Subnets define who provides compute, who deploys workloads, and what the network exists to do.
The Network
A subnet is a controlled slice of shared compute. It gives operators, providers, and developers a shared network with a clear purpose.
Define the rules. Set the mission.
Launch a subnet and decide who provides compute, who deploys, and what the network exists to do. Build a community around a purpose.
Learn more →Supply compute. Earn from demand.
Contribute hardware to subnets that match your machines. Earn from real workloads — not speculation. Any GPU, CPU, or accelerator.
Learn more →Deploy workloads. Build products.
Point at a subnet, deploy your container, and the network handles placement and scale. No capacity planning. No surprise bills.
Learn more →Why subnets
Most of the world's compute isn't connected. It sits in labs, offices, and homes — unused. Subnets bring it together around a mission.
Create a subnet around a purpose. Research, a product, a company, a movement. The mission is the magnet.
They contribute GPUs, spare machines, lab hardware. Your community becomes your infrastructure. No procurement. No budget cycle.
Deploy your workload. It lands on available machines. The compute is there because the people are. Costs fall as the community grows.
Purpose
Open compute is useful. Purpose makes it thrive. Subnets give people a reason to participate, support workloads, and build around the same network.
Run large-scale experiments on global compute.
Spin up a subnet around a problem. People join because it matters. They contribute GPUs. You run jobs. Results improve as the network grows.
Controlled compute with clear boundaries.
Create a private subnet. Decide who can join and what can run. Keep sensitive workloads contained. Scale without giving up control.
Turn a product into infrastructure.
Launch a subnet around your product. Users bring machines because they care. Usage becomes supply. Costs fall as the network grows.
Infrastructure
Subnets run on open compute. They combine local control with global reach.
Anyone can build. Anyone can provide.
No approval process to get started. Create a subnet, contribute hardware, or deploy workloads — the network is permissionless by default.
GPUs, CPUs, ARM, x86, edge, and cloud.
The network treats all hardware as compute capacity. Different architectures, different vendors, different locations — one unified system.
Runs across vendors, regions, and clusters.
Workloads are placed on the best available hardware regardless of where it sits. The network resolves geography so you don't have to.
More demand brings more providers.
Supply grows with the network. Every new provider adds capacity, every subnet can attract its own supply. Every new provider means your plan stretches further.