Compute Capacity,
Precisely Allocated.
Rent the infrastructure you need, from a development workspace to an entire Kubernetes cluster. Choose metered use or a listed commitment and review the complete price before starting.
Use Cases
Compute for the Workloads Providers Actually Run
Interactive AI workstations
Launch SSH, VS Code, and Jupyter-ready workspaces on provider GPUs without reserving an entire cluster when a single node is enough.
Kubernetes namespaces
Give teams a bounded namespace with quota, service-account access, and familiar kubectl workflows for model serving, batch jobs, and experiments.
Dedicated GPU nodes
Reserve the whole machine when workloads need predictable local disk, full GPU visibility, or direct access to the node capacity.
Multi-node clusters
Rent coordinated capacity for distributed training, internal platforms, test environments, and infrastructure teams that already operate on Kubernetes.
Provider monetization
List idle lab, office, datacenter, or workstation capacity with visible pricing, hardware reports, validation status, and uptime history.
Heterogeneous hardware labs
Expose NVIDIA, AMD, CPU-only, and specialized AI workstation capacity through one control plane instead of forcing every machine into one shape.
Built on Kubernetes
Infrastructure Professionals Already Trust
Everything in Compute³ is built on Kubernetes.
- Not Kubernetes-inspired.
- Not Kubernetes-compatible.
- Kubernetes-native.
This allows workloads to run on mature, battle-tested infrastructure patterns already trusted by enterprises, research organizations, AI teams, and cloud providers.
Compute³ is designed around the same orchestration principles used to operate modern large-scale infrastructure.
Secure by Default
Your Infrastructure Stays Private
No public IPs. No port forwarding. No firewall gymnastics.
Joining Compute³ does not require exposing your infrastructure to the internet.
All communication between provider nodes and the platform occurs through secure WireGuard tunnels established outbound from the provider network.
Providers do not need:
- Static IP addresses
- Publicly routable IPs
- Port forwarding
- Firewall rule changes
- Complex networking configuration
A node can be onboarded from a home lab, office, datacenter, or cloud environment with minimal networking requirements.
Compute³ securely connects workloads to infrastructure while keeping provider networks isolated and private.
Rent Exactly What You Need
From a Single Workspace to an Entire Cluster
Infrastructure should adapt to the workload. Compute³ supports multiple consumption models:
- Development workspace with SSH access
- Dedicated Kubernetes namespace
- Dedicated node
- Multi-node cluster
Users can consume exactly the amount of infrastructure they need. No oversized instances. No wasted resources. No paying for capacity that sits idle.
Precise allocation means better utilization and lower costs.
Supported configurations include:
Hosts Stay in Control
The provider decides what resources are available, what can be rented, capacity limits, availability schedules, and tenant permissions.
Compute³ provides orchestration while preserving provider control.
Hardware Without Artificial Limits
Bring the Hardware You Actually Own
Compute³ is designed around real-world infrastructure rather than a narrow list of approved configurations.
Providers are not forced into artificially restricted hardware profiles. Modern infrastructure is heterogeneous. The platform embraces that reality.
Transparent by Design
No Black Boxes
Review onboarding status, benchmark results, allocations and recorded earnings throughout the provider lifecycle. Renters see the complete rental price before accepting.
Provider Docs
Host Setup That Matches ComputeCubed
Provider setup is documented publicly and adapted to the ComputeCubed control plane. It does not require public IP addresses, per-GPU port forwarding, or a separate marketplace daemon.
The host flow is based on a one-time join command, outbound WireGuard connectivity, k3s-agent registration, hardware collection, validation, then listing controls for nodes and clusters.
Read host setup docsPrepare the Linux node
Use a server-oriented Linux install, keep firmware and GPU drivers stable, and verify the hardware locally before onboarding.
Create or select a cluster
ComputeCubed joins provider machines to a managed k3s control plane, so each node inherits the cluster GPU type and rental modes.
Run the one-time join command
The command installs required WireGuard tools, creates an outbound tunnel, starts k3s-agent, and registers hardware with the platform.
Review validation and publish
Hardware collection, network checks, disk tests, GPU checks, and pricing controls determine when a node or cluster is ready to list.
Provider Linux
Recommended Linux Distributions
Start with a server-oriented Linux installation and working GPU drivers. Ubuntu Server is our recommended starting point for your first node. Other distributions may also work when they meet the host requirements.
Ubuntu Server
22.04 LTS · 24.04 LTS
Recommended starting points for preparing your first provider node.
Why Compute³
A Modern Approach to Compute Infrastructure
Kubernetes-native
Built entirely on standard K8s primitives for enterprise-grade orchestration.
Secure Connectivity
Outbound-only WireGuard tunnels ensure provider networks remain completely private.
No Network Hassle
No public IPs required. No port forwarding. Seamless onboarding from anywhere.
Precise Allocation
From a single namespace to multi-node clusters, rent exactly what the workload demands.
Hardware Flexibility
Broad compatibility across GPU vendors and configurations. Mixed topologies supported.
Provider Autonomy
Granular control over sharing policies, schedules, and resource limits.
Transparent Metrics
Open benchmarking and clear visibility into node performance and allocation status.
Transparent Pricing
Hosts set their earnings price. Renters see the complete rental price, including marketplace markup, before accepting.
Enterprise Foundations
Designed for production workloads, AI teams, and serious infrastructure operations.
Pricing
Simple and Transparent
Know the price before you rent.
Rental charges, host earnings and Stripe transfers are recorded separately.
Set the price you want to earn for your resources. ComputeCubed adds a markup that can vary by listing. Renters review the complete price before accepting, and accepted prices stay fixed for that rental and its recurring periods.
A successful €10 top-up adds exactly €10 to Rental Balance; €10 is the minimum. ComputeCubed covers payment-processing and platform-billed Stripe costs. Hosts receive their full eligible earnings in their Stripe account; Stripe may charge that account directly for bank payouts.
- No listing fees
- No host earnings commission
- No ComputeCubed withdrawal fee
What providers earn and what renters pay should never be a mystery.
Host price → renter price
One billable hour with an illustrative 15% markup. Actual markup varies by listing.
Host earns
€1.00
Markup
€0.15
Renter pays
€1.15
Hosts earn their stated resource price. The renter sees the complete price before accepting. Short metered sessions round to whole cents.
Early Access
Join the Closed Beta
Compute³ is currently in closed beta. Access is granted through a lightweight approval process that helps us onboard providers and renters responsibly while the platform scales.