Self-Hosted n8n vs Custom Automation Cost at Scale

You install n8n on a small server, point a few workflows at it, and the bill is a rounding error. A 2 vCPU box with 4GB of RAM runs to roughly $10 a month, and the Community Edition licence is genuinely free. For a side project, or a workflow that fires a handful of times a day, this is the right answer and nothing here should talk you out of it.

The trouble starts later — quietly, after the tool has stopped being a toy and started running something you depend on. The free tool stopped being free the moment it became business-critical. That is the point this note is about: where self-hosting an automation platform begins to cost more than owning the real thing, and how to see the crossover before it costs you a bad month.

What "free" actually means at low volume

Credit where it is due. n8n's pricing model is fairer than most. An execution is one complete workflow run, regardless of how many nodes it contains — a 3-step workflow and a 30-step workflow both cost exactly one execution. That is a meaningful contrast with task-metered tools, where every step in a workflow counts separately and a five-step flow burns five units each time it runs.

On the self-hosted side, the marginal cost of an extra execution is close to zero. You are paying for a server, not for usage. A single small VPS — call it $10–14 a month all-in with a domain and off-site backups — will happily process far more than a paid cloud tier allows. So at low and predictable volume, self-hosting wins on raw arithmetic, and it is not close.

This is exactly why the conversation gets dangerous. The early numbers are so good that they set an expectation the system cannot keep once it grows up.

n8n cost per execution: where the cloud line moves

If you stay on n8n's hosted plans rather than self-host, the per-execution cost is the number to watch. On the Starter plan at €24/month for 2,500 executions, you are paying roughly $0.008 per execution. Move up to Pro at €60/month for 10,000 executions and it falls to about $0.005. The Business plan sits at €800/month for 40,000 executions, with overage sold in large blocks (one published figure is €4,000 for an extra 300,000 executions).

Run the volume forward and the model bites. A modest setup — 50 active workflows, each triggering 100 times a day — produces around 150,000 executions a month. That is well past what the Pro tier covers and past the Business tier's 40,000 too. Hosted n8n at that scale means the Business plan or an enterprise quote; self-hosted handles the same load on a server costing under $15 a month. So the obvious move is to self-host. And that is correct — provided you account for what hosting it actually requires.

The hidden costs of self-hosted n8n

Here is where the spreadsheet that justified self-hosting quietly stops telling the truth. The server is the cheapest line. The expensive lines are the ones nobody puts on the invoice.

• Engineer hours. Updates, monitoring checks, patching the OS, Docker and n8n itself, backup verification and debugging. Honest time-tracking on a simple instance lands around 1–3 hours a month. Once it is business-critical and properly run, published estimates put it nearer 10–20 hours a month — and a DevOps engineer's loaded time is not cheap.
• Monitoring and alerting. Datadog, New Relic or a self-hosted equivalent runs roughly $30–150 a month. Without it, you find out a workflow died when a customer tells you.
• Redundancy and uptime. Geographic failover, data replication and the architecture to survive a node dying add a further $100–300 a month.
• Scaling architecture. A single container does not survive growth. Real load means queue mode, a separate Redis cache, a worker tier and a managed PostgreSQL database — the same distributed shape any serious system needs. Memory demand climbs with concurrent executions, and teams routinely jump from 4GB to 16GB+ within months.
• Your time at 2am. The cost that never makes the budget: the emergency response when the thing that runs your business stops at the worst possible moment.

Add it up honestly and the picture changes completely. A self-hosted n8n setup run to a professional standard — proper monitoring, backups, redundancy — is widely costed at $300–800 a month before staff time. The "$10 server" was never the real cost of self-hosting n8n in 2026. It was the cost of the part you could see.

When no-code reaches its limits — the cost you can't price

Money is only half the story. The other half is what happens to the workflow itself as the business leans on it.

Visual builders are wonderful until the logic gets real. Branching, retries, error handling, conditional fan-out — the things any production process needs — turn a clean canvas into what practitioners bluntly call a spaghetti workflow. The visual overhead that made it approachable starts adding latency in complex loops, and the usual fix is to drop into custom code nodes anyway. At that point you have written code, but inside a tool that makes it harder to test, version and reason about than if you had simply written code.

This is the limit, and it is a behaviour you can feel before you can measure it: changes get slower, nobody quite trusts a deploy, and the person who built it becomes the only person who can safely touch it. The platform that was meant to remove a bottleneck has become one. None of that appears on the hosting bill, which is precisely why it does the most damage.

Self-hosted n8n vs custom build cost: the honest crossover

So when is custom automation cheaper than n8n, and when is it just expensive ego? The crossover is not a single number — it is the meeting of three lines.

Volume. Tens of thousands of executions a month, where hosted tiers force you onto Business-plus pricing and self-hosting forces you into a real distributed architecture you now have to operate.

Criticality. The workflow touches revenue, compliance or customers directly, so an outage is not an inconvenience — it is a number on a P&L. The moment downtime has a price, the value of owning the failure modes goes up sharply.

Complexity. The logic has outgrown the canvas. You are already writing code inside the tool, fighting its execution model rather than being served by it.

When all three are true, the comparison flips. A focused custom build — a small service that does your specific job and nothing else — carries no per-execution fee, no platform lock-in, and no visual sprawl. It can be tested, version-controlled and deployed like any other software you depend on. Crucially, the ongoing cost stops scaling with usage and starts scaling with change: you pay to evolve it, not to run each transaction. For a high-volume, business-critical process, that is usually the cheaper curve over any horizon that matters.

The reverse is just as true, and we will say it plainly: if your volume is modest, your logic is simple, and an outage means "fix it next morning," a custom build is the wrong spend. Self-hosted n8n — or even the hosted plan — is the right tool, and we would tell you so rather than sell you a project you do not need.

How to read your own numbers

Before anyone quotes you anything, do the maths the brochures skip. Take your real monthly execution count and your honest growth rate. Add the engineer hours your current setup eats — measured, not guessed — at a real loaded rate. Add monitoring, backups and redundancy at the figures above. Then put a pound value on an hour of downtime for the process in question.

That total is your true run cost today. Compare it to the all-in cost of a system built to do exactly your job — including the part where it largely stops costing you per run. If self-hosting still wins, keep it; you have made a good decision with open eyes. If it does not, you have found the moment the free tool started charging rent.

The mistake almost everyone makes is comparing a custom build against the $10 server, when the fair comparison is against the $300–800-a-month operation the server quietly grew into — plus the hours and the risk. Seen that way, "custom is expensive" stops being obviously true. We work this calculation with founders constantly, and the answer genuinely goes both ways. What matters is that you are comparing the real cost, not the advertised one.

If you want a second set of eyes on where your automation sits on that curve — and an honest read on whether a build is even worth it yet — that is the kind of question we like being asked early, before a bad month forces it.