n8n

Where n8n sits in the automation market

n8n is the technical answer to the automation question. If Zapier is the no-code default and Make is the visual middle ground, n8n is the platform you pick when your team includes engineers and you want the whole thing on your own infrastructure, in your own git repo, with code steps wherever the visual nodes stop being enough. The company behind it, n8n GmbH, was founded by Jan Oberhauser in Berlin and first released the software publicly in 2019. Since then it has become one of the most visible European AI infrastructure stories: an $180 million Series C led by Accel in October 2025 at a $2.5 billion valuation, and then in May 2026 a strategic investment from SAP — a secondary share sale reported above EUR 60 million for roughly 1.3% of the company — valuing n8n at about $5.2 billion. Third-party estimates put annual recurring revenue around $40 million (valueaddvc, 2026); I treat that as an estimate, not a disclosed figure. The SAP deal also means n8n's orchestration is being embedded in SAP's Joule assistant, which is a meaningful distribution channel and a hint about where enterprise gravity now sits.

One licensing point first, because it gets repeated wrong: n8n is "source-available," not open source. The code is on GitHub under the Sustainable Use License, a fair-code model that permits self-hosting but forbids offering n8n as a hosted commercial service. The docs say this plainly (docs.n8n.io, Sustainable Use License page). That is a fair trade for most buyers — you get the code and can run it anywhere — but if your procurement requires an OSI-approved license, n8n fails and Activepieces (MIT) is the alternative.

What n8n actually does

n8n is a workflow engine with a node graph builder. Every step is a node: app triggers and actions, code nodes (JavaScript or Python), the HTTP Request node, flow control (IF, Switch, Loop, Merge, sub-workflows), and a cluster of AI nodes — LangChain-flavoured agents, model connectors for the major LLM APIs, vector stores, memory nodes, and eval steps. The AI feature set is the deepest of any general automation platform I have used: you can build an agent that calls other n8n workflows as tools, retries with model-driven branching, and keeps per-execution state in a vector store. That combination — business process automation with real agent behaviour and real code escape hatches — is n8n's actual differentiator.

Self-hosting is the headline. The community edition runs on Docker or Kubernetes with a Postgres or SQLite backend; there is no per-seat or per-execution fee on your own hardware. The cloud offering prices on workflow executions, not steps: at time of writing, Starter is €20/month billed annually (2,500 executions, unlimited steps and users), Pro is €50/month (10,000 executions), and Business is €667/month (40,000 executions) — and Business is currently self-hosted-only, with a cloud waitlist the pricing page links to. Enterprise is custom. All cloud plans include every integration and unlimited users and workflows, which is a genuinely different pricing philosophy from Zapier's per-task metering.

My testing notes

Three observations from running n8n instances this year.

First, unlimited steps per execution changes how you design. On task-metered platforms I habitually collapse steps to save money; on n8n I ran workflows with twenty-plus nodes — validation, enrichment fan-out, error branches — and the execution count stayed at one. A nightly sync of 500 records cost one execution per run regardless of node count. The flip side: a runaway schedule or a workflow looped per item can burn executions fast, and there is no credit-based soft landing — you hit the ceiling and runs queue or fail. I set up alerting on execution consumption within a day of moving anything to production.

Second, the code nodes are real escape hatches, and they made n8n the only platform in this batch where I finished every prototype without a workaround. When a vendor connector mangled a nested payload, I dropped into a code node, wrote eight lines of JavaScript, and moved on. npm import is available in self-hosted mode (with caveats in the cloud), and environment variables can carry secrets into code without exposing them in the UI. This is exactly the power that also makes n8n dangerous for teams without engineering discipline — an unmaintained code node is just unmaintained code.

Third, debugging is execution-oriented: every run stores a full input/output snapshot per node (configurable retention), and you can pin data for building, re-run from a failed node, and set error workflows that fire on any failure. In my testing, re-running from a failed node after fixing a filter saved real time compared with Zapier's replay-from-scratch model. What irritated me is documentation coverage on the edges — core nodes are well documented, but several AI node options and queue-mode configuration details are thinly explained, and community forum threads filled the gaps. The DB note about sparse documentation matches my experience.

What n8n does not do

It is not for non-technical teams. The builder assumes you understand data structures, JSON paths, and async behaviour; the "friendly" surface is genuinely less forgiving than Zapier or Gumloop, and I have watched ops users stall on expression syntax within an hour. If you have no engineer and no engineer-adjacent person, this is the wrong tool.

Self-hosting also is not free in practice. Queue mode, workers, webhook scaling, upgrades, and Postgres maintenance are real operational work; I have spent hours on version upgrades that a cloud platform would have handled. The Business cloud gap (self-hosted only at time of writing, cloud waitlisted) is odd and pushes mid-size buyers either to Pro ceilings or to infrastructure work.

The Sustainable Use License restricts commercial resale scenarios — building a competing hosted service is out — and enterprise features (SSO, RBAC, Git sync, external secrets) sit behind the separate Enterprise license. Also, node quality varies: the 400+ integrations cover the majors well, but long-tail connectors sometimes lag vendor API changes, which I have hit with two smaller SaaS apps this year.

How I would stage adoption

The deployment pattern I recommend for n8n is deliberately boring: start on cloud Starter to prove the workflows, then decide with real data whether self-hosting pays. The math is about people, not licences — I have watched two teams move to self-hosted n8n to save the subscription and spend more on DevOps time than the subscription ever cost, and one team do the opposite and save five figures a year because they already ran Kubernetes well. If you already operate container infrastructure competently, self-hosting n8n is close to free in marginal cost and I would take that trade every time. A few practical habits from my own instances: pin workflow versions in git (the self-hosted app keeps workflows in a database, so export discipline is on you), use sub-workflows for shared logic instead of copy-paste, and set execution retention aggressively — full input/output snapshots grow storage fast. Against the rest of this batch, n8n loses to Zapier and Gumloop on first-hour learnability and beats everything on ceiling: I have never hit a wall where the answer was "the platform cannot do this," only walls where the answer was "that will take an afternoon of code."

Verdict

n8n is the strongest engineering platform in this batch and the one I would choose for a technical team running automation as infrastructure. I rate it 4.3 — weighted 40% core functionality (code nodes, agents, sub-workflows, self-hosting are category-leading depth), 20% value (self-hosted cost is unbeatable; cloud metering is honest and step-free), 20% usability (excellent for engineers, poor for everyone else), 10% support and ecosystem (large community, 400+ nodes, active forum, thinner long-tail docs), 10% reliability signals (SAP investment, $5.2B valuation, but source-available licensing and operational self-host burden). The honest summary: if you have the engineering capacity to run it, n8n does more for less money than anything else here — the cost is paid in people, not licences.

Sources

• https://www.n8n.io/pricing
• https://docs.n8n.io/n8n-community-license/sustainable-use-license
• https://github.com/n8n-io/n8n/blob/master/LICENSE.md
• https://en.wikipedia.org/wiki/N8n
• https://theaiinsider.tech/2025/10/24/n8n-secures-180m-series-c-at-2-5b-valuation-to-scale-ai-orchestration-and-automation-platform
• https://heise.de/en/news/Berlin-based-AI-startup-n8n-reaches-2-5-billion-dollar-valuation-10748098.html
• https://startupdaily.news/2026/05/13/n8n-s-valuation-doubles-to-5-2bn-following-sap-strategic-investment
• https://valueaddvc.com/blog/n8n-valuation-2026-5-2b-40m-arr-and-how-it-compares-to-zapier-and-make
• https://trendingtopics.eu/sap-bets-big-on-ai-invests-in-n8n-at-a-5-2-billion-valuation
• https://seedtable.com/companies/n8n/funding-rounds/series-c-2025-10
• https://docs.n8n.io/