Plugsky is an AI data center for GPU capacity. Rent inference capacity for your models and agents, or contribute an idle GPU and earn 60% of the revenue Plugsky routes to it.
Plugsky connects AI workloads with GPU capacity. You can buy inference capacity through an OpenAI-compatible API, or turn your own machine into a verified GPU node.
Run chat, RAG, agents, and batch inference on Plugsky's GPU network without buying hardware or managing a cluster.
Contribute an idle GPU to the shared network. Plugsky handles auth, metering, routing, and payments; you keep 60% of routed revenue.
Plugsky lets you rent GPU capacity for AI inference and training, or rent out your own GPU and earn 60% of the revenue. Shared GPU inference is priced at $0.30 input / $2.40 output per 1M tokens; isolated sovereign GPU environments are available for enterprises.
Buyers connect in minutes. GPU owners run a local preflight, register the machine, install the worker, and wait for approval before jobs are routed.
https://api.plugsky.com/v1./nodes; the installer shows a one-line command.The shared network is for non-sensitive workloads only. Every payload is classified before routing. Regulated or confidential data belongs in a dedicated, private, or air-gapped environment.
This is the actual GPU Share view: live network status, a one-step node installer, and per-node jobs and earnings. The screenshot is from Plugsky's validation node; your earnings depend on routed jobs and uptime.
Plugsky keeps 40% as a gross platform share and pays the GPU operator 60%. Move the sliders to model your own token volume and how much of your capacity is actually routed.
Illustrative estimate. Actual earnings depend on routed jobs, model demand, uptime, and your node's throughput.
Rates used: $0.30 per 1M input tokens and $2.40 per 1M output tokens. Operator share: 60%. Platform share: 40%.
At the token volume and utilization above.
At 60% routed utilization, 500M input and 100M output tokens per month.
Token-priced. No idle GPU bill. Best for bursty inference, chat, RAG, and agents.
Reserved capacity, private tenant, predictable throughput. Quoted per deployment.
Operators keep 60% of customer revenue. Payouts monthly via Stripe Connect, $50 minimum.
Whole-GPU hourly rental is a different product from token-priced shared inference. These are published market rates checked on 2026-10-27, so you can compare the models honestly.
| GPU | VRAM | Published hourly range | Source |
|---|---|---|---|
| RTX 4090 | 24 GB | $0.34 – $0.89 / hr | Runpod |
| RTX 5090 | 32 GB | $0.69 – $1.19 / hr | Runpod |
| A100 80GB | 80 GB | $1.19 – $1.79 / hr · $1.99 – $2.79 / hr | Runpod · Lambda |
| H100 80GB | 80 GB | $1.99 – $3.99 / hr · $3.99 / hr | Runpod · Lambda |
| H200 | 141 GB | $3.59 – $5.29 / hr | Runpod |
| B200 | 180 GB | $5.98 – $7.99 / hr · $6.69 / hr | Runpod · Lambda |
Sources: Runpod GPU pricing (updated 2026-10-08) and Lambda AI cloud pricing, checked 2026-10-27. Rates exclude tax and change with supply, commitment, and region. Plugsky's shared network is token-priced, not an hourly whole-GPU rental; dedicated and sovereign capacity is quoted separately.
The current shared model is Plugsky GPU 27B: a 27B, IQ2-mix, Apache-2.0 model with a 32K context window, tools, and JSON mode. The worker is outbound-only.
| Requirement | Minimum | Recommended | Why it matters |
|---|---|---|---|
| VRAM / unified memory | 7 GB | 16 GB | Model weights, KV cache, and context |
| System RAM | 7 GB | 16 GB | Worker and model loading headroom |
| Throughput | 4 tok/s | 12 tok/s | Job completion time and routing quality |
| Network | Outbound HTTPS | Stable broadband | No inbound port or public endpoint |
| Uptime | Best effort | Always-on | More routed jobs and predictable earnings |
Run the preflight before registering: curl -fsSL https://plugsky.com/gpu-worker/preflight.js | node -. Windows: iwr -useb https://plugsky.com/gpu-worker/preflight.js | node -.
If you want your own GPU without joining the shared network, Plugsky can deploy an isolated environment: a dedicated node, private VPC, on-prem, or air-gapped. Your hardware stays in your perimeter; Plugsky provides the control plane, OpenAI-compatible API, metering, and support.
Sovereign and dedicated deployments are enterprise-scoped and quoted per environment. They are not the public shared GPU network.
A buyer gets an OpenAI-compatible API without cluster management. A GPU owner gets metering, routing, payments, and a real revenue share.
Operators keep 60% of routed revenue; 40% is Plugsky's gross platform share.
Stripe Connect payouts with a $50 minimum. Track pending and paid earnings.
Per-job tokens, latency, earnings, node health, and payout history.
The worker is outbound-only. No public endpoint or router change required.
Same chat-completions shape, SDKs, tools, JSON mode, and streaming.
Plugsky GPU 27B supports long prompts, tools, and JSON on the shared network.
UAE/GCC and global routes, with region-locked options for enterprise.
Dedicated, VPC, on-prem, or air-gapped. Your GPU, your perimeter.
| Shared GPU network | Dedicated GPU | Your own GPU (sovereign) | |
|---|---|---|---|
| Best for | Bursty inference, chat, RAG, agents | Steady production, predictable throughput | Sensitive data, full control, air-gapped |
| Pricing | Token-priced: $0.30 / $2.40 per 1M | Custom quote | Enterprise license + support |
| Data boundary | Non-sensitive workloads only | Private tenant | Your perimeter / your region |
| Hardware | Plugsky + verified operators | Reserved GPU | Your GPU or a dedicated node |
| Setup | Minutes | Days | Enterprise scoping |
| Training / fine-tuning | Inference only | By arrangement | By arrangement |
| Inbound ports | None for workers | Private network | None / private network |
Not sure which one fits? Start on the shared API to establish quality and latency, then move sensitive or steady workloads to dedicated or sovereign capacity.
Plugsky lets you rent GPU capacity for AI inference and training, or rent out your own GPU and earn 60% of the revenue Plugsky routes to it. Shared GPU inference is token-priced; dedicated and isolated sovereign environments are available for enterprises.
Shared GPU inference is priced by tokens at $0.30 input and $2.40 output per 1M tokens, not by the hour. Dedicated whole-GPU and isolated sovereign deployments are quoted per deployment.
Yes. Register a compatible macOS, Linux, or Windows machine at plugsky.com/nodes, run the preflight, install the worker, and after approval it can receive non-sensitive inference jobs. Operators keep 60% of routed revenue.
The current Plugsky GPU 27B model needs at least 7 GB of VRAM or unified memory and 7 GB of system RAM. 16 GB is recommended; 32K context needs 16 GB or more. Minimum throughput is 4 tokens per second, with 12 tokens per second recommended.
Earnings depend on how many tokens Plugsky routes to your node, your uptime, and your share of total demand. Operators keep 60% of the customer price. The calculator above shows the math for your own token-volume estimate; it is illustrative, not a guarantee.
The customer pays the published token price. 60% goes to the GPU operator and 40% is Plugsky's gross platform share, before payment processing. Stripe fees are deducted from the platform side, not from the operator.
The shared GPU network is for non-sensitive workloads and classifies payloads before routing. For sensitive data, Plugsky offers dedicated and isolated sovereign environments: private tenant, region-locked routing, VPC, on-prem, or air-gapped deployment, with customer-managed keys and audit logs.
Yes, by arrangement. Plugsky can deploy its control plane and OpenAI-compatible API into your environment while your GPU hardware stays inside your perimeter. This is an enterprise-scoped deployment, not the public shared network.
No. The shared network is explicitly for non-sensitive workloads. Regulated, confidential, or customer data should use a dedicated or isolated sovereign environment.
Plugsky serves the GCC and global routes, with UAE/GCC and other regions available depending on the deployment. Enterprise customers can require region-locked or in-country routing.
Earnings are tracked per job in the operator console. Payouts run monthly through Stripe Connect, with a $50 minimum payout. Operators need a connected Stripe account.
No. The worker is outbound-only. It does not require an inbound port or a public endpoint.
The shared network is for inference. Training and fine-tuning are available on dedicated or isolated sovereign deployments by arrangement.
For capacity, create a Plugsky account and connect through the OpenAI-compatible API. For your own GPU, run the preflight and register at /nodes. For a dedicated or sovereign quote, contact the GPU team.
Start with the shared API, register a GPU node, or scope an isolated sovereign environment with our team.