Disclosure: I co-founded NxFlowAI, an AI automation agency. I have no stake in either option, but I do have a stake in you hiring help.
"Self-host it, it is free" is one of the most common pieces of advice small businesses get about n8n. The software is free to run under its licence terms. The system is not. If you are working out AI automation costs when you self-host, these are the questions that decide whether self-hosting is actually cheaper.
1. Who is on call?
A cloud plan includes someone else keeping the service up. Self-hosted, that someone is you or your contractor. If an automation that replies to customers stops on a Saturday, who notices and who fixes it?
2. How do updates happen?
Self-hosted orchestrators need regular updates for features and security fixes. Each update can change node behaviour. Budget time for updating, testing and occasionally rolling back.
3. Where do credentials live?
Your workflows hold API keys for your CRM, email, WhatsApp provider and model provider. Self-hosted, you are responsible for encrypting them, restricting access and rotating them. That is real work.
4. What about backups?
Workflow definitions, credentials and execution history all need backing up, and the restore needs testing at least once.
5. How do you see failures?
Execution logs on a small server fill disks. You need log retention settings, alerting on failures, and someone who reads the alerts.
6. What does scale look like?
A small instance is fine at low volume. Long-running or high-volume workflows may need queue mode and workers, which means more infrastructure to manage.
A fair comparison
Write two columns: cloud plan cost at your expected volume (from the vendor's current pricing page, dated), and self-host cost = server + the hours of setup, updates, backups and incidents multiplied by what those hours cost you.
For many small businesses with no in-house engineer, the cloud plan wins once the hours are counted honestly. For businesses with a technical person who already runs servers, or with data that must stay on their own infrastructure, self-hosting can make sense.
What does not change
Either way, the expensive parts are designing the workflow, testing it on real data, and reviewing AI output. The orchestrator choice is a smaller decision than it feels.
If you are earlier than this and still deciding whether you need a custom flow at all, see the custom-versus-packaged trade-off for SMB automations.
Disclosure (again): I am Ujjwal Dubey, co-founder of NxFlowAI (nxflowai.com), a Mumbai-based AI automation agency working with businesses worldwide remotely.
