Archive node hosting cost: cloud vs dedicated infrastructure in 2026

An Ethereum archive node holds the full historical state of the chain. It is already large, and it grows continuously as the chain produces new state. The compute profile is steady. The storage profile is not. That single fact is what makes archive node hosting cost behave differently from a validator or a full node. It is where cloud and dedicated infrastructure diverge the most.

This page covers where the cost actually goes. It also covers why a quote that looks fine in month one does not hold by month twelve.

What an archive node actually requires

A validator needs little storage. A full node needs a few terabytes. An archive node is a different shape of workload. It stores every historical state, so it carries a large and continuously growing storage footprint. It also serves deep historical queries with sustained random reads.

The compute and memory lines are predictable. They do not move much month to month. Storage is the line that grows. On cloud it is also the line billed in the way that compounds.

Cloud cost: compute is fixed, storage is not

On shared-tenant cloud the bill splits into two different behaviours. Compute and memory run at a predictable rate that does not change as the node grows. Storage is the problem.

Cloud block storage is billed per gigabyte per month. An archive node adds new state every month and never gives any back. The storage line climbs every month and never comes down. The instance you sized at the start is undersized within the first year. The bill has been rising the whole way there.

There is no ceiling in that model. You pay for exactly what the chain produces. That reads as efficient until you add up two years of continuous growth and the meter has never stopped.

Dedicated cost: a fixed line that does not compound

On dedicated component-priced infrastructure the same workload is a fixed monthly cost. Storage is priced per terabyte provisioned, not per gigabyte consumed. The rate does not change as the node grows.

The dedicated line is flat. You provision storage ahead of the growth curve and keep the same rate as the node expands. The cloud line climbs every month the chain produces state. That difference only shows up when you look past the first invoice.

The part cloud does not solve

Cost is not the only reason archive workloads end up on dedicated infrastructure. The other reason is I/O. Archive queries are sustained random reads against the full state. That is one of the hardest patterns for shared storage to serve predictably. You compete with every other tenant on the same fabric for the same throughput. Local NVMe on owned hardware removes that contention. The reads land on disks nothing else is touching.

For a workload measured on tail latency and query throughput, that predictability is worth as much as the cost difference. A cheaper archive node that stalls under load is not cheaper.

The short version

A standard cloud instance sized for an archive node today is undersized within the first year. The storage bill compounds the whole way. Dedicated infrastructure prices the same workload as a flat line, with the I/O profile archive queries actually need.

See also: ETH validator hosting cost: AWS vs dedicated for the validator workload in the same series, and why we run dedicated infrastructure for Web3 for the architecture behind it.

Frequently asked questions

What makes archive node hosting cost different from a validator?

Storage. An archive node stores the full historical state and grows continuously, so on shared-tenant cloud the storage bill compounds month over month while compute stays flat. A validator needs very little storage, so its cost stays stable.

Why does an archive node cost more on cloud than on dedicated infrastructure?

Cloud block storage is billed per gigabyte consumed with no ceiling, so the bill climbs every month the chain produces new state. Dedicated infrastructure prices storage per terabyte provisioned at a fixed rate, so the line stays flat as the node grows.

Do archive nodes need local NVMe storage?

Archive queries are sustained random reads against the full state, which shared storage struggles to serve predictably under tenant contention. Local NVMe on owned hardware removes that contention and gives the predictable I/O archive workloads need.