (730, 511, 365 and 183 hours a month) for every card with a median rate card, and on each card page.
- [Reserved GPU pricing](https://gpueconomy.com/reserved): every live 1-year, 3-year and committed rate providers publish, card by card, each against the same provider's on-demand rate and linked to its source. The page also takes requests for reserved quotes, which GPU Economy passes to up to three providers listing a rate for the card, cheapest first; a provider may pay a fee if one becomes a contract, disclosed there. Requests and fees change no rate, table order or index value.
- [Fees beyond the GPU-hour](https://gpueconomy.com/fees): what each provider bills besides the hourly rate - whether the rate includes vCPU, RAM and local disk, data out (egress), persistent storage, what a stopped instance keeps billing, and the billing step - read by hand from the provider's own pages, each answer linked to its page with the day it was checked. "Not published" means the pages were read and do not say. Not added up and not part of any rate or the index; the same facts are on each provider page (#fees), in providers.json (fees) and in the MCP tool provider_info.
- [Price spread by GPU](https://gpueconomy.com/spread): for every card listed by 3+ providers, the cheapest, median and dearest listed on-demand rate, the cheapest-to-dearest ratio and the hyperscaler premium - the dispersion behind the index. From four providers it also gives the middle half of rate cards (25th to 75th percentile of the same rate cards the median is taken over), steadier than the cheapest-to-dearest ratio; the same band is on each card page and in gpus.json (rate_card_p25/p75_usd_per_gpu_hour).
- [Cheapest by provider type](https://gpueconomy.com/cheapest): for every GPU, the lowest listed on-demand rate from hyperscalers and specialist GPU clouds, and beside them the cheapest community capacity or marketplace floor (RunPod Community Cloud, the Vast.ai floor), which never counts as on-demand, each named and linked - the data answer to "cheapest cloud GPU for X". Also the cheapest on-demand card with at least 24, 48, 80 and 141 GB of memory, with provider and capture time, and the memory step where the cheapest rate rises most.
- [GPU pages](https://gpueconomy.com/): per model - retail purchase price alongside the full rental table. Every reserved and spot rate shows its difference in dollars and per cent from the same provider's on-demand rate for the same card, region and scope; where providers price the card as a whole 8-GPU node, a dated sentence gives the cheapest and dearest node per hour.
- [GPU price history](https://gpueconomy.com/): /gpu/{model}/history - weekly median and cheapest listed on-demand rate since 2025 (history reconstructed from a public price catalog, then our own scrape), a like-for-like first-vs-last table per provider and region, static chart; every figure linked to its source.
- [Provider pages](https://gpueconomy.com/): per provider - full price list and served regions.
- [Providers](https://gpueconomy.com/providers): every tracked cloud GPU provider - who runs the capacity, how many GPUs carry a live rate, the cheapest on-demand rate with its source link; each name opens /providers/{id}.
- [Provider comparisons](https://gpueconomy.com/compare): /compare/{a}-vs-{b} - two providers side by side on the GPUs both list, cheaper one per GPU.
- [GPU comparisons](https://gpueconomy.com/gpus): /gpus/{a}-vs-{b} - two GPUs side by side: specs, median rate card and cheapest rental rate, providers listing both.

## Every GPU page

One page per card, /gpu/{id}: every provider's listed rate for that card with its
source link and read time, the retail purchase price where sellers carry it, and the
break-even against renting. The same rows as data: /gpu/{id}/prices.json.

- [NVIDIA A10](https://gpueconomy.com/gpu/a10): Ampere, 24 GB, PCIe - history https://gpueconomy.com/gpu/a10/history; data https://gpueconomy.com/gpu/a10/prices.json
- [NVIDIA A100 40GB](https://gpueconomy.com/gpu/a100-40): Ampere, 40 GB, SXM - history https://gpueconomy.com/gpu/a100-40/history; data https://gpueconomy.com/gpu/a100-40/prices.json
- [NVIDIA A100 80GB](https://gpueconomy.com/gpu/a100-80): Ampere, 80 GB, SXM - history https://gpueconomy.com/gpu/a100-80/history; data https://gpueconomy.com/gpu/a100-80/prices.json
- [NVIDIA A10G](https://gpueconomy.com/gpu/a10g): Ampere, 24 GB, PCIe - history https://gpueconomy.com/gpu/a10g/history; data https://gpueconomy.com/gpu/a10g/prices.json
- [NVIDIA A2](https://gpueconomy.com/gpu/a2): Ampere, 16 GB, PCIe - data https://gpueconomy.com/gpu/a2/prices.json
- [NVIDIA A30](https://gpueconomy.com/gpu/a30): Ampere, 24 GB, PCIe - history https://gpueconomy.com/gpu/a30/history; data https://gpueconomy.com/gpu/a30/prices.json
- [NVIDIA A40](https://gpueconomy.com/gpu/a40): Ampere, 48 GB, PCIe - history https://gpueconomy.com/gpu/a40/history; data https://gpueconomy.com/gpu/a40/prices.json
- [NVIDIA B200](https://gpueconomy.com/gpu/b200): Blackwell, 192 GB, SXM - history https://gpueconomy.com/gpu/b200/history; data https://gpueconomy.com/gpu/b200/prices.json
- [NVIDIA B300](https://gpueconomy.com/gpu/b300): Blackwell, 288 GB, SXM - history https://gpueconomy.com/gpu/b300/history; data https://gpueconomy.com/gpu/b300/prices.json
- [Intel Gaudi 2](https://gpueconomy.com/gpu/gaudi-2): Gaudi, 96 GB, OAM - data https://gpueconomy.com/gpu/gaudi-2/prices.json
- [NVIDIA GB10 (DGX Spark)](https://gpueconomy.com/gpu/gb10): Blackwell, 128 GB, Superchip - data https://gpueconomy.com/gpu/gb10/prices.json
- [NVIDIA GB200 NVL72](https://gpueconomy.com/gpu/gb200): Blackwell, 186 GB, NVL72 - history https://gpueconomy.com/gpu/gb200/history; data https://gpueconomy.com/gpu/gb200/prices.json
- [NVIDIA GB300 NVL72](https://gpueconomy.com/gpu/gb300): Blackwell, 288 GB, NVL72 - history https://gpueconomy.com/gpu/gb300/history; data https://gpueconomy.com/gpu/gb300/prices.json
- [NVIDIA GH200](https://gpueconomy.com/gpu/gh200): Hopper, 96 GB, Superchip - history https://gpueconomy.com/gpu/gh200/history; data https://gpueconomy.com/gpu/gh200/prices.json
- [NVIDIA GeForce GTX 1070 Ti](https://gpueconomy.com/gpu/gtx-1070-ti): Pascal, 8 GB, PCIe - data https://gpueconomy.com/gpu/gtx-1070-ti/prices.json
- [NVIDIA GeForce GTX 1080](https://gpueconomy.com/gpu/gtx-1080): Pascal, 8 GB, PCIe - data https://gpueconomy.com/gpu/gtx-1080/prices.json
- [NVIDIA GeForce GTX 1080 Ti](https://gpueconomy.com/gpu/gtx-1080-ti): Pascal, 11 GB, PCIe - data https://gpueconomy.com/gpu/gtx-1080-ti/prices.json
- [NVIDIA GeForce GTX 1660 SUPER](https://gpueconomy.com/gpu/gtx-1660-super): Turing, 6 GB, PCIe - data https://gpueconomy.com/gpu/gtx-1660-super/prices.json
- [NVIDIA GeForce GTX 1660 Ti](https://gpueconomy.com/gpu/gtx-1660-ti): Turing, 6 GB, PCIe - data https://gpueconomy.com/gpu/gtx-1660-ti/prices.json
- [NVIDIA H100 80GB (form factor not stated)](https://gpueconomy.com/gpu/h100): Hopper, 80 GB - history https://gpueconomy.com/gpu/h100/history; data https://gpueconomy.com/gpu/h100/prices.json
- [NVIDIA H100 NVL](https://gpueconomy.com/gpu/h100-nvl): Hopper, 94 GB, PCIe - history https://gpueconomy.com/gpu/h100-nvl/history; data https://gpueconomy.com/gpu/h100-nvl/prices.json
- [NVIDIA H100 PCIe](https://gpueconomy.com/gpu/h100-pcie): Hopper, 80 GB, PCIe - history https://gpueconomy.com/gpu/h100-pcie/history; data https://gpueconomy.com/gpu/h100-pcie/prices.json
- [NVIDIA H100 SXM](https://gpueconomy.com/gpu/h100-sxm): Hopper, 80 GB, SXM - history https://gpueconomy.com/gpu/h100-sxm/history; data https://gpueconomy.com/gpu/h100-sxm/prices.json
- [NVIDIA H200](https://gpueconomy.com/gpu/h200): Hopper, 141 GB, SXM - history https://gpueconomy.com/gpu/h200/history; data https://gpueconomy.com/gpu/h200/prices.json
- [NVIDIA H200 NVL](https://gpueconomy.com/gpu/h200-nvl): Hopper, 141 GB, PCIe - history https://gpueconomy.com/gpu/h200-nvl/history; data https://gpueconomy.com/gpu/h200-nvl/prices.json
- [NVIDIA L4](https://gpueconomy.com/gpu/l4): Ada, 24 GB, PCIe - history https://gpueconomy.com/gpu/l4/history; data https://gpueconomy.com/gpu/l4/prices.json
- [NVIDIA L40](https://gpueconomy.com/gpu/l40): Ada, 48 GB, PCIe - history https://gpueconomy.com/gpu/l40/history; data https://gpueconomy.com/gpu/l40/prices.json
- [NVIDIA L40S](https://gpueconomy.com/gpu/l40s): Ada, 48 GB, PCIe - history https://gpueconomy.com/gpu/l40s/history; data https://gpueconomy.com/gpu/l40s/prices.json
- [AMD Instinct MI300X](https://gpueconomy.com/gpu/mi300x): CDNA 3, 192 GB, OAM - history https://gpueconomy.com/gpu/mi300x/history; data https://gpueconomy.com/gpu/mi300x/prices.json
- [AMD Instinct MI325X](https://gpueconomy.com/gpu/mi325x): CDNA 3, 256 GB, OAM - history https://gpueconomy.com/gpu/mi325x/history; data https://gpueconomy.com/gpu/mi325x/prices.json
- [AMD Instinct MI350X](https://gpueconomy.com/gpu/mi350x): CDNA 4, 288 GB, OAM - history https://gpueconomy.com/gpu/mi350x/history; data https://gpueconomy.com/gpu/mi350x/prices.json
- [AMD Instinct MI355X](https://gpueconomy.com/gpu/mi355x): CDNA 4, 288 GB, OAM - history https://gpueconomy.com/gpu/mi355x/history; data https://gpueconomy.com/gpu/mi355x/prices.json
- [NVIDIA Tesla P100 12GB](https://gpueconomy.com/gpu/p100-12): Pascal, 12 GB, PCIe - data https://gpueconomy.com/gpu/p100-12/prices.json
- [NVIDIA Quadro P4000](https://gpueconomy.com/gpu/quadro-p4000): Pascal, 8 GB, PCIe - data https://gpueconomy.com/gpu/quadro-p4000/prices.json
- [NVIDIA Quadro RTX 4000](https://gpueconomy.com/gpu/quadro-rtx-4000): Turing, 8 GB, PCIe - data https://gpueconomy.com/gpu/quadro-rtx-4000/prices.json
- [NVIDIA Quadro RTX 6000](https://gpueconomy.com/gpu/quadro-rtx-6000): Turing, 24 GB, PCIe - history https://gpueconomy.com/gpu/quadro-rtx-6000/history; data https://gpueconomy.com/gpu/quadro-rtx-6000/prices.json
- [NVIDIA RTX 2000 Ada](https://gpueconomy.com/gpu/rtx-2000-ada): Ada Lovelace, 16 GB, PCIe - data https://gpueconomy.com/gpu/rtx-2000-ada/prices.json
- [NVIDIA GeForce RTX 2060 12GB](https://gpueconomy.com/gpu/rtx-2060-12): Turing, 12 GB, PCIe - data https://gpueconomy.com/gpu/rtx-2060-12/prices.json
- [NVIDIA GeForce RTX 2080 Ti](https://gpueconomy.com/gpu/rtx-2080-ti): Turing, 11 GB, PCIe - data https://gpueconomy.com/gpu/rtx-2080-ti/prices.json
- [NVIDIA GeForce RTX 3060](https://gpueconomy.com/gpu/rtx-3060): Ampere, 12 GB, PCIe - data https://gpueconomy.com/gpu/rtx-3060/prices.json
- [NVIDIA GeForce RTX 3060 Ti](https://gpueconomy.com/gpu/rtx-3060-ti): Ampere, 8 GB, PCIe - data https://gpueconomy.com/gpu/rtx-3060-ti/prices.json
- [NVIDIA GeForce RTX 3070](https://gpueconomy.com/gpu/rtx-3070): Ampere, 8 GB, PCIe - history https://gpueconomy.com/gpu/rtx-3070/history; data https://gpueconomy.com/gpu/rtx-3070/prices.json
- [NVIDIA GeForce RTX 3070 Ti](https://gpueconomy.com/gpu/rtx-3070-ti): Ampere, 8 GB, PCIe - data https://gpueconomy.com/gpu/rtx-3070-ti/prices.json
- [NVIDIA GeForce RTX 3080](https://gpueconomy.com/gpu/rtx-3080): Ampere, 10 GB, PCIe - history https://gpueconomy.com/gpu/rtx-3080/history; data https://gpueconomy.com/gpu/rtx-3080/prices.json
- [NVIDIA GeForce RTX 3080 Ti](https://gpueconomy.com/gpu/rtx-3080-ti): Ampere, 12 GB, PCIe - history https://gpueconomy.com/gpu/rtx-3080-ti/history; data https://gpueconomy.com/gpu/rtx-3080-ti/prices.json
- [NVIDIA GeForce RTX 3090](https://gpueconomy.com/gpu/rtx-3090): Ampere, 24 GB, PCIe - history https://gpueconomy.com/gpu/rtx-3090/history; data https://gpueconomy.com/gpu/rtx-3090/prices.json
- [NVIDIA GeForce RTX 3090 Ti](https://gpueconomy.com/gpu/rtx-3090-ti): Ampere, 24 GB, PCIe - history https://gpueconomy.com/gpu/rtx-3090-ti/history; data https://gpueconomy.com/gpu/rtx-3090-ti/prices.json
- [NVIDIA RTX 4000 Ada](https://gpueconomy.com/gpu/rtx-4000-ada): Ada Lovelace, 20 GB, PCIe - history https://gpueconomy.com/gpu/rtx-4000-ada/history; data https://gpueconomy.com/gpu/rtx-4000-ada/prices.json
- [NVIDIA RTX 4000 SFF Ada](https://gpueconomy.com/gpu/rtx-4000-sff-ada): Ada Lovelace, 20 GB, PCIe - history https://gpueconomy.com/gpu/rtx-4000-sff-ada/history; data https://gpueconomy.com/gpu/rtx-4000-sff-ada/prices.json
- [NVIDIA GeForce RTX 4060](https://gpueconomy.com/gpu/rtx-4060): Ada Lovelace, 8 GB, PCIe - data https://gpueconomy.com/gpu/rtx-4060/prices.json
- [NVIDIA GeForce RTX 4060 Ti](https://gpueconomy.com/gpu/rtx-4060-ti): Ada Lovelace, 16 GB, PCIe - data https://gpueconomy.com/gpu/rtx-4060-ti/prices.json
- [NVIDIA GeForce RTX 4070](https://gpueconomy.com/gpu/rtx-4070): Ada Lovelace, 12 GB, PCIe - data https://gpueconomy.com/gpu/rtx-4070/prices.json
- [NVIDIA GeForce RTX 4070 SUPER](https://gpueconomy.com/gpu/rtx-4070-super): Ada Lovelace, 12 GB, PCIe - data https://gpueconomy.com/gpu/rtx-4070-super/prices.json
- [NVIDIA GeForce RTX 4070 Ti](https://gpueconomy.com/gpu/rtx-4070-ti): Ada Lovelace, 12 GB, PCIe - data https://gpueconomy.com/gpu/rtx-4070-ti/prices.json
- [NVIDIA GeForce RTX 4070 Ti SUPER](https://gpueconomy.com/gpu/rtx-4070-ti-super): Ada Lovelace, 16 GB, PCIe - data https://gpueconomy.com/gpu/rtx-4070-ti-super/prices.json
- [NVIDIA GeForce RTX 4080](https://gpueconomy.com/gpu/rtx-4080): Ada Lovelace, 16 GB, PCIe - history https://gpueconomy.com/gpu/rtx-4080/history; data https://gpueconomy.com/gpu/rtx-4080/prices.json
- [NVIDIA GeForce RTX 4080 SUPER](https://gpueconomy.com/gpu/rtx-4080-super): Ada Lovelace, 16 GB, PCIe - history https://gpueconomy.com/gpu/rtx-4080-super/history; data https://gpueconomy.com/gpu/rtx-4080-super/prices.json
- [NVIDIA RTX 4090](https://gpueconomy.com/gpu/rtx-4090): Ada, 24 GB - history https://gpueconomy.com/gpu/rtx-4090/history; data https://gpueconomy.com/gpu/rtx-4090/prices.json
- [NVIDIA RTX 4500 Ada Generation](https://gpueconomy.com/gpu/rtx-4500-ada): Ada Lovelace, 24 GB, PCIe - data https://gpueconomy.com/gpu/rtx-4500-ada/prices.json
- [NVIDIA RTX 5000 Ada Generation](https://gpueconomy.com/gpu/rtx-5000-ada): Ada Lovelace, 32 GB, PCIe - history https://gpueconomy.com/gpu/rtx-5000-ada/history; data https://gpueconomy.com/gpu/rtx-5000-ada/prices.json
- [NVIDIA GeForce RTX 5060](https://gpueconomy.com/gpu/rtx-5060): Blackwell, 8 GB, PCIe - data https://gpueconomy.com/gpu/rtx-5060/prices.json
- [NVIDIA GeForce RTX 5060 Ti](https://gpueconomy.com/gpu/rtx-5060-ti): Blackwell, 16 GB, PCIe - data https://gpueconomy.com/gpu/rtx-5060-ti/prices.json
- [NVIDIA GeForce RTX 5070](https://gpueconomy.com/gpu/rtx-5070): Blackwell, 12 GB, PCIe - data https://gpueconomy.com/gpu/rtx-5070/prices.json
- [NVIDIA GeForce RTX 5070 Ti](https://gpueconomy.com/gpu/rtx-5070-ti): Blackwell, 16 GB, PCIe - data https://gpueconomy.com/gpu/rtx-5070-ti/prices.json
- [NVIDIA GeForce RTX 5080](https://gpueconomy.com/gpu/rtx-5080): Blackwell, 16 GB, PCIe - history https://gpueconomy.com/gpu/rtx-5080/history; data https://gpueconomy.com/gpu/rtx-5080/prices.json
- [NVIDIA RTX 5090](https://gpueconomy.com/gpu/rtx-5090): Blackwell, 32 GB - history https://gpueconomy.com/gpu/rtx-5090/history; data https://gpueconomy.com/gpu/rtx-5090/prices.json
- [NVIDIA RTX 5880 Ada](https://gpueconomy.com/gpu/rtx-5880-ada): Ada, 48 GB, PCIe - data https://gpueconomy.com/gpu/rtx-5880-ada/prices.json
- [NVIDIA RTX 6000 Ada](https://gpueconomy.com/gpu/rtx-6000-ada): Ada, 48 GB, PCIe - history https://gpueconomy.com/gpu/rtx-6000-ada/history; data https://gpueconomy.com/gpu/rtx-6000-ada/prices.json
- [NVIDIA RTX A2000](https://gpueconomy.com/gpu/rtx-a2000): Ampere, 6 GB, PCIe - history https://gpueconomy.com/gpu/rtx-a2000/history; data https://gpueconomy.com/gpu/rtx-a2000/prices.json
- [NVIDIA RTX A4000](https://gpueconomy.com/gpu/rtx-a4000): Ampere, 16 GB, PCIe - history https://gpueconomy.com/gpu/rtx-a4000/history; data https://gpueconomy.com/gpu/rtx-a4000/prices.json
- [NVIDIA RTX A5000](https://gpueconomy.com/gpu/rtx-a5000): Ampere, 24 GB, PCIe - history https://gpueconomy.com/gpu/rtx-a5000/history; data https://gpueconomy.com/gpu/rtx-a5000/prices.json
- [NVIDIA RTX A6000](https://gpueconomy.com/gpu/rtx-a6000): Ampere, 48 GB, PCIe - history https://gpueconomy.com/gpu/rtx-a6000/history; data https://gpueconomy.com/gpu/rtx-a6000/prices.json
- [NVIDIA RTX PRO 4500 Blackwell](https://gpueconomy.com/gpu/rtx-pro-4500): Blackwell, 32 GB, PCIe - history https://gpueconomy.com/gpu/rtx-pro-4500/history; data https://gpueconomy.com/gpu/rtx-pro-4500/prices.json
- [NVIDIA RTX PRO 5000 Blackwell](https://gpueconomy.com/gpu/rtx-pro-5000): Blackwell, 48 GB, PCIe - history https://gpueconomy.com/gpu/rtx-pro-5000/history; data https://gpueconomy.com/gpu/rtx-pro-5000/prices.json
- [NVIDIA RTX PRO 6000 Blackwell](https://gpueconomy.com/gpu/rtx-pro-6000): Blackwell, 96 GB, PCIe - history https://gpueconomy.com/gpu/rtx-pro-6000/history; data https://gpueconomy.com/gpu/rtx-pro-6000/prices.json
- [NVIDIA RTX PRO 6000 Blackwell Max-Q](https://gpueconomy.com/gpu/rtx-pro-6000-max-q): Blackwell, 96 GB, PCIe - history https://gpueconomy.com/gpu/rtx-pro-6000-max-q/history; data https://gpueconomy.com/gpu/rtx-pro-6000-max-q/prices.json
- [NVIDIA T4](https://gpueconomy.com/gpu/t4): Turing, 16 GB, PCIe - history https://gpueconomy.com/gpu/t4/history; data https://gpueconomy.com/gpu/t4/prices.json
- [NVIDIA TITAN RTX](https://gpueconomy.com/gpu/titan-rtx): Turing, 24 GB, PCIe - data https://gpueconomy.com/gpu/titan-rtx/prices.json
- [NVIDIA TITAN V](https://gpueconomy.com/gpu/titan-v): Volta, 12 GB, PCIe - data https://gpueconomy.com/gpu/titan-v/prices.json
- [NVIDIA TITAN Xp](https://gpueconomy.com/gpu/titan-xp): Pascal, 12 GB, PCIe - data https://gpueconomy.com/gpu/titan-xp/prices.json
- [NVIDIA V100](https://gpueconomy.com/gpu/v100): Volta, 16 GB, SXM - history https://gpueconomy.com/gpu/v100/history; data https://gpueconomy.com/gpu/v100/prices.json
- [NVIDIA V100 32GB](https://gpueconomy.com/gpu/v100-32): Volta, 32 GB - data https://gpueconomy.com/gpu/v100-32/prices.json

## Every provider page

One page per provider, /providers/{id}: its full price list, the regions it serves and
how its prices are read. The class is the operator class the Price Index is also cut by.
A provider we considered and do not read keeps a page that says why.

- [airgpu](https://gpueconomy.com/providers/airgpu): specialist - data https://gpueconomy.com/providers/airgpu/prices.json
- [Airon](https://gpueconomy.com/providers/airon): specialist - data https://gpueconomy.com/providers/airon/prices.json
- [Atlantic.Net](https://gpueconomy.com/providers/atlantic): specialist - data https://gpueconomy.com/providers/atlantic/prices.json
- [Atlas Cloud](https://gpueconomy.com/providers/atlascloud): specialist - data https://gpueconomy.com/providers/atlascloud/prices.json
- [Amazon Web Services](https://gpueconomy.com/providers/aws): hyperscaler - data https://gpueconomy.com/providers/aws/prices.json
- [Microsoft Azure](https://gpueconomy.com/providers/azure): hyperscaler - data https://gpueconomy.com/providers/azure/prices.json
- [Bamboozle Web Services](https://gpueconomy.com/providers/bamboozle): specialist - data https://gpueconomy.com/providers/bamboozle/prices.json
- [Barrack AI](https://gpueconomy.com/providers/barrack): specialist - data https://gpueconomy.com/providers/barrack/prices.json
- [Baseten](https://gpueconomy.com/providers/baseten): specialist - data https://gpueconomy.com/providers/baseten/prices.json
- [Beam](https://gpueconomy.com/providers/beam): specialist - data https://gpueconomy.com/providers/beam/prices.json
- [Bentaus](https://gpueconomy.com/providers/bentaus): specialist - data https://gpueconomy.com/providers/bentaus/prices.json
- [Bitdeer AI](https://gpueconomy.com/providers/bitdeer): specialist - data https://gpueconomy.com/providers/bitdeer/prices.json
- [Canopy Wave](https://gpueconomy.com/providers/canopywave): specialist - data https://gpueconomy.com/providers/canopywave/prices.json
- [Cato Digital](https://gpueconomy.com/providers/cato): specialist - data https://gpueconomy.com/providers/cato/prices.json
- [centron](https://gpueconomy.com/providers/centron): specialist - data https://gpueconomy.com/providers/centron/prices.json
- [Cerebrium](https://gpueconomy.com/providers/cerebrium): specialist - data https://gpueconomy.com/providers/cerebrium/prices.json
- [Civo](https://gpueconomy.com/providers/civo): specialist - data https://gpueconomy.com/providers/civo/prices.json
- [Clore.ai](https://gpueconomy.com/providers/clore): marketplace - data https://gpueconomy.com/providers/clore/prices.json
- [CloudGPU](https://gpueconomy.com/providers/cloudgpu): specialist - data https://gpueconomy.com/providers/cloudgpu/prices.json
- [CloudPe](https://gpueconomy.com/providers/cloudpe): specialist - no rate of ours; why: https://gpueconomy.com/add-a-provider#not-listed
- [CloudRift](https://gpueconomy.com/providers/cloudrift): specialist - data https://gpueconomy.com/providers/cloudrift/prices.json
- [Cloud.ru Evolution](https://gpueconomy.com/providers/cloudru): specialist - data https://gpueconomy.com/providers/cloudru/prices.json
- [CoreWeave](https://gpueconomy.com/providers/coreweave): specialist - data https://gpueconomy.com/providers/coreweave/prices.json
- [Crusoe](https://gpueconomy.com/providers/crusoe): specialist - data https://gpueconomy.com/providers/crusoe/prices.json
- [Cudo Compute](https://gpueconomy.com/providers/cudo): specialist - data https://gpueconomy.com/providers/cudo/prices.json
- [Cyfuture AI](https://gpueconomy.com/providers/cyfuture): specialist - no rate of ours; why: https://gpueconomy.com/add-a-provider#not-listed
- [Verda](https://gpueconomy.com/providers/datacrunch): specialist - data https://gpueconomy.com/providers/datacrunch/prices.json
- [Dataoorts](https://gpueconomy.com/providers/dataoorts): specialist - data https://gpueconomy.com/providers/dataoorts/prices.json
- [Daytona](https://gpueconomy.com/providers/daytona): specialist - data https://gpueconomy.com/providers/daytona/prices.json
- [Deep Infra](https://gpueconomy.com/providers/deepinfra): specialist - data https://gpueconomy.com/providers/deepinfra/prices.json
- [Denvr Dataworks](https://gpueconomy.com/providers/denvr): specialist - data https://gpueconomy.com/providers/denvr/prices.json
- [E2E Networks](https://gpueconomy.com/providers/e2e): specialist - data https://gpueconomy.com/providers/e2e/prices.json
- [EcoHash](https://gpueconomy.com/providers/ecohash): specialist - data https://gpueconomy.com/providers/ecohash/prices.json
- [Edgevana](https://gpueconomy.com/providers/edgevana): specialist - data https://gpueconomy.com/providers/edgevana/prices.json
- [Elice Cloud](https://gpueconomy.com/providers/elice): specialist - data https://gpueconomy.com/providers/elice/prices.json
- [EmpirioLabs](https://gpueconomy.com/providers/empiriolabs): specialist - data https://gpueconomy.com/providers/empiriolabs/prices.json
- [Enverge](https://gpueconomy.com/providers/enverge): specialist - data https://gpueconomy.com/providers/enverge/prices.json
- [fal](https://gpueconomy.com/providers/fal): specialist - data https://gpueconomy.com/providers/fal/prices.json
- [FarmGPU](https://gpueconomy.com/providers/farmgpu): specialist - data https://gpueconomy.com/providers/farmgpu/prices.json
- [Fireworks AI](https://gpueconomy.com/providers/fireworks): specialist - data https://gpueconomy.com/providers/fireworks/prices.json
- [FlexAI](https://gpueconomy.com/providers/flexai): specialist - data https://gpueconomy.com/providers/flexai/prices.json
- [FluidCore](https://gpueconomy.com/providers/fluidcore): specialist - data https://gpueconomy.com/providers/fluidcore/prices.json
- [FPT AI Factory](https://gpueconomy.com/providers/fpt): specialist - data https://gpueconomy.com/providers/fpt/prices.json
- [Gcore](https://gpueconomy.com/providers/gcore): specialist - data https://gpueconomy.com/providers/gcore/prices.json
- [Google Cloud](https://gpueconomy.com/providers/gcp): hyperscaler - data https://gpueconomy.com/providers/gcp/prices.json
- [Geodd](https://gpueconomy.com/providers/geodd): specialist - data https://gpueconomy.com/providers/geodd/prices.json
- [GMI Cloud](https://gpueconomy.com/providers/gmi-cloud): specialist - data https://gpueconomy.com/providers/gmi-cloud/prices.json
- [GPU.ai](https://gpueconomy.com/providers/gpuai): specialist - data https://gpueconomy.com/providers/gpuai/prices.json
- [GPUwerk](https://gpueconomy.com/providers/gpuwerk): specialist - data https://gpueconomy.com/providers/gpuwerk/prices.json
- [Helios Cloud](https://gpueconomy.com/providers/helios): specialist - data https://gpueconomy.com/providers/helios/prices.json
- [Hetzner](https://gpueconomy.com/providers/hetzner): specialist - data https://gpueconomy.com/providers/hetzner/prices.json
- [HexGrid](https://gpueconomy.com/providers/hexgrid): specialist - data https://gpueconomy.com/providers/hexgrid/prices.json
- [Hikube](https://gpueconomy.com/providers/hikube): specialist - data https://gpueconomy.com/providers/hikube/prices.json
- [Hivenet](https://gpueconomy.com/providers/hivenet): specialist - data https://gpueconomy.com/providers/hivenet/prices.json
- [Hostinger](https://gpueconomy.com/providers/hostinger): specialist - data https://gpueconomy.com/providers/hostinger/prices.json
- [Hot Aisle](https://gpueconomy.com/providers/hotaisle): specialist - data https://gpueconomy.com/providers/hotaisle/prices.json
- [HPC-AI.COM](https://gpueconomy.com/providers/hpcai): specialist - data https://gpueconomy.com/providers/hpcai/prices.json
- [Hugging Face](https://gpueconomy.com/providers/huggingface): specialist - data https://gpueconomy.com/providers/huggingface/prices.json
- [Hyperbolic](https://gpueconomy.com/providers/hyperbolic): specialist - data https://gpueconomy.com/providers/hyperbolic/prices.json
- [Hyperstack](https://gpueconomy.com/providers/hyperstack): specialist - data https://gpueconomy.com/providers/hyperstack/prices.json
- [immers.cloud](https://gpueconomy.com/providers/immers): specialist - data https://gpueconomy.com/providers/immers/prices.json
- [Inferless](https://gpueconomy.com/providers/inferless): specialist - data https://gpueconomy.com/providers/inferless/prices.json
- [Infomaniak](https://gpueconomy.com/providers/infomaniak): specialist - data https://gpueconomy.com/providers/infomaniak/prices.json
- [Intelion Cloud](https://gpueconomy.com/providers/intelion): specialist - data https://gpueconomy.com/providers/intelion/prices.json
- [IONOS Cloud](https://gpueconomy.com/providers/ionos): specialist - data https://gpueconomy.com/providers/ionos/prices.json
- [ionstream](https://gpueconomy.com/providers/ionstream): specialist - no rate of ours; why: https://gpueconomy.com/add-a-provider#not-listed
- [Jarvislabs](https://gpueconomy.com/providers/jarvislabs): specialist - data https://gpueconomy.com/providers/jarvislabs/prices.json
- [Koyeb](https://gpueconomy.com/providers/koyeb): specialist - data https://gpueconomy.com/providers/koyeb/prices.json
- [Krutrim Cloud](https://gpueconomy.com/providers/krutrim): specialist - data https://gpueconomy.com/providers/krutrim/prices.json
- [Lambda](https://gpueconomy.com/providers/lambda): specialist - data https://gpueconomy.com/providers/lambda/prices.json
- [Latitude.sh](https://gpueconomy.com/providers/latitude): specialist - no rate of ours; why: https://gpueconomy.com/add-a-provider#not-listed
- [Leafcloud](https://gpueconomy.com/providers/leafcloud): specialist - data https://gpueconomy.com/providers/leafcloud/prices.json
- [Lightning AI](https://gpueconomy.com/providers/lightning): specialist - data https://gpueconomy.com/providers/lightning/prices.json
- [Limestone Networks](https://gpueconomy.com/providers/limestone): specialist - data https://gpueconomy.com/providers/limestone/prices.json
- [Akamai Cloud (Linode)](https://gpueconomy.com/providers/linode): specialist - data https://gpueconomy.com/providers/linode/prices.json
- [Lium](https://gpueconomy.com/providers/lium): marketplace - data https://gpueconomy.com/providers/lium/prices.json
- [Lyceum](https://gpueconomy.com/providers/lyceum): specialist - data https://gpueconomy.com/providers/lyceum/prices.json
- [Massed Compute](https://gpueconomy.com/providers/massed): specialist - data https://gpueconomy.com/providers/massed/prices.json
- [Meshive](https://gpueconomy.com/providers/meshive): specialist - data https://gpueconomy.com/providers/meshive/prices.json
- [MimicPC](https://gpueconomy.com/providers/mimicpc): specialist - data https://gpueconomy.com/providers/mimicpc/prices.json
- [Modal](https://gpueconomy.com/providers/modal): specialist - data https://gpueconomy.com/providers/modal/prices.json
- [MWS Cloud Platform](https://gpueconomy.com/providers/mws): specialist - data https://gpueconomy.com/providers/mws/prices.json
- [Nebius](https://gpueconomy.com/providers/nebius): specialist - data https://gpueconomy.com/providers/nebius/prices.json
- [Nebula Block](https://gpueconomy.com/providers/nebulablock): specialist - data https://gpueconomy.com/providers/nebulablock/prices.json
- [Neysa](https://gpueconomy.com/providers/neysa): specialist - data https://gpueconomy.com/providers/neysa/prices.json
- [Northflank](https://gpueconomy.com/providers/northflank): specialist - data https://gpueconomy.com/providers/northflank/prices.json
- [Nova Cloud](https://gpueconomy.com/providers/novacloud): specialist - data https://gpueconomy.com/providers/novacloud/prices.json
- [Octa.space](https://gpueconomy.com/providers/octa): marketplace - data https://gpueconomy.com/providers/octa/prices.json
- [OpenRelay](https://gpueconomy.com/providers/openrelay): specialist - data https://gpueconomy.com/providers/openrelay/prices.json
- [Oracle Cloud (OCI)](https://gpueconomy.com/providers/oracle): hyperscaler - data https://gpueconomy.com/providers/oracle/prices.json
- [OVHcloud](https://gpueconomy.com/providers/ovhcloud): specialist - data https://gpueconomy.com/providers/ovhcloud/prices.json
- [packet.ai](https://gpueconomy.com/providers/packet-ai): specialist - data https://gpueconomy.com/providers/packet-ai/prices.json
- [Paperspace](https://gpueconomy.com/providers/paperspace): specialist - data https://gpueconomy.com/providers/paperspace/prices.json
- [Phala Cloud](https://gpueconomy.com/providers/phala): specialist - data https://gpueconomy.com/providers/phala/prices.json
- [QuantaCloud](https://gpueconomy.com/providers/quantacloud): specialist - data https://gpueconomy.com/providers/quantacloud/prices.json
- [Replicate](https://gpueconomy.com/providers/replicate): specialist - data https://gpueconomy.com/providers/replicate/prices.json
- [RunComfy](https://gpueconomy.com/providers/runcomfy): specialist - data https://gpueconomy.com/providers/runcomfy/prices.json
- [Runcrate](https://gpueconomy.com/providers/runcrate): specialist - data https://gpueconomy.com/providers/runcrate/prices.json
- [RunPod](https://gpueconomy.com/providers/runpod): specialist - data https://gpueconomy.com/providers/runpod/prices.json
- [Runware](https://gpueconomy.com/providers/runware): specialist - data https://gpueconomy.com/providers/runware/prices.json
- [Salad](https://gpueconomy.com/providers/salad): specialist - data https://gpueconomy.com/providers/salad/prices.json
- [Scaleway](https://gpueconomy.com/providers/scaleway): specialist - data https://gpueconomy.com/providers/scaleway/prices.json
- [Seeweb](https://gpueconomy.com/providers/seeweb): specialist - data https://gpueconomy.com/providers/seeweb/prices.json
- [Selectel](https://gpueconomy.com/providers/selectel): specialist - data https://gpueconomy.com/providers/selectel/prices.json
- [Servercore](https://gpueconomy.com/providers/servercore): specialist - data https://gpueconomy.com/providers/servercore/prices.json
- [Sesterce](https://gpueconomy.com/providers/sesterce): specialist - data https://gpueconomy.com/providers/sesterce/prices.json
- [STACKIT](https://gpueconomy.com/providers/stackit): specialist - data https://gpueconomy.com/providers/stackit/prices.json
- [STN](https://gpueconomy.com/providers/stn): specialist - data https://gpueconomy.com/providers/stn/prices.json
- [SwissGPU](https://gpueconomy.com/providers/swissgpu): specialist - data https://gpueconomy.com/providers/swissgpu/prices.json
- [SynpixCloud](https://gpueconomy.com/providers/synpix): specialist - data https://gpueconomy.com/providers/synpix/prices.json
- [Taiga Cloud (Northern Data Group)](https://gpueconomy.com/providers/taiga): specialist - data https://gpueconomy.com/providers/taiga/prices.json
- [Tata Communications](https://gpueconomy.com/providers/tata): specialist - data https://gpueconomy.com/providers/tata/prices.json
- [TensorCloud](https://gpueconomy.com/providers/tensorcloud): specialist - data https://gpueconomy.com/providers/tensorcloud/prices.json
- [TensorDock](https://gpueconomy.com/providers/tensordock): marketplace - no rate of ours; why: https://gpueconomy.com/add-a-provider#not-listed
- [TensorPool](https://gpueconomy.com/providers/tensorpool): specialist - data https://gpueconomy.com/providers/tensorpool/prices.json
- [TensorWave](https://gpueconomy.com/providers/tensorwave): specialist - data https://gpueconomy.com/providers/tensorwave/prices.json
- [TheAI Cloud](https://gpueconomy.com/providers/theai): specialist - data https://gpueconomy.com/providers/theai/prices.json
- [Thunder Compute](https://gpueconomy.com/providers/thunder): specialist - data https://gpueconomy.com/providers/thunder/prices.json
- [Together AI](https://gpueconomy.com/providers/together): specialist - data https://gpueconomy.com/providers/together/prices.json
- [Trooper.AI](https://gpueconomy.com/providers/trooper): specialist - data https://gpueconomy.com/providers/trooper/prices.json
- [Vagon](https://gpueconomy.com/providers/vagon): specialist - data https://gpueconomy.com/providers/vagon/prices.json
- [Vast.ai](https://gpueconomy.com/providers/vast-ai): marketplace - data https://gpueconomy.com/providers/vast-ai/prices.json
- [VESSL AI](https://gpueconomy.com/providers/vessl): specialist - data https://gpueconomy.com/providers/vessl/prices.json
- [Voltage Park](https://gpueconomy.com/providers/voltage-park): specialist - no rate of ours; why: https://gpueconomy.com/add-a-provider#not-listed
- [Vultr](https://gpueconomy.com/providers/vultr): specialist - data https://gpueconomy.com/providers/vultr/prices.json
- [Yandex Cloud](https://gpueconomy.com/providers/yandex): specialist - data https://gpueconomy.com/providers/yandex/prices.json
- [Yotta Labs](https://gpueconomy.com/providers/yottalabs): specialist - data https://gpueconomy.com/providers/yottalabs/prices.json
- [YTL AI Cloud](https://gpueconomy.com/providers/ytl): specialist - data https://gpueconomy.com/providers/ytl/prices.json

## Every GPU comparison

Two cards side by side, /gpus/{a}-vs-{b}: specs, median rate card, cheapest rental rate
and the providers listing both. These pairs are chosen, not generated.

- [NVIDIA A100 40GB vs NVIDIA A100 80GB](https://gpueconomy.com/gpus/a100-40-vs-a100-80)
- [NVIDIA A100 80GB vs NVIDIA H100 SXM](https://gpueconomy.com/gpus/a100-80-vs-h100-sxm)
- [NVIDIA B200 vs NVIDIA B300](https://gpueconomy.com/gpus/b200-vs-b300)
- [NVIDIA GB200 NVL72 vs NVIDIA B200](https://gpueconomy.com/gpus/gb200-vs-b200)
- [NVIDIA H100 SXM vs NVIDIA B200](https://gpueconomy.com/gpus/h100-sxm-vs-b200)
- [NVIDIA H100 SXM vs NVIDIA H100 PCIe](https://gpueconomy.com/gpus/h100-sxm-vs-h100-pcie)
- [NVIDIA H100 SXM vs NVIDIA H200](https://gpueconomy.com/gpus/h100-sxm-vs-h200)
- [NVIDIA H100 SXM vs AMD Instinct MI300X](https://gpueconomy.com/gpus/h100-sxm-vs-mi300x)
- [NVIDIA H200 vs NVIDIA B200](https://gpueconomy.com/gpus/h200-vs-b200)
- [NVIDIA H200 vs AMD Instinct MI300X](https://gpueconomy.com/gpus/h200-vs-mi300x)
- [NVIDIA L40S vs NVIDIA RTX A6000](https://gpueconomy.com/gpus/l40s-vs-rtx-a6000)
- [NVIDIA RTX 4090 vs NVIDIA L40S](https://gpueconomy.com/gpus/rtx-4090-vs-l40s)
- [NVIDIA RTX 4090 vs NVIDIA RTX PRO 6000 Blackwell](https://gpueconomy.com/gpus/rtx-4090-vs-rtx-pro-6000)
- [NVIDIA RTX PRO 6000 Blackwell vs NVIDIA L40S](https://gpueconomy.com/gpus/rtx-pro-6000-vs-l40s)

Provider-against-provider comparisons, /compare/{a}-vs-{b}, are generated for every
pair that quoted two or more of the same cards on-demand within 7 days - too many to
list here without burying the pages above. All of them, and every other URL on this
site, are in the sitemap: https://gpueconomy.com/sitemap-index.xml (https://gpueconomy.com/sitemap-0.xml).

## Fetching

Every file is static: no key, no rate limit. Send a User-Agent header - any value. The hosting platform (Cloudflare Pages) refuses Python's default urllib agent (`Python-urllib/3.x`) with HTTP 403 "error code: 1010"; curl, requests, Node and Go clients and crawlers with their own agent are served normally. Example: `urllib.request.Request('https://gpueconomy.com/prices.json', headers={'User-Agent': 'my-analysis/1.0'})`.

## Citing

Cite as: GPU Economy, "<page title>", <date>. <URL> - the date is the page's
"Data updated" date, or `data_updated_at` in a JSON file. Example: GPU Economy,
"Cloud GPU Price Index", September 18, 2026. https://gpueconomy.com/price-index. /data,
/spread, /buy-vs-rent and each /market/{YYYY-MM} review end with a "Cite this
page" block giving this line, a Markdown and an APA form, filled in. A hint, not a condition: the data is CC BY 4.0.

Why the date belongs in the citation: every rate here is re-read hourly and a
quote not re-read within 48 hours stops counting (unconfirmed, then stopped), so
an undated figure can describe a price that no longer exists. For the Price
Index it matters more: a week can be withheld, and a withheld week is cited as
"not published for the week of X: N of 8 providers confirmed", never as the
previous week's value. Every figure also carries `methodology_version`; add it
when a figure is quoted for comparison over time.

Attribute as "GPU Economy" and link the specific page. Each price carries the
source URL it was read from; cite that source for the price itself, and cite us
for the normalization and the cross-provider comparison.

The Cloud GPU Price Index is also archived as a citable dataset: Zenodo DOI
10.5281/zenodo.22706541 (https://doi.org/10.5281/zenodo.22706541, resolves to the
latest dated snapshot) and Hugging Face gpueconomy/cloud-gpu-price-index. Cite
https://gpueconomy.com/price-index for the current week; cite the DOI with its
version for a figure frozen as it was used.

## Link policy

Every rental rate is read first hand, from the company that bills for the GPU-hour - its own pricing page or API - never from an aggregator or a republished rate card; a marketplace floor is read from the marketplace, which bills for the host's price. Retail purchase prices are the stated exception: resellers' own listings and Google Shopping, each figure naming its store. Source links are direct: every price links to the page or API it was read from, stored as the source serves it, with no tracking or affiliate parameters and no redirect of ours. Since September 19, 2026 the separate "Deploy" button beside a rate goes through /go/ on this site, which counts the click (time and quote only, no personal data, bots not counted) and redirects to the provider: since October 1, 2026 to the provider's referral link where GPU Economy has one, otherwise to the same source page; the price link itself stays direct. Seven providers publish prices through an API (AWS, Azure, Cudo, Google Cloud, Oracle, RunPod, Vast.ai): their rows link the provider's price page and say "read from the provider's API", with the endpoint where it can be opened without a key of your own - Google's Cloud Billing Catalog cannot, so those rows point at Google's page instead. A figure read from an API is often more exact than the page beside it, which may round it or default to another region. Referral links (providers.json, referral_url) are never used for price source links. No referral arrangement changes which prices a table shows, which providers it includes or their order, and neither inclusion in the Price Index nor a position in any table is for sale. Changes to this policy are dated at https://gpueconomy.com/methodology#link-policy.

## Who publishes this

GPU Economy sells, resells and hosts no GPU capacity, brokers none, and takes no commission on what a reader rents from a listed provider. It issues, prices and settles no contract, option, swap or other instrument against the Price Index or any other figure on this site, and holds no position in any market it measures. No provider in the Price Index, or in any table here, is owned or operated by the publisher or shares ownership with it; if that ever changes, that provider's row is marked in the table where its price appears, not only on the methodology page. A referral link, if a programme is ever joined, does not change any of this: it moves no price, no table and no index membership. Section published September 15, 2026; changes to it are dated at https://gpueconomy.com/methodology#publisher.


Providers are listed free when they publish a public rate card, and no listing, position or index inclusion can be bought. Suggest a provider, or report a wrong price with the provider's page that shows the right one, at https://gpueconomy.com/add-a-provider (email data@gpueconomy.com). A price sent by email is never published: every price links to the provider's own page. Providers, price sources and single quotes we considered and do not list, each with the reason and the day checked, are at https://gpueconomy.com/add-a-provider#not-listed.

## Limits

- Listed public prices only. Negotiated, committed-use and enterprise contract rates are not visible to us and are not included.
- Other H100 indices print different numbers for the same week, mostly for three reasons: posted rate cards (this index, every quote linked) against executed transaction prices (for example Ornn's OCPI), whether the four hyperscalers are in the panel, and the statistic (a median here, a trimmed or volume-weighted mean elsewhere). Compare the methods before the numbers: https://gpueconomy.com/methodology#other-indices. Two of them are becoming futures settlement prices: CME lists Silicon Data H100 and B200 Rental Index Futures on NYMEX from October 5, 2026 (pending regulatory review), and ICE announced futures on Ornn's OCPI on May 19, 2026 (subject to approval, no listing date as of September 19, 2026). The GPU Economy index is a median of public rate cards with a written specification (https://gpueconomy.com/price-index/specification); settling an instrument on it is free under CC BY 4.0, and a settlement licence adds binding terms, signed values and custom indices (https://gpueconomy.com/data#settlement): https://gpueconomy.com/methodology#index-futures
- Availability is not tracked: a listed price does not mean capacity is free right now.
- Spot and preemptible rates move faster than our scrape interval; treat them as indicative.
- A quote not re-read within 48 hours of the newest scrape is marked unconfirmed beside its price (after 7 days: stopped), shows when it was last confirmed, and is left out of every computed figure, the Price Index included. prices.json carries `status` and `last_confirmed_at` on every row. A Vast.ai floor withheld as too thin at the newest read is marked thin market instead (status thin_market, with withheld_at and withheld_reason): it keeps its last published floor and is equally left out of every figure. Each row also carries its change over 7 days (listed_7d_ago, read_7d_ago_at, change_7d_pct: against the same quote's read 7 days earlier, null until a week of reads exists).
- Every live price also says since when that exact figure has been listed ("unchanged since {date}"; `listed_since` in prices.json), from our own hourly reads only. "Unchanged since first read" means it has not moved since we started reading it - the real age may be longer.