# gpu.ai > gpu.ai | Compute; 1 record: free tier, limits, identity cost and evidence. HTML https://fshy.ajam.dev/p/gpu/ | JSON https://fshy.ajam.dev/api/v1/entity/gpu.json | how to read a record: https://fshy.ajam.dev/llms-full.txt ## gpu.ai - `compute/gpu-ai` | paid-only | evidence documented | binding volume-quota | category GPU cloud | tier on-demand - identity a placeholder: card and phone unknown | verified 2026-11-03T07:36:21.683Z (live) | stale on 2026-12-04 - url https://gpu.ai/ - free tier: no free tier: a GPU marketplace that routes each launch to the cheapest connected provider, billed per second. Live cheapest rows from its public pricing API on 2026-11-03: H100 SXM $3.99 an hour (1 GPU, us-east, 4 available); A100 80 GB $1.07 an hour (ca-central, 2 available), $1.23 (us-central) and $1.25 (us-east); L40S $0.80 an hour (us-central, 92 available) and $1.09 (eu-west). The same API call returned no RTX 4090 or L4 row with units available that day. | unit: none | metered on: none | read on: 2026-11-03 | source: [1] - per-object cap: not applicable: the unit is a GPU-hour, not an object | source: [2] - retention: not stated in the pages read | source: [2] - egress: not stated in the pages read | metered: no | source: [2] - identity: email | detail: unverified: the getting-started guide says to create an account and add credit, and a deposit page shows both card deposits and USDC deposits on five chains, so a stablecoin balance is an alternative to a card. The same guide says that during private beta the browser login is internal-only and a key must be requested from the team. No signup form was read, so email is the assumed minimum and card, phone and company checks are not established. - interface: rest, cli, ssh - limits: - volume-quota (binding): rows are per offering with an available count (for example 92 L40S units in us-central, 2 A100 80 GB in ca-central); capacity is pooled from other providers and a row with zero units is omitted by default [1] - limits note: capacity is aggregated from several providers, so the available count of a row changes between reads; the docs state the on-demand server can be shared with other tenants (each on their own GPU) - binding note: each row has a small available count, so the live stock of a given GPU in a given region is the binding number, not a time cap - interface detail: a REST API at api.gpu.ai/v1, a gpu CLI, Python and TypeScript SDKs and an MCP server; instances are reached over SSH. The catalog and pricing endpoints need no key. - tooling: REST API (bearer key), the gpu CLI, official Python and TypeScript SDKs and an MCP server; an instance is created with POST /v1/instances and reached over SSH. Each pricing row carries deployment_type vm or container, a tier of on_demand or spot, and an available count. - composes with: the pricing endpoint is public and priced live, so it can be polled without an account to find the cheapest A100 80 GB, L40S or H100 row before launching. - reliability: unverified: the docs page is versioned V1 May 2026 and its samples were last checked against a demo environment on 2026-08-06; no status page or incident history was read - notes: the same vendor also sells serverless token-metered inference and reserved multi-node clusters by quote; this record prices the on-demand GPU rows. - evidence note: the prices are read from the vendor's own public pricing API on 2026-11-03; RTX 4090 and L4 appear only in the price index (reported) - evidence: - [3] 2026-11-03: the vendor's own pricing API lists an H100 SXM at $3.99 an hour, 1 GPU, us-east, 4 available | capture: sha256:2d4c746f001f4b26850709032c81bb844a642edd80dba8457caa88a404fc8b1e | quote: "gpu_type":"h100_sxm","gpu_count":1,"cpu_cores":27,"ram_gb":287,"storage_gb":0,"region":"us-east","tier":"on_demand","price_per_hour":3.99,"available":4 - [3] 2026-11-03: the vendor's own pricing API lists an A100 80 GB at $1.07 an hour in ca-central, 2 available | capture: sha256:2d4c746f001f4b26850709032c81bb844a642edd80dba8457caa88a404fc8b1e | quote: "gpu_type":"a100_80gb","gpu_count":1,"cpu_cores":28,"ram_gb":120,"storage_gb":100,"region":"ca-central","tier":"on_demand","price_per_hour":1.07,"available":2 - [3] 2026-11-03: the vendor's own pricing API lists an L40S at $0.80 an hour in us-central, 92 available | capture: sha256:2d4c746f001f4b26850709032c81bb844a642edd80dba8457caa88a404fc8b1e | quote: "gpu_type":"l40s","gpu_count":1,"cpu_cores":12,"ram_gb":72,"storage_gb":625,"region":"us-central","tier":"on_demand","price_per_hour":0.8,"available":92 - [4] 2026-11-03: the price index lists an RTX 4090 at $0.89 an hour | capture: sha256:08731e5b6fb101a8b17e0aba129c1b295ddc67e97d8d0b34a2fe99bb4ad4dd17 | quote: | [RTX 4090](https://computeprices.com/providers/gpu/gpus/rtx4090)1× | $0.89/hr | - [4] 2026-11-03: the price index lists an L4 at $0.59 an hour | capture: sha256:08731e5b6fb101a8b17e0aba129c1b295ddc67e97d8d0b34a2fe99bb4ad4dd17 | quote: | [L4](https://computeprices.com/providers/gpu/gpus/l4)1× | $0.59/hr | - [5] 2026-11-03: the first steps are an account and prepaid credit | capture: sha256:190d291883286744e17a7f4a2810301bc083836c95c1be0ce5c77c81e2e47553 | quote: Create an account, add credit, and launch your first cloud GPU on GPU.ai in minutes. - [5] 2026-11-03: during the private beta a key is requested from the team | capture: sha256:190d291883286744e17a7f4a2810301bc083836c95c1be0ce5c77c81e2e47553 | quote: internal-only; request a `gpuai_live_*` key from the team and export it as - [5] 2026-11-03: the catalog and pricing endpoints are public and need no key | capture: sha256:190d291883286744e17a7f4a2810301bc083836c95c1be0ce5c77c81e2e47553 | quote: Read-only catalog endpoints (`/v1/gpu-types`, `/v1/pricing`) need no key at all - [6] 2026-11-03: balance can be funded by crypto deposits as well as card deposits | capture: sha256:f1bdca27a0139812ecb04222b62e2db036b797aee030af76a60c33d0cb5953e6 | quote: crypto deposits qualify exactly like card deposits - [2] 2026-11-03: an instance bills by the second from boot until terminated | capture: sha256:fad187a81c3a720b40bdb67bda6768bb02f55af037fe0d15bcff1c4b1a95111e | quote: A created instance bills by the second from the moment it boots until you terminate it. - [7] 2026-11-03: an on-demand GPU is dedicated to the renter, but the physical host is shared with other tenants | capture: sha256:3487f27395a99c32dfd8c3704a4e3aeeddfeb8326344dee937e6163b69e6dafd | quote: The physical server it sits in, however, is shared with other tenants (each on their own GPU) - [8] 2026-11-03: the access surface is a REST API, SDKs and a CLI | capture: sha256:93bc7d392364decaa04758d36b5e5c5f5840396446dd62a043a244662c6f9400 | quote: A REST API, official Python and TypeScript SDKs, a CLI, and a marketplace of GPU providers, wired through a single binary. - [8] 2026-11-03: the docs list an MCP server for coding agents | capture: sha256:93bc7d392364decaa04758d36b5e5c5f5840396446dd62a043a244662c6f9400 | quote: Connect Claude Code, Claude Desktop, or Cursor straight to GPU.ai. - [9] 2026-11-03: the vendor aggregates capacity from other clouds | capture: sha256:e456eaa524578c42b60f3fef70a5fe0e8ff31839d6bbf33937ec4b6a65adbbdf | quote: provider-agnostic: the same console, CLI and API across every cloud we aggregate - sources: - [1] https://api.gpu.ai/v1/pricing - [2] https://gpu.ai/docs/guides/gpu-instances - [3] https://api.gpu.ai/v1/pricing?include_unavailable=false&limit=200 - [4] https://computeprices.com/providers/gpu - [5] https://gpu.ai/docs/guides/getting-started - [6] https://gpu.ai/docs/guides/stablecoin-deposits - [7] https://gpu.ai/pricing - [8] https://gpu.ai/docs - [9] https://gpu.ai/