ICPay Blog

Technology

ICP Subnets and Nodes Explained

Behind every canister is a subnet — a group of independent machines that all run the same code and agree on the same state. Understanding nodes, subnets, and consensus explains how the Internet Computer stays tamper-proof without you trusting anyone. Here is the physical layer of the network.

August 16, 2026 · 6 min read

The hierarchy

The Internet Computer has four layers, and it helps to hold them in order:

  • The Internet Computer — the whole network.
  • Subnets — groups of nodes that execute canisters.
  • Nodes — physical machines run by independent node providers.
  • Canisters — the applications living on a subnet.

Nodes are not your servers

A node is a physical machine participating in the network. Crucially, the nodes are not owned by you or by one company — they are operated by independent node providers across multiple data centres and countries. That distribution is what makes the network censorship-resistant: no single operator, government, or cloud provider controls it.

Subnets: groups of nodes that act as one

A subnet is a set of nodes that collectively execute a group of canisters. Think of it as a single fault-tolerant computer made of many machines. When a canister is deployed, it runs on every node of its subnet; every node maintains the full canister state.

If one node fails or goes offline, the subnet continues — the other nodes still hold the complete state. There is no single point of failure, and no backup service to configure, because the replication is built into the network.

Consensus: agreeing on the truth

Multiple machines running the same code still need a way to agree on the order and outcome of calls. That is the job of consensus: each subnet runs a Byzantine-fault-tolerant protocol so that the nodes reach agreement on the next state even if some nodes are faulty or malicious.

Because the computation is deterministic, every honest replica computes the same result; the consensus layer then commits that result as the new canister state. The outcome is a tamper-proof record: to change state, you would have to corrupt a majority of a subnet's nodes, which are operated by different providers in different places.

Why subnets make ICP fast

Subnets do not just add redundancy — they add parallelism. Different subnets process different canisters at the same time, which is how the network scales beyond the limits of a single chain. This design is what allows the theoretical throughput of thousands of transactions per second and finality in about one to two seconds.

Different subnets can also have different trust and performance profiles — some are high-throughput, others are purpose-built for fiduciary applications like Bitcoin custody.

The hierarchy in practice

When you open ICPay, the wallet canister you are talking to runs on a specific subnet, and every call you make is executed and agreed on by that subnet's nodes. Your balances, transfers, and cloud buckets are all records in replicated canister state — verified by consensus, not by a single server you have to trust.

Useful links

Related reading