What n8n actually costs
Cloud against self-hosting, with the hours nobody puts in the spreadsheet.
By Ahsan Ahmad, Chief Executive Officer · Last reviewed
n8n Cloud is a monthly subscriptionwith an execution allowance. Self-hosting is free for internal business use under n8n's Sustainable Use License but, in our estimate, costs $14 to $150 a month in infrastructure plus real maintenance time - upgrades, backups, monitoring and the occasional broken migration.
The break-even is not about executions. It is about whether your own hours are free. At even two hours a month of maintenance valued at $75 an hour, self-hosting carries $150 of hidden monthly cost, which is more than n8n's entry-level Cloud plans cost. Self-hosting wins clearly on data residency and at high volume; Cloud wins almost everywhere else.
Pricing last checked against n8n.io on 22 September 2026. Plans and execution definitions change - verify before budgeting.
What counts as an execution in n8n?
An execution is one run of a workflow, not one node. A workflow with forty nodes that fires once is a single execution, and a loop that processes 500 records inside one run is still a single execution. n8n's pricing page defines it the same way: a single run of the entire workflow, however many steps it has or how much data it processes.
It is the thing almost everyone gets wrong, and it usually works in your favour.
What multiplies executions is trigger frequency, not workflow complexity. A polling trigger checking every minute is roughly 43,200 executions a month whether or not it finds anything - and that single design decision moves people between plans more often than their actual workload does. Switch pollers to webhooks wherever the source supports it, and widen the interval where it does not.
What does it cost to self-host n8n?
By our estimate, a small production n8n instance costs about $14 to $150 a month in infrastructure, depending on workload, and roughly $90 to $440 a month once one to four hours of your own maintenance time are counted at $75 an hour. The licence costs nothing for internal use; the time is the real cost.
Infrastructure is the easy half. Here is a realistic monthly total at three workload sizes.
| Line item | Light | Moderate | Heavy |
|---|---|---|---|
| VPS | $12 (2 vCPU, 4GB) | $30 (4 vCPU, 8GB) | $70 (8 vCPU, 16GB) |
| Managed Postgres, or self-run | $0 (on-box) | $15 | $40 |
| Backups and object storage | $2 | $5 | $12 |
| Monitoring and alerting | $0 (self-hosted) | $0-10 | $10-25 |
| Infrastructure subtotal | $14 | $50-60 | $132-147 |
| Your time, at 1-4 hrs/month | 1 hr | 2 hrs | 4 hrs |
| Time cost at $75/hr | $75 | $150 | $300 |
| Realistic total | ~$89 | ~$205 | ~$440 |
The time row is the one left out of every comparison that concludes self-hosting is free. It covers version upgrades, a database that fills up, credentials expiring, and the afternoon a queue-mode worker stops picking up jobs. None of it is hard; all of it is yours.
Is n8n Cloud or self-hosting cheaper?
For most small teams, Cloud is cheaper once your own hours are priced in. Self-hosting wins when data residency or an air-gapped environment is required, when volume runs to hundreds of thousands of executions a month, or when you already operate infrastructure and one more service costs you almost nothing to run.
Self-hosting wins when
- Data residency or an air-gapped environment is a hard requirement. This is the strongest reason and it is not a cost argument at all.
- Volume is genuinely high - hundreds of thousands of executions a month, where any per-execution plan gets expensive.
- You already run infrastructure and the marginal operational burden of one more service is near zero.
- You need to modify n8n itself, or run community nodes a hosted plan will not allow.
Cloud wins when
- Your team is small and nobody wants to be on call for an automation platform.
- Volume is modest and predictable.
- You would rather spend the two hours a month building workflows than maintaining the thing that runs them.
- You want upgrades and security patches to be someone else's problem.
Our own view, stated plainly because it is the useful part: most teams under fifty people should use Cloud. The self-hosting saving is real only if you genuinely would not have spent that time on anything else, and in a small team that is rarely true.
What hidden costs come with n8n Cloud and self-hosting?
Self-hosting hides operational costs: upgrades, downtime, backups you must test, scaling work with queue mode and workers, and a dependency on whoever set it up. Cloud hides fewer, but leaving it means exporting workflows and rebuilding infrastructure. API and egress costs are the same either way.
| Self-hosted | Cloud | |
|---|---|---|
| Upgrades | Yours, including the breaking ones | Automatic |
| Downtime | Yours to notice and fix | Theirs, with a status page |
| Backups | Yours to configure and, crucially, to test | Included |
| Scaling | Queue mode, workers, Redis - real work | A plan change |
| Bus factor | Whoever set it up | None |
| Egress and API costs | Same either way | Same either way |
| Leaving | You already have everything | Export workflows, rebuild infrastructure |
When should you stop using n8n and build custom?
When error handling matters more than the happy path, when state has to persist across runs, when changes need tests before they ship, when a workflow grows past about thirty nodes, or when the AI part has become the product. We use n8n on client work and recommend it regularly; recognising that point early saves a rewrite.
- Error handling matters more than the happy path. When a failure at step seven must not leave steps one to six half-applied, you need transactions, idempotency keys and a dead-letter queue. That is code.
- State has to persist across runs. A workflow that needs to know what it did yesterday is an application with a database, not an automation.
- You need tests before a change ships. There is no good way to unit-test a visual workflow, and by the time that matters, it really matters.
- The canvas passed about thirty nodes. Once nobody can hold the workflow in their head, the visual editor has stopped being an advantage and started being the problem.
- The AI part became the product. Evaluation harnesses, retries that distinguish a rate limit from a bad prompt, cost control per request - none of that fits in a node.
That last case is the one we are usually called about. If you are there, AI workflow automation describes how we rebuild it, and pricing will give you a budget range. If the workflow needs to plan rather than follow a fixed route, that is AI agent development instead. If you are not there yet, stay on n8n - we will tell you so on the call.
Choosing between n8n and Zapier in the first place, or deciding whether you need an n8n automation consultant at all? Read workflow automation consulting: n8n vs Zapier.
Frequently asked questions
Free in licence terms for internal business use under n8n's Sustainable Use License, a fair-code licence - but not free in total cost. Budget $14 to $150 a month in infrastructure depending on workload, plus your own time for upgrades, backups, monitoring and the occasional broken migration. The licence fee is the smallest line in the calculation.
One run of a workflow, not one node - n8n's own pricing page defines an execution as a single run of the entire workflow, however many steps it has. A workflow with forty nodes that runs once is one execution. This is the single most misunderstood part of n8n pricing, and it usually means people over-estimate what a plan will cost them - a loop processing 500 records inside one run is still one execution.
Almost always, once you price your own hours honestly. If self-hosting takes even two hours a month of your time and you value that at $75 an hour, that is $150 of hidden cost - more than n8n's entry-level Cloud plans cost. Self-hosting wins on data residency, on unlimited high-volume workloads, and when you genuinely enjoy running infrastructure.
When error handling matters more than the happy path, when state has to persist across runs, when you need real tests before a change ships, or when the workflow has grown past roughly thirty nodes and nobody can reason about it any more. Those are engineering problems, and a visual editor makes them harder rather than easier.
Yes, and for classification, extraction and simple routing it works well. It gets difficult when you need an evaluation harness, retries that distinguish a rate limit from a bad prompt, or cost control per request. Those are the point at which the AI part is the product rather than a step in a workflow.
Your workflow data stays on your infrastructure, yes. Anything a node sends to a third-party API - including a model provider - still leaves, exactly as it would on Cloud. Self-hosting controls where the orchestration runs, not where the integrations send data.
Outgrown your n8n workflows?
Send us the canvas. We will tell you honestly whether it needs rebuilding or just tidying, and give you a fixed price for whichever it is.