GPUDECK · THE TRACKED UNIVERSE · ABSENCE IS A DECISION, NOT IGNORANCE
GPUDeck's published price record only carries sources with an affirmative documented legal basis under methodology §8 — either a licence or permission, or a recorded reasoned conclusion that no permission is required for that specific document. Silence is never a basis, and terms GPUDeck could not locate are not silence. This register shows the whole universe we track — 21 marketplaces, 54 direct clouds, 5 hyperscalers and 19 long-tail dedicated hosts (99 sources) — and exactly why each is or is not in the record yet. Alphabetical order; nothing here is a ranking or a recommendation.
Every entry below is an aggregator: it sells capacity contributed by third-party operators rather than hardware it owns. That has a consequence worth stating plainly, because this page lists marketplaces and direct clouds as though they were parallel. A machine can appear in both. Most of these feeds — including Lium’s, the one GPUDeck publishes — do not identify which operator runs a given node, so where a card is listed on a marketplace and by a direct provider, GPUDeck cannot rule out that it is the same physical machine counted twice. This is the reason no figure anywhere on this site is computed across sources.
| PROVIDER | TYPE | API | PRICE RIGHTS | 5090 | 4090 | 3090 | BLACKWELL PRO | REFERRAL | STATUS |
|---|---|---|---|---|---|---|---|---|---|
| Akash | Decentralized bidding market | Open API (/v1/providers, rich per-host) | UNKNOWN | ✓ | ✓ | ✓ | ✓ | — | API VERIFIED / RIGHTS PENDING |
| Aquanode AGG | Aggregator | Aggregated live feed | UNKNOWN | ✓ | ✓ | ✓ | ✓ | — | NOT TESTED |
| Clore.ai | P2P host marketplace | API exists | PROHIBITED | ✓ | ✓ | ✓ | ✓ | — | TERMS DO NOT PERMIT REPUBLICATION |
| FluxEdge | Distributed marketplace | Not tested | UNKNOWN | ◑ | ✓ | ◑ | ◑ | — | NOT TESTED |
| GPU.ai AGG | Multi-source aggregator | Not tested | UNKNOWN | ✓ | ✓ | ✓ | ✓ | — | NOT TESTED |
| Hyperbolic | Supplier-network cloud | Not tested | UNKNOWN | UNKNOWN | ✓ | ✓ | — | — | NOT TESTED |
| io.net | Distributed GPU cloud | Not tested | UNKNOWN | ✓ | ✓ | ✓ | ◑ | — | NOT TESTED |
| Lium AGG | Bittensor marketplace | Public no-auth nodes feed, versioned, built for listing sites | ALLOWED (archived) | ✓ | ✓ | ✓ | ✓ | — | PUBLISHED ON /PRICES (DIRECT FEED, OPERATOR GO 2026-08-26) |
| Nosana | Distributed job marketplace | Live per-model market pages | UNKNOWN | ✓ | ✓ | ✓ | ✓ | — | NOT TESTED |
| OctaSpace | Distributed marketplace | API exists | PROHIBITED | ✓ | ✓ | ✓ | ◑ | — | TERMS DO NOT PERMIT REPUBLICATION |
| OpenRelay | Distributed dedicated GPUs | Availability API per docs (endpoint unverified) | UNKNOWN | ✓ | ✓ | ✓ | ✓ | — | NOT TESTED |
| Prime Intellect AGG | Aggregated compute exchange | Not tested | UNKNOWN | ✓ | ✓ | ✓ | ✓ | — | NOT TESTED |
| QuickPod | Community host marketplace | Public endpoints per docs (unverified) | UNKNOWN | ✓ | ✓ | ✓ | ◑ | — | NOT TESTED |
| RunPod (Community) | Community cloud | API exists | UNKNOWN | ✓ | ✓ | ✓ | ✓ | — | API VERIFIED / RIGHTS PENDING |
| SaladCloud | Distributed consumer-GPU containers | Inventory auth-gated | UNKNOWN | ✓ | ✓ | ✓ | — | — | API VERIFIED / RIGHTS PENDING |
| Shadeform AGG | Multi-cloud aggregator | Not tested | UNKNOWN | ✓ | ✓ | ◑ | ✓ | — | NOT TESTED |
| SimplePod | GPU marketplace | API enumerates models + availability (unverified) | UNKNOWN | ◑ | ◑ | ◑ | ◑ | — | NOT TESTED |
| Spheron AGG | Multi-provider marketplace | No reachable public API (whitelist-gated) | UNKNOWN | ✓ | ✓ | UNKNOWN | ✓ | — | NO PUBLIC API |
| TensorDock | Independent-host marketplace | v2 API authenticates; inventory empty at last check | UNKNOWN | UNKNOWN | UNKNOWN | UNKNOWN | UNKNOWN | — | SOURCE REACHABLE / NO CURRENT INVENTORY |
| Theta EdgeCloud | Community + hosted cloud | Not tested | UNKNOWN | ✓ | ✓ | ✓ | ◑ | — | NOT TESTED |
| Vast.ai | P2P host marketplace | API exists | PROHIBITED | ✓ | ✓ | ✓ | ✓ | DISCLOSED REFERRAL | TERMS DO NOT PERMIT REPUBLICATION |
✓ verified offering · ◑ supported, supply intermittent · — not found as current offering · UNKNOWN = unknown/not tested · AGG = aggregator (upstream supply tracked; one physical GPU is never counted twice — a directly authorized source supersedes an aggregator carrying the same provider).
The GPU columns track the taxonomy the record models: RTX 30/40/50 series, RTX A-series, RTX 6000 Ada, the RTX PRO Blackwell family (2000/4000/4500/5000/6000 incl. Server Edition), L4/L40/L40S, A100, H100/H200, B200/B300, GB200/GB300 and AMD MI300X/MI325X where offered. A PROHIBITED rights status means the provider's prices reach GPUDeck only through an openly licensed intermediary, or not at all — never by scraping.
AceCloud (public pricing API; redistribution rights unresolved - quarantined) · AIME · Beam · BreezeHost · Cerebrium · CloudClusters · CloudPe · CloudRift · Cloudzy · Contabo · CoreWeave · Cyfuture · Database Mart · DataOorts · EcoHash · Edgevana · Elastx · EmpirioLabs · Enverge · Exoscale · Fal.ai · Geodd · GPUaaS · GPUhub · Hetzner · HOSTKEY · Hostiko · Hydra Host · HyperAI · Hyperstack · Koyeb · Latitude.sh · LeaderGPU · Leafcloud · Leaseweb · Massed Compute · Nebius · NeuralRack · Nova Cloud · Novita AI · Oblivus · OVHcloud · Packet.ai · Qubrid · Runcrate · RunPod (Secure) · Seeweb (EU; RTX PRO 6000 96GB) · Sesterce · SwissGPU · SynpixCloud · Thunder Compute · Trooper · Verda · Yotta
Akamai Cloud · AWS · Google Cloud (incl. Cloud Run serverless RTX PRO 6000) · Microsoft Azure · Oracle Cloud
Regional and sales-led operators renting physical consumer-GPU systems, usually monthly. Tracked at lower verification priority; none is in the published price record.
1Legion · AlphaVPS · Barrack.ai · BHK Cloud · BTCServers · ComputeThisHub · DaylightCompute · GDEP Solutions · GPU-Mart · GPURental.jp · HiveNet · Immers.cloud · LayerStack · Menkisys · ServerMO · SkyStark · SuperRendersFarm · TurboRender · UltraRender
GetDeploying, GPU Finder, GPU.fund, AIMC, GPUHour, GPUPerHour, CloudGPUPrice, GPURental.net, GPUTracker
Comparison/data services - they aggregate other people's inventory; not rentable supply.
Render Network, Livepeer, Gensyn, generic Bittensor compute subnets, AIOZ, Bacalhau
Compute networks where a specific GPU SKU cannot currently be selected and paid for.
The record's measured side grows by replication, not by scale claims: CONTRIBUTING-DATA.md is the protocol for publishing your own fleet's goodput cells as a second, independently labelled fleet — never averaged into the first.
Publish daily workload, GPU, attempts, completions and occupied GPU-hours under your own fleet identifier. Fleets are never averaged; earnings and worker identity are not accepted. Submitting does not make data part of the record — protocol validation is the gate.