If you run n8n for clients, here's the short version: self-host the moment you pass two or three clients. n8n Cloud prices per instance, so the agency model — one isolated instance per client — turns into a $500-plus monthly bill that scales linearly with your client count. One self-hosted box runs dozens of clients on unlimited executions for the price of a couple of coffees a week. Past the third client, the math stops being close.
We run 200+ workflows across our client base on self-hosted n8n, and the single most common question we get from other agency owners is some version of "is Cloud fine, or do I need to self-host?" Usually they ask it after their n8n bill quietly became their third-biggest software cost. So let me give you the answer we'd give a friend, with the actual numbers.
The pricing model is the whole story
Most "n8n Cloud vs self-hosted" posts compare features. For an agency, features barely matter — the billing unit is what decides your margin.
n8n Cloud charges per instance, and it meters executions. That's fine for one company automating its own operations. It's a trap for an agency, because the clean way to run client work is to isolate each client, and on Cloud each isolated instance is another paid plan. Your cost grows in a straight line with your client count, and every polling trigger eats into an execution quota you're already paying for.
Self-hosted n8n flips both of those. It's the same open-source software, it has no execution cap, and one server can hold many clients. Your infrastructure cost is basically flat while your client count climbs. That gap — linear vs flat — is the entire decision.
The real cost-per-client math
Here's what it actually looks like once you have a handful of clients. Cloud figures use publicly listed 2026 tiers (Starter ~$24/mo, Pro ~$60/mo per instance); self-hosted assumes one right-sized VPS running everyone.
| Clients | n8n Cloud (isolated, Pro tier) | Self-hosted (one VPS) | Monthly difference | |---|---|---|---| | 1 | ~$60 | ~$25 | ~$35 | | 3 | ~$180 | ~$25–40 | ~$140–155 | | 5 | ~$300 | ~$30–40 | ~$260–270 | | 10 | ~$600 | ~$40–60 | ~$540–560 |
The crossover is the third client. One or two clients, Cloud's hands-off convenience can be worth the premium. By client three the difference is a real line item, and by client ten you're paying roughly ten times more to not manage a server — while also fighting execution limits the self-hosted stack never has.
And that table is generous to Cloud. It assumes you never blow past an execution tier. In practice, one client with a five-minute polling trigger burns ~8,600 executions a month on its own — enough to drain a Starter plan in about nine days and push you to upgrade. On self-hosted, that same workflow costs nothing extra.
When Cloud is genuinely the right call
I'm not anti-Cloud. There are cases where paying the premium is the correct business decision:
- You have one or two clients. The savings don't justify owning infrastructure yet. Ship on Cloud, revisit at client three.
- You have zero ops appetite. If nobody on your team will ever babysit a server, a crashed self-hosted box that takes down every client at 2 a.m. is worse than a bigger invoice. Be honest about this.
- A client mandates a vendor-managed SLA. Some enterprise clients want a named provider on the hook. Give that one client Cloud and self-host the rest — you don't have to pick one model for the whole roster.
That last point is the one most people miss: it's not all-or-nothing. Run your book on a self-hosted instance and peel off the one compliance-heavy client onto dedicated infrastructure. Best of both.
The part nobody puts on the pricing page: hidden costs
Self-hosting is free software, not free. The costs just move from a subscription line to your time:
- Updates and backups. You own version pinning, database backups, and SSL renewal. Automate the backups off-box on day one — a corrupted database with no backup is the only truly unrecoverable failure here.
- Monitoring. Without uptime and error alerts, you find out a client's automation died when the client tells you. Wire up monitoring before you onboard anyone.
- Right-sizing. The cheapest 1GB VPS will fall over on heavy payloads or concurrent runs. An out-of-memory crash hits every client at once, so the extra $20/month of RAM headroom is the cheapest insurance you'll buy. (This is exactly the uptime problem our own product, pingitnow.com, exists to catch early.)
Realistically this is a few hours a month once it's set up. That's the honest trade: a few hours of ops for a few hundred dollars of saved margin, every month, forever.
How we run multi-tenant n8n
The safe way to put many clients on one instance is boring on purpose. Give every client their own folder or project, their own stored credentials, and their own webhook path prefix. Never share one client's API keys across another's workflows. Keep environment secrets in the host, not pasted into nodes. Lock the editor to your team — clients get outcomes, not canvas access. Back up daily, monitor everything, and pin your version so updates are deliberate.
If a client has strict data-residency or compliance requirements, don't contort the shared instance to satisfy them. Give that client a dedicated container or a second instance and leave the rest on the shared box. Escalate isolation only where it's actually required.
This is the same architecture discipline behind the workflows in our rundown of the n8n workflows every agency needs, and it's what makes migrations sane — if you're coming off Zapier or Make, our guide to migrating Zapier and Make workflows to n8n walks the move step by step. Self-hosting is also what makes the deeper integrations practical, like the custom n8n Hyros nodes we built for attribution pipelines — the kind of thing you can't do inside a locked-down cloud tenant.
"What about managed n8n hosting?"
Fair question, because there's a third option the two-way comparison hides. Managed self-hosting providers run the open-source n8n for you on infrastructure they own — you get unlimited executions and a single flat-ish bill without touching a server. For 2026 the going rate is roughly $20–$40/month total for an instance that holds multiple clients, which is dramatically cheaper than stacking Cloud instances.
It's a real middle path, and for a lot of agencies it's the right one. You keep the cost profile of self-hosting — flat, not linear — while outsourcing the updates and uptime you'd otherwise own. The trade-off is control: you're trusting a third party with your instance and your clients' data, and you're limited to what their platform exposes. If you need deep customization, custom nodes, or specific data-residency guarantees, raw self-hosting still wins. If you just want off the Cloud per-instance treadmill without hiring for ops, managed hosting closes most of the gap.
My rule of thumb: raw self-host if you have someone who can own the box, managed if you don't, Cloud only for one or two clients or a mandated SLA. All three beat "a separate Cloud plan per client," which is the one setup that quietly eats your margin.
So which one for you?
- One or two clients, or no ops appetite? Start on n8n Cloud. Don't over-engineer before you have to.
- Three or more clients and someone who can own a server? Self-host. The margin you keep pays for the ops time many times over, and unlimited executions let you build for the client instead of the meter.
- Mostly self-hosted but one client needs a managed SLA? Do both. One dedicated instance for them, one shared box for everyone else.
The framing that keeps agencies stuck is treating this as a tooling question. It's a margin question. When automation is a flat cost and your client count grows, every new retainer is nearly pure margin on the infra side. When it scales linearly with Cloud instances, you're renting away the exact thing that should make an automation agency profitable.
If you're already running client automations and you're not sure whether your setup is costing you margin — or you want it architected right the first time so you're not migrating clients off Cloud in six months — that's the call we make with agency owners every week. Get a free automation audit and we'll look at your client count, your workflow load, and your current bill, then tell you straight whether to self-host, stay on Cloud, or run a hybrid — even if the honest answer is "you're fine where you are."
Carlos Aragon is the founder of VIXI, an AI-first automation agency in Allen, TX (Dallas area), running 200+ automation workflows for clients on self-hosted n8n. Pricing figures reflect publicly listed 2026 tiers; confirm current numbers directly on the n8n pricing page and in the n8n docs before you build, as tiers change.