Earn revenue from idle data center capacity
Connect your servers to Akash and let tenants bid on unused CPU, GPU, memory, and storage. You set the minimum price. You keep operational control of your cluster.
What it replaces
Selling capacity directly means building a sales channel, a billing system, and a support function, then waiting out enterprise procurement cycles. On Akash, your nodes watch the marketplace and bid automatically when a workload matches your hardware and your price floor. Tenants fund the lease before the workload starts, and payment streams to you for as long as it runs.
| Selling capacity directly | Providing on Akash | |
|---|---|---|
| Customer acquisition | Sales team, marketing spend, procurement cycles. | Your nodes bid automatically on matching workloads. |
| Billing | Custom invoicing, collections, cross-border payments. | Escrowed upfront, paid out continuously during the lease. |
| Provisioning | Manual setup per tenant. | Ansible playbook, Kubernetes-based provider service. |
Global Grid. No Off Switch.
Access a global network engineered for high availability. While centralized clouds rely on single points of failure, Akash uses a distributed protocol to keep your workloads independent and resilient against system-wide failure.
61
Active Providers
16k
vCPUs
447
GPUs
93 TB
Memory
784 TB
Storage
How it works for operators
Pricing
You set the floor
You set the minimum price for CPU cores, memory, storage, and each GPU model in your provider configuration. Your node only bids at or above it. No reseller discount schedule and no margin dictated to you.
Payments
No invoicing
Tenants fund a lease escrow before the workload starts, and payment settles to your provider account continuously while the lease runs. No collections, no cross-border invoicing, no defaults.
Hardware requirements
Akash runs on standard x86_64 and ARM64 Linux servers. Existing bare-metal inventory can be grouped into profiles that target different kinds of tenant workload.
| Tier | Minimum spec | What runs on it | Setup |
|---|---|---|---|
| GPU nodes | NVIDIA GPUs, identical model per node (RTX 4090, A100, H100). | Training, LLM inference, rendering. | GPU worker nodes with the NVIDIA container runtime. |
| General compute | 12+ CPU cores, 48 GB+ RAM, 500 GB+ NVMe. | Web applications, database replicas, CI/CD runners. | Ansible, via the official provider playbook. |
| Storage | High-throughput enterprise NVMe arrays. | Container databases, model weight storage, persistent state. | Dedicated storage controllers in the cluster. |
FAQs
A Kubernetes cluster on Ubuntu 24.04 LTS (x86_64), a public IP, and the ports the provider service and tenant ingress need. Setup runs through the official Provider Playbook (Ansible) or the Provider Console. GPU nodes use the NVIDIA GPU Operator, installed as part of the playbook.
Each lease runs in its own Kubernetes namespace, and containers run without elevated privileges by default.
A tenant's workload failing is not counted against you. Your provider record reflects whether your service accepted and fulfilled leases and stayed reachable, not what happens inside a tenant's container. Kubernetes restarts failed containers according to the tenant's own configuration.
Start earning
Put unallocated racks and idle GPUs on an open market. Set your price, keep control of the cluster, get paid per lease.