Add a GPU cloud
GPU Economy lists any GPU cloud that publishes a public rate card. A listing is free and cannot be bought: providers are found by us or suggested by anyone, and every price on the site links to the provider’s own page.
As of , we read the prices of 63 providers for 53 GPU models: Akamai Cloud (Linode), Amazon Web Services, Atlantic.Net, Atlas Cloud, Baseten, Beam, Bentaus, Canopy Wave, Cerebrium, Civo, Clore.ai, Cloud.ru Evolution, CloudRift, CoreWeave, Crusoe, Cudo Compute, Daytona, Deep Infra, Denvr Dataworks, E2E Networks, EcoHash, EmpirioLabs, Enverge, fal, Gcore, GMI Cloud, Google Cloud, HexGrid, Hot Aisle, Hugging Face Inference Endpoints, Hyperbolic, Hyperstack, immers.cloud, Intelion Cloud, Jarvislabs, Koyeb, Lambda, Leafcloud, Lyceum, Massed Compute, Microsoft Azure, Modal, Nebius, Neysa, Northflank, Nova Cloud, Oracle Cloud (OCI), OVHcloud, packet.ai, Paperspace, Replicate, RunPod, Scaleway, Seeweb, Selectel, Sesterce, Taiga Cloud (Northern Data Group), TensorWave, Thunder Compute, Together AI, Vast.ai, Verda, Vultr.
What a provider needs to be listed
- A public pricing page, reachable without an account, a login or a sales call. A price we cannot link to is not published, whoever sends it.
- An hourly price per GPU, or per instance with its GPU count. We divide a node price by its GPU count ourselves and say so next to the figure.
- A stated price. “Contact sales” gives us nothing to link to. An advertised “from” price is shown tagged as such and kept out of the Price Index, and we may wait for a full rate card before listing a provider that publishes only that.
- Prices readable without a browser: in the page’s HTML or a public API, JSON being best. A table drawn only by JavaScript we cannot read, and we do not run headless browsers.
- A GPU we track: AMD Instinct MI300X, AMD Instinct MI325X, AMD Instinct MI350X, AMD Instinct MI355X, NVIDIA A10, NVIDIA A100 40GB, NVIDIA A100 80GB, NVIDIA A10G, NVIDIA A30, NVIDIA A40, NVIDIA B200, NVIDIA B300, NVIDIA GB200 NVL72, NVIDIA GB300 NVL72, NVIDIA GeForce RTX 3060, NVIDIA GeForce RTX 3070, NVIDIA GeForce RTX 3080, NVIDIA GeForce RTX 3080 Ti, NVIDIA GeForce RTX 3090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 4070 Ti, NVIDIA GeForce RTX 4080, NVIDIA GeForce RTX 4080 SUPER, NVIDIA GeForce RTX 5070 Ti, NVIDIA GeForce RTX 5080, NVIDIA GH200, NVIDIA H100 80GB (form factor not stated), NVIDIA H100 NVL, NVIDIA H100 PCIe, NVIDIA H100 SXM, NVIDIA H200, NVIDIA H200 NVL, NVIDIA L4, NVIDIA L40, NVIDIA L40S, NVIDIA Quadro RTX 6000, NVIDIA RTX 4000 Ada, NVIDIA RTX 4000 SFF Ada, NVIDIA RTX 4090, NVIDIA RTX 5000 Ada Generation, NVIDIA RTX 5090, NVIDIA RTX 6000 Ada, NVIDIA RTX A2000, NVIDIA RTX A4000, NVIDIA RTX A5000, NVIDIA RTX A6000, NVIDIA RTX PRO 4500 Blackwell, NVIDIA RTX PRO 5000 Blackwell, NVIDIA RTX PRO 6000 Blackwell, NVIDIA RTX PRO 6000 Blackwell Max-Q, NVIDIA T4, NVIDIA V100, NVIDIA V100 32GB. Other models are recorded as unmapped and added deliberately, never guessed.
- Any currency. A price listed in another currency is shown as listed, next to its USD conversion at the dated official rate: the ECB’s, or the Bank of Russia’s for RUB and VND, which the ECB does not publish. Only USD listings enter the Price Index.
What we never do
- Sell a listing, a position in a table or inclusion in the Price Index. Price tables are sorted by price.
- Publish a price sent to us by email, a quote, or a negotiated rate.
- Let a referral programme change which prices a table shows or in what order — see the link policy.
Considered and not listed
Every GPU cloud, price source and single quote we looked at and do not list, with the reason and the day we last checked, ordered by reason and then by name. No price is shown here: a figure we do not list is not published anywhere on this site. Where the reason is that we cannot read the prices, the limit is ours, not the provider’s.
| Provider or source | Reason | What we saw | Checked | Look again |
|---|---|---|---|---|
| Alta Technologies retail source | No public rate to link to | A reseller that sells datacenter GPUs by quote only, with no public price to link to. | — | |
| Cherry Servers | No public rate to link to (HTTP 200) | Its GPU offer lives at /dedicated-gpu-servers (found in sitemap-2.xml). The hourly figures printed there are for servers without a GPU: each card reads "GPU Optional up to 1 GPU" beside a rate such as 0.318 USD/hr, while the GPU itself - A2, A16, A100, T4, Quadro K2200/K4200 are named - is "Configure Server" or "Chat with an Expert" with no rate of its own. | ||
| Elastx | No public rate to link to (HTTP 200) | A Swedish OpenStack and Kubernetes operator. Its pricing page carries no figure at all - it describes hourly billing and asks the reader to get in touch for a price. | ||
| FluidStack | No public rate to link to (HTTP 200) | Its sitemap lists only the home page, jobs, news and legal pages - no priced path exists to probe - and the home page it serves carries no rate, no "per hour" and no pricing section of any kind. Capacity is arranged by talking to the company. | ||
| GPUaaS | No public rate to link to (HTTP 200) | Its cluster pages name the cards it offers (B300, B200, H200, H100, A100, RTX 5090, RTX 4090, RTX 6000) and print no rate for any of them; the pages lead to a checkout rather than a published price. | ||
| IBM Cloud | No public rate to link to (HTTP 200) | Rendered GPU product and pricing pages name A100, L40S, MI300X and B300 with no published hourly rate. | ||
| IREN (IREN Cloud) | No public rate to link to (HTTP 200) | iren.com/solutions/gpu-cloud/ai-cloud answers 200 and lists five cards - H100, H200, B200, B300 and GB300 NVL72 - each with its specs and the words "Contact sales" where a rate would be; the page offers "exclusive pricing" when you "deal directly with IREN". Its sitemap carries no pricing path at all. Nothing to read until a rate is printed. | ||
| Nebul | No public rate to link to (HTTP 200) | Its GPU page names B300, L40S and L4 and prints no price, rendered as a visitor's browser draws it (314 KB of rendered HTML, not one currency figure); the page offers "Talk to us" instead. The served HTML had already shown no figure on 2026-09-27. | ||
| Nscale | No public rate to link to (HTTP 200) | Its sitemap has no pricing page at all; the priced paths are per-card pages (/gpu-nodes/nvidia-h100, -h200, -gb200, -gb300-nvl72), and all four are served as static HTML with no figure on them - the call to action is "Reserve GPUs" and "Get in touch". The page even carries a schema.org Offer whose priceSpecification names the currency (USD) and the unit (1 GPU node) and leaves the price out. | ||
| Sakura Internet | No public rate to link to (HTTP 200) | Its GPU line-up prices services, not named cards: the container product DOK at "0.016 yen/second (57.6 yen/hour) and up" and the VM product VRT at "481 yen/hour and up", tax included, each for a service whose card is one of V100 or H100; the bare-metal product PHY is "please contact us". Floors for a service are not a rate for a card. Re-probed 2026-09-28; yen is no longer a reason. | ||
| SF Compute | No public rate to link to (HTTP 200) | The pricing path redirects to a login; the home page is a spot and reserved market with no rate card. | ||
| Timeweb Cloud | No public rate to link to (HTTP 200) | Its GPU page names the cards it rents (H200, H100, A100, A5000, A4000, A2000, A30, A2, T4) with benchmark scores, and prints no hourly rate for any of them in the served HTML: the only rouble figures are a minimum advance payment and grant sums. Re-probed 2026-09-28, when the currency stopped being a reason. | ||
| Voltage Park | No public rate to link to | The pricing page gives only a “starts at” figure for H100s, and its on-demand block reads “Contact for pricing”; the rate card itself sits in the logged-in dashboard. | ||
| WhiteFiber | No public rate to link to (HTTP 200) | Its real priced path (/gpu-pricing, found in its own sitemap after a guessed /pricing returned 404) is static HTML carrying exactly one sentence about price - "Contact us to discuss pricing based on your specifications" - and not a single figure. | ||
| 1Legion | On hold: our decision (HTTP 200) | Reserved-only 8-GPU servers, and the page prints two different 12-month rates for the same card (RTX 5090 from $1.00 in the table, $0.59 in the comparison; H100 SXM $2.40 vs $1.99). | ||
| AIEEV | On hold: our decision (HTTP 200) | Rendered pricing page (Wix) shows no GPU-hour price. | ||
| Alibaba Cloud | On hold: our decision (HTTP 200) | Rendered ECS price list shows no GPU instance rate without choosing region and family by hand. | ||
| AtmosCompute | On hold: our decision (HTTP 200) | Per-card pages list shifting configurations that look like resold marketplace offers (0 vCPU rows, 45 GB A40, 20 GB RTX 3090); what machine a row is cannot be verified on the page. | ||
| BreezeHost | On hold: our decision (HTTP 200) | One card (RTX 5090), out of stock; the on-demand rate appears only in the page title, the body prints a spot "from" price. | ||
| Cleura | On hold: our decision (HTTP 200) | Its public price list (the JSON its pricing page loads) carries VMs and storage but no GPU; no GPU rate found after rendering. | ||
| Ekasunucu | On hold: our decision (HTTP 200) | Rendered GPU server page names H100 and A100 with no hourly price (monthly servers, TRY by default). | ||
| Exoscale | On hold: our decision | Its price file lists the same figures under EUR, CHF and USD, and the page switches the currency label without changing them, so we cannot tell which currency a GPU rate is in. | ||
| GPUhub | On hold: our decision (HTTP 200) | Rendered page and its own open API disagree about tenfold (RTX PRO 6000 $1.16/hr on the page, $11.02 per GPU-hour in the API); which is the price cannot be told. | ||
| Hinode | On hold: our decision (HTTP 200) | Machines priced by VRAM and RAM "from" (L4 24 GB from $1.75/hr), the cheapest being fractions of a GPU; no whole-GPU rate to compare. | ||
| HyperAI | On hold: our decision (HTTP 200) | GPU hours come with a paid membership (the free plan has a GPU quota of 0 hours), so the hourly rate on the page is not the whole price of an hour. | ||
| ionstream | On hold: our decision (HTTP 200) | Its solutions page prints one "from" rate per GPU-hour for each card (B200, B300), with no term. The company told us in writing that these rates assume about a quarter of the server paid up front and that the B200 is no longer available. Until the page states the terms a rate is sold on, it cannot be listed as an hourly rental price. | ||
| Latitude.sh | On hold: our decision | Every GPU plan is priced “Starting At”, the cheapest location. Our decision: not listed until plans are priced per location, so that a floor never enters the Price Index as a rate. | ||
| OmegaGradient | On hold: our decision (HTTP 200) | Open marketplace catalogue with one to four offers per card - below the three verified offers a marketplace floor needs. | ||
| Qubrid | On hold: our decision (HTTP 200) | GPU prices sit behind a "GPU Compute" tab that takes a click; rendering shows only per-token model prices, and clicking is not reading. | ||
| QuickPod | On hold: our decision (HTTP 200) | Rendered home page carries no GPU-hour price; offers appear only inside the signed-in app. | ||
| SpotGPUs | On hold: our decision (HTTP 200) | Prices far below every other source (A100 80GB $0.29, GH200 $0.79 on demand) and payment in crypto; not taken until the offering can be verified as a rentable machine at that rate. | ||
| Theta EdgeCloud | On hold: our decision (HTTP 200) | The pricing path renders "Page not found"; no GPU rate on the rendered site. | ||
| ConoHa (GMO) | We cannot read the prices (HTTP 403) | Cloudflare answers its robots.txt itself with an interactive challenge ("Just a moment...", HTTP 403) to a plain client, so not even the crawling rules can be read, let alone a price page. A challenge is a refusal: it was asked once and not retried, and no browser user agent was tried. | ||
| Cyfuture AI | We cannot read the prices (HTTP 403) | Its static price page reads from a desk, but answers 403 Forbidden to every hourly read from our GitHub Actions runners (2026-09-28, 20:37 and 21:37 UTC), so no quote was ever stored. A refusal is not worked around; the parser module was withdrawn and is re-added only if a re-probe from the runners is answered. | ||
| io.net | We cannot read the prices (HTTP 401) | No pricing page (/pricing 404); the hardware price API answers 401 without authentication. | ||
| Lium | We cannot read the prices | A marketplace where each provider sets its own ask per pod, so a floor can only be taken from the per-pod feed - and its own robots.txt disallows that feed for every agent, including the group naming AI crawlers. The model-level feed it does allow gives only the single cheapest machine and a reference price for models with nothing available, and we publish neither. | ||
| NeevCloud | We cannot read the prices (HTTP 403) | Its price page answers 403 to a plain request from our own user agent; bot protection is never worked around, so the page cannot be read. | ||
| Novita AI | We cannot read the prices | GPU instance prices are drawn in the browser from a keyed API; none are in the page HTML. | — | |
| Oblivus | We cannot read the prices (HTTP 520) | Its pricing page answers our reader with a 520 error and then refuses every request from the same address for about a minute; reading it would mean getting around that protection, which we do not do. | ||
| Salad | We cannot read the prices | The price table ships empty and a script fills it from an API that needs an organization key; we read prices from page HTML or open APIs only. | — | |
| ServerSupply retail source | We cannot read the prices (HTTP 403) | A reseller of datacenter GPUs whose site refuses our reader, so its listings cannot be linked or checked. | — | |
| TensorDock | We cannot read the prices | Its marketplace API, where machines and their prices were listed, returns an empty host list since TensorDock was acquired by Voltage Park. | ||
| UpCloud | We cannot read the prices (HTTP 403) | robots.txt is served and open, but the pricing page answers a plain client with HTTP 403, so the rates cannot be read. Asked once, not retried, and no browser user agent was tried. | ||
| Genesis Cloud | Not trading (HTTP 522) | Reported to be in liquidation, and its site does not load: the origin server is unreachable. | — | |
| Hexabyte | Not trading (HTTP 200) | Swedish host (Umea) with four NVIDIA L4 plans priced by the hour in kronor - l4.gpu1 3.91, l4.gpu2 4.28, l4.gpu3 4.82, l4.dual-gpu 10.69 for two cards, ex VAT - but the service cannot be bought. The table is headed "Prisexempel" (price examples) and the only call to action under every plan, and in the page's hero, is "Ga med i vantelistan": join the waiting list. The figures are drawn by a script; a headless render under our own user agent showed them, so this is not a reading problem. Publishing them would put an hourly rate on the site that nobody can order. | ||
| TheAI Cloud | Not trading (HTTP 200) | Service starts 2026-10-15; on-demand has "no price published yet", only a reserved B300 "from" rate. | ||
| 3NV | Not a GPU-hour (HTTP 200) | An Estonian VM host, not a GPU cloud: its pricing page sells vCPU, RAM and NVMe by the month in euro (3.49 EUR per vCPU, 2.99 EUR per GB of RAM) and the word "GPU" does not appear on it at all. The aggregator that listed it names GPUs only in its own "alternatives" sidebar. | ||
| Akash Network | Not a GPU-hour (HTTP 200) | Prices are "derived from recent provider bids ... weighted by GPU count" - an average, which the marketplace rule excludes (a real ask, never an average). | — | |
| compute.cheap | Not a GPU-hour (HTTP 200) | Prices set against a "Market" column - a reseller of other clouds' capacity, not an operator's own rate card. | — | |
| GPUOutlet | Not a GPU-hour (HTTP 200) | Rendered, the page is a "Reserved GPU Capacity Marketplace" - resale of reserved capacity, not an hourly rate card of its own. | — | |
| GreenNode | Not a GPU-hour (HTTP 200) | Its pricing page prices GPU instances per 30 days in Vietnamese dong, and prints hourly figures only for platform products ("Starting at 2,356 VND/hour"), never for a named card. Not a GPU-hour. The dong is also a rate our fx_rate column cannot carry to six decimals (about 0.000038 USD), so a dong source waits for that column to widen. Re-probed 2026-09-28. | ||
| Impossible Cloud | Not a GPU-hour (HTTP 200) | Long-term cluster listings for future delivery (GB300 NVL72 "2027 Q1", 36-month terms, "Request this cluster") - capacity sold ahead, not an hourly rate. | ||
| Lade | Not a GPU-hour (HTTP 200) | No GPU product. Its pricing page sells compute credits and CPU instances (from 0.0017 USD/hour, 1.25 USD/month upward) and names no GPU at all. | ||
| Leaseweb | Not a GPU-hour (HTTP 200) | GPU servers are dedicated servers billed by the month, not by the hour. The pricing page is an Angular shell, but the shop backend it names itself (shop-api.leaseweb.com, SAP Commerce OCC) answers our user agent with no key: 787 GPU offers, e.g. a Dell R750 (16SFF+8NVMe) GPU 2x Xeon 4314 with an L4 in Amsterdam at EUR 395.30. The product's own sapUnit is {"code":"MON","name":"Months"} and delivery is quoted at five days. Readable, but not an hourly rate - out of scope the way LeaderGPU is. | ||
| Ravion | Not a GPU-hour (HTTP 200) | Not a GPU cloud: it is the successor to Flightcontrol, a deployment platform priced per module (compute stays in the customer's own cloud account). Its pricing page mentions GPUs zero times. | ||
| Runcrate | Not a GPU-hour | A broker across other clouds: each GPU is shown at the lowest of several offers from different providers and regions (its own API gives the minimum and maximum), so the figure is another operator's price, not a rate card of its own. | — | |
| Spheron | Not a GPU-hour (HTTP 200) | Its own pricing page calls the figures "live rates across 5+ providers" and "the cheapest live on-demand rate per GPU on the Spheron marketplace", with a "Supplier Matchmaking" service that sources from a partner network - a broker of other operators' offers, and a single cheapest ask rather than a floor over a window. | — | |
| Thalassa Cloud | Not a GPU-hour (HTTP 200) | A European Kubernetes and VM platform priced in euro per vCPU, per GB and per volume-hour; its pricing page names no GPU. | ||
| Baseten · every GPU, on-demand one quote | Quote retired | Never serverless: the four figures come from "Dedicated Deployments" (GPU instances billed per minute) and the word "serverless" is absent from www.baseten.co/pricing/ - Baseten's serverless product is "Model APIs", priced per million tokens and not published by us. Republished with the market unset, so it follows price_type (on_demand), from the same figures. Price audit 2026-09-21. | — | |
| CoreWeave · NVIDIA A100 80GB, spot () one quote | Quote retired | CoreWeave spot is priced per region: the page prints one rate card for North America and one for Europe. This quote carried the cheaper of the two with an empty region, so the figure could not be found in either table by a reader following the link. Replaced by one quote per region from 2026-09-19; no figure was deleted. | — | |
| CoreWeave · NVIDIA B200, spot () one quote | Quote retired | CoreWeave spot is priced per region: the page prints one rate card for North America and one for Europe. This quote carried the cheaper of the two with an empty region, so the figure could not be found in either table by a reader following the link. Replaced by one quote per region from 2026-09-19; no figure was deleted. | — | |
| CoreWeave · NVIDIA B300, spot () one quote | Quote retired | CoreWeave spot is priced per region: the page prints one rate card for North America and one for Europe. This quote carried the cheaper of the two with an empty region, so the figure could not be found in either table by a reader following the link. Replaced by one quote per region from 2026-09-19; no figure was deleted. | — | |
| CoreWeave · NVIDIA H100 SXM, spot () one quote | Quote retired | CoreWeave spot is priced per region: the page prints one rate card for North America and one for Europe. This quote carried the cheaper of the two with an empty region, so the figure could not be found in either table by a reader following the link. Replaced by one quote per region from 2026-09-19; no figure was deleted. | — | |
| CoreWeave · NVIDIA H200, spot () one quote | Quote retired | CoreWeave spot is priced per region: the page prints one rate card for North America and one for Europe. This quote carried the cheaper of the two with an empty region, so the figure could not be found in either table by a reader following the link. Replaced by one quote per region from 2026-09-19; no figure was deleted. | — | |
| CoreWeave · NVIDIA L40, spot () one quote | Quote retired | CoreWeave spot is priced per region: the page prints one rate card for North America and one for Europe. This quote carried the cheaper of the two with an empty region, so the figure could not be found in either table by a reader following the link. Replaced by one quote per region from 2026-09-19; no figure was deleted. | — | |
| CoreWeave · NVIDIA L40S, spot () one quote | Quote retired | CoreWeave spot is priced per region: the page prints one rate card for North America and one for Europe. This quote carried the cheaper of the two with an empty region, so the figure could not be found in either table by a reader following the link. Replaced by one quote per region from 2026-09-19; no figure was deleted. | — | |
| GMI Cloud · NVIDIA H100 SXM, on-demand one quote | Quote retired | Never shown to be SXM: the GMI Cloud card reads "NVIDIA H100 GPU" with no form factor or memory, and h100-sxm rested on the research doc, not the page. Read as the form-factor-neutral h100 from 2026-09-19. | — | |
| ionstream · every GPU, on-demand one quote | Quote retired | ionstream no longer read: its CRO wrote (2026-09-28) that the "from" per-GPU-hour rates assume about 25% of the server paid up front and that the B200 is no longer available; the page states neither, so the rates cannot be published as on-demand. | — | |
| Microsoft Azure · NVIDIA H200, spot one quote | Quote retired | Azure lists H200 spot at the on-demand price in every region (32 of 32, 2026-09-25): no spot market; parser emits spot only where it differs (commit 4f57fb0) | — | |
| Nebius · every GPU, reserved one quote | Quote retired | Never a commitment: Nebius column 4 is "Preemptible, GPU-hour" (page and 2026-09-07 fixture). Read as committed until 2026-09-15, republished as spot from the same figures. Price audit 2026-09-15. | — | |
| packet.ai · every GPU, on-demand (Dynamic) one quote | Quote retired | packet.ai Dynamic is a shared card, read as market shared from 2026-09-27 (owner's decision); the on_demand quotes before that are superseded | — | |
| Paperspace · NVIDIA V100, on-demand (Paperspace) one quote | Quote retired | The Paperspace x8 node was "V100-32G", the 32 GB card, read as our v100 (the 16 GB card) from 2026-09-08; the parser reports it as unmapped from 2026-09-29, as immers and Cyfuture do. | — | |
| RunPod · every GPU, spot one quote | Quote retired | RunPod API returned spot equal to on-demand for every GPU and tier, 2026-09-07 to 2026-09-11; the parser emits spot only when it differs | — | |
| Thunder Compute · NVIDIA H100 PCIe, on-demand one quote | Quote retired | Thunder Compute's redesigned pricing page (2026-09-29) names its H100 an "H100 SXM" at the same $3.20 it listed as "H100 PCIe" until 2026-09-28; the card is read as the page names it, as h100-sxm, from 2026-09-29. | — | |
| Vast.ai · NVIDIA A100 80GB, on-demand one quote | Quote retired | Never an A100 80GB: Vast gpu_name "A100 SXM4" covers the 40 GB and the 80 GB card and only gpu_ram tells them apart, so the floor was taken over both books and published the 40 GB machines (13 of 17 offers) as 80 GB - $0.5111 and $0.5891 per GPU against a real 80 GB floor of $1.13-$1.70. Parser now sends a gpu_ram window; the 40 GB book is a100-40. Price audit 2026-09-17. | — | |
| Vultr · every GPU, on-demand one quote | Quote retired | Vultr no longer read: its plans API (the only machine-readable price source) is disallowed for every agent by api.vultr.com/robots.txt, and the L40S plans it listed had zero locations - a rate nobody could order. 2026-09-18. | — | |
| Vast.ai · NVIDIA H100 NVL (Multi-GPU machines) one quote | Too few offers for a floor | Withheld: 2 verified offer(s), a floor needs 3. The floor is published again once enough offers are verified. | — | |
| Vast.ai · NVIDIA H100 PCIe one quote | Too few offers for a floor | Withheld: 2 verified offer(s), a floor needs 3. The floor is published again once enough offers are verified. | — | |
| Vast.ai · NVIDIA L4 (Multi-GPU machines) one quote | Too few offers for a floor | Withheld: 1 verified offer(s), a floor needs 3. The floor is published again once enough offers are verified. | — | |
| Vast.ai · NVIDIA RTX 4000 Ada one quote | Too few offers for a floor | Withheld: 2 verified offer(s), a floor needs 3. The floor is published again once enough offers are verified. | — | |
| Vast.ai · NVIDIA RTX 5000 Ada Generation one quote | Too few offers for a floor | Withheld: 1 verified offer(s), a floor needs 3. The floor is published again once enough offers are verified. | — | |
| Vast.ai · NVIDIA V100 one quote | Too few offers for a floor | Withheld: 1 verified offer(s), a floor needs 3. The floor is published again once enough offers are verified. | — |
Suggest a provider
Add your GPU cloudPrices come from your public pricing page, never from this form. Listing is free and can’t be bought. How providers are listed, and who is not, and why
Or email data@gpueconomy.com with the provider’s name, the address of its public pricing page, the API documentation if prices come from an API, and the GPU models it rents. You do not have to work for the provider.
What happens next: we read the page, run our parser against the live source without writing anything, and check the numbers against other providers for the same card. If the page meets the conditions above, the provider appears with every price linked to that page; if it does not, it is not listed yet. We do not promise a date.
What the form keeps
Your email, the link you sent, the page the form was on and the time — no IP address and no browser details. The email is used only about your listing and never shared; to have a submission deleted, write to the address above.
A price on this site is wrong?
Email data@gpueconomy.com with the page on this site and the provider’s page that shows the correct price. Every price here names where it was read and when; a correction is checked against that page, and a mistake is fixed.
Data updated: