A blockchain node costs what its server, disk, bandwidth, and your time cost - nothing more, nothing less. A small testnet node on a modest VPS can run for under $25 a month. An Ethereum mainnet archival node can cost several times that, mostly because of disk. This post breaks the bill into its buckets, compares full and archival nodes with real disk numbers, and gives you a simple model to estimate your own monthly cost before you commit to a server.
If you want the validator side of the same question, read our validator cost breakdown first. If you are choosing where to host, our VPS in India guide covers location and latency. This post is about the monthly bill itself: what you pay for, what changes it, and what people forget to budget.
The four cost buckets
Every node bill has the same four buckets. First, compute: the CPU and RAM the client needs to keep up with the chain. A light testnet node runs fine on 2 vCPU and 4 GB of RAM. A mainnet full node wants 4 to 8 vCPU and 16 to 32 GB of RAM, and a Solana validator wants far more. Second, disk: the chain data itself, which only grows. Third, bandwidth: peers syncing from you, snapshots, and RPC traffic if you serve it. Fourth, your time: upgrades, monitoring, and the occasional 2am incident. The first three show up on an invoice. The fourth shows up in your calendar, and it is the bucket most people underprice.
Compute and disk are the two that move the bill most. A node that falls behind the chain tip burns CPU catching up, and a disk that fills up stops the node entirely. Budget for both with headroom, not at the exact minimum the docs list. The docs give you the floor. Your bill lives above it.
Full node vs archival node: the disk decides the bill
The single biggest cost decision is node type. A full node keeps the current state and recent history, and prunes the rest. An archival node keeps everything, so indexers and explorers can query any block in history. That difference is measured in terabytes. Public measurements from ethereum.org in September 2026 put a Geth full node at about 1.8 TB, a Geth path-based archive node at over 2 TB, with full trie history near 6.5 TB, and an Erigon full archive node at over 2.7 TB. Those numbers were current as of September 2026 and will have grown since, so treat them as a floor and check the client docs on the day you size disk.
Disk is not just capacity, it is speed. Chain clients read and write constantly, so slow disk means slow sync and missed attestations. Budget for NVMe storage on a mainnet node, not the cheapest spinning disk or shared volume a provider offers. The price gap between a slow 2 TB volume and a fast one is small next to the cost of a node that cannot keep up. Our full vs archival guide and our archival setup guide cover which type your use case actually needs. Most apps need a full node. Only indexers, explorers, and analytics tools need archival, and paying for archival when you do not need it is the most common wasted spend in node hosting.
A sample monthly model
The ranges below are illustrative, built from commonly listed VPS and dedicated server tiers as of November 2026. Providers change prices often, so confirm the current price on the provider page before you commit. The point of the model is the shape of the bill, not a quote.
A small testnet or light node: a 2 vCPU, 4 GB RAM VPS with 80 to 160 GB of disk, roughly $10 to $25 a month. Enough to learn on, run a testnet validator, or serve a small internal tool. A mainnet full node: 4 to 8 vCPU, 16 to 32 GB RAM, 2 TB of fast NVMe disk, roughly $40 to $120 a month depending on provider and region. An Ethereum mainnet archival node: the same compute with 3 to 7 TB of NVMe, roughly $150 to $400 a month, and the disk is most of it. A validator on a proof-of-stake chain costs about the same as a full node for infrastructure; the stake itself is separate and is not a monthly cost, it is capital you lock.
Bandwidth can surprise you. Most VPS plans include a generous transfer allowance, but a busy public RPC endpoint or a node serving many peers can exceed it. If you plan to serve public RPC traffic, read our private RPC setup guide and budget for egress separately. Egress fees on some clouds cost more than the server itself when traffic spikes.
Hidden costs operators forget
Five costs hide outside the server invoice. Disk growth: chains grow every day, so a disk that fits today will not fit in six months. Plan a growth margin of at least 30 percent and set a disk alert at 70 percent full. Snapshots and backups: a full chain snapshot is terabytes, and storing it somewhere safe costs money every month. Upgrades: client upgrades need a maintenance window, a snapshot before you start, and sometimes a resync if something goes wrong. Monitoring: the tools are often free, but the alerting, dashboards, and on-call time are not. And finally, recovery: if a disk dies, restoring from a snapshot or resyncing from genesis can take days, and that downtime is a cost too. Our Grafana monitoring guide shows which four numbers to watch so a full disk pages you before it stops the node.
How to cut cost without cutting uptime
Start with the cheapest correct node type. Run a full node unless an indexer or explorer genuinely needs archival history. Prune aggressively where the client allows it, and put chain data on its own volume so a runaway log or snapshot cannot fill the system disk. Use committed or reserved pricing for nodes you know will run for a year, but only after a month on pay-as-you-go proves the sizing is right. Set up disk and sync alerts before launch week, not during it. And keep one spare snapshot you have actually tested restoring from, because an untested backup is not a backup, it is a hope. If the numbers still do not work, a managed node service can be cheaper than a badly sized self-hosted one - you pay a flat fee and stop paying for your own mistakes. That is a real option, not a failure.
Estimate your cost in 10 minutes
Write down four numbers: the chain, the node type, the disk size the client docs list today, and the traffic you expect to serve. Price the server and disk at two providers, add 30 percent disk headroom, add a monitoring alert for disk and sync lag, and write down who gets paged when an alert fires. If the total surprises you, the surprise is usually the disk or the bandwidth, not the CPU. Re-run the estimate every quarter, because chain data grows and provider prices move. A node whose cost you never re-check is a node whose bill quietly doubles.
Running a node is cheap for what it does, and expensive when it is sized wrong. The fix is not a cheaper provider, it is a correct estimate made before you launch. If you want help sizing a node for a launch, or a second pair of eyes on a bill that grew faster than the chain, book a free infra audit with DevOpsBY. We run nodes, RPC, and validator infrastructure for Web3 teams, and we will tell you plainly when a smaller server is the right answer.
Disclaimer: This post is educational only, not financial advice. Figures are listing or project data as of 2026-11-09 and may change - always confirm version numbers, requirements, and dates in the official docs before acting on them.
