Best VPS for n8n in 2026
Updated Aug 2026 · 8 plans benchmarked
n8n spends almost all of its life idle — a Node process waiting on a webhook or a cron trigger, comfortable in well under a gigabyte. The sizing decision is about the spikes: a workflow that pulls a large API response, processes a spreadsheet, or fans out over hundreds of items will briefly want several times its idle footprint, and that is the moment a 1 GB box dies.
- Idle is cheap, bursts are not. n8n sits under 1 GB doing nothing, but a single execution that loads a large payload into memory can multiply that — 2 GB is the honest floor, and the burst profile is what you should size for, not the average.
- Move off the default SQLite to Postgres as soon as the instance matters. Execution history grows fast, SQLite gets unhappy under concurrent writes, and Postgres is also the prerequisite for queue mode.
- Queue mode (a main process plus Redis plus separate workers) is what you switch to when workflows start overlapping. It needs more RAM than a single container, so leave headroom or plan a second box for workers.
- Reliability matters more than speed — a webhook that arrives while your server is rebooting is simply lost. Prefer a provider with a solid uptime record, and keep the encryption key and database backed up, since losing the key means losing every stored credential.
The picks
The value pick: a CX22 (2 vCPU / 4 GB) gives n8n roughly four times its idle footprint, which is exactly the burst headroom that keeps a heavy workflow from taking the instance down. Hard to beat on price for an always-on service.
The most RAM per dollar of the group — the sort of headroom that lets you run n8n in queue mode with Postgres, Redis and a couple of workers on one box without watching the memory graph.
Hourly billing suits automation work — spin up a second instance to test a risky workflow, then destroy it. Custom sizing also fits queue mode nicely, letting you add RAM for workers without buying vCPUs you'll never use.
The polished option: predictable pricing, a managed Postgres you can point n8n at instead of running your own, and documentation that saves time when you're wiring up webhooks and TLS. You pay a premium for that.
The picks, side by side
Prices verified Aug 2026 · How we benchmark →
Common questions
How much RAM does n8n need?
An idle n8n instance fits in well under 1 GB, but that is the wrong number to size on. Give it 2 GB so a burst — a large API response, a file to parse, a loop over hundreds of items — has somewhere to go, and 4 GB if you run heavy workflows or queue mode.
Should I use SQLite or Postgres with n8n?
SQLite is fine for a first look; Postgres is the answer for anything you depend on. Execution history grows quickly, SQLite struggles once workflows run concurrently, and queue mode requires Postgres anyway — switching later means migrating data, so start there if you can.
Can I self-host n8n commercially?
n8n is fair-code under a Sustainable Use License, not a classic open-source licence: internal business use is permitted, but reselling n8n or offering it as a hosted service to others is not. Read the current licence terms yourself before building a product on top of it.