I’ve been running n8n self-hosted for about two years now, and honestly, I’m starting to wonder if we’re actually saving money or just fooling ourselves. On paper, self-hosted seems cheap—no recurring vendor fees, right? But when I actually sit down and calculate everything, the picture gets messy.
There’s the obvious stuff: server costs, database maintenance, infrastructure scaling when workflows spike. Then there’s the hidden layer—DevOps time. We have someone spending maybe 30-40% of their week just keeping the lights on. Security patches, backups, monitoring, troubleshooting when things break at 2 AM.
But here’s what really got my attention recently. We’ve been bolting on different AI model subscriptions as we expand what we’re doing. OpenAI key here, Anthropic subscription there. Each one is its own billing relationship, its own authentication nightmare, its own cost center that nobody’s really tracking.
I’ve been reading that some platforms now let you access 300+ models under one subscription. And from what I can tell, the execution-based pricing model means you’re only paying for actual runtime—not per workflow or per task like Zapier. One case study I found mentioned they’re seeing 7x cost savings on certain high-volume operations compared to other platforms.
I’m not ready to flip the switch yet, but I’m genuinely curious: has anyone actually done this migration math? Not the sexy pitch deck version, but the real spreadsheet where you factor in all the DevOps overhead, the scattered AI subscriptions, the infrastructure costs creeping up every year? What did your actual TCO look like on both sides?
I went through this exact calculation about six months ago. The shocking part wasn’t the compute costs—we’d already budgeted for that. It was realizing how much we were spending on keeping people busy with operational work instead of building new automations.
With self-hosted n8n, we had a full-time engineer just managing the platform. Database optimization, dependency updates, scaling when load spiked. That person could’ve been building features instead.
When I switched to execution-based pricing platform, the math flipped. We eliminated that person from platform maintenance entirely. That alone justified the move, even if monthly platform costs were higher.
The other thing that surprised me was the AI subscription sprawl. We had four separate API keys across our team. Consolidating those into one subscription actually saved us money and made governance way simpler. Audit was happy because we could finally track spending by department.
One thing to factor in that often gets overlooked: your cost per workflow execution actually matters way more than your platform cost. We were getting penny-wise and pound-foolish with n8n self-hosted.
If you’re running high-volume workflows—like data processing, email generation, API polling—the infrastructure costs multiply fast. A cloud-native platform that charges based on execution time can actually be cheaper because you’re not maintaining unused capacity.
We calculated our average cost per 100 workflow runs. Self-hosted was around $1.20 when you factored in server overhead. Managed platform with execution pricing came out to $0.45. It looked small until you multiply it across millions of runs a month.
The TCO comparison really depends on your use case and team size. For us, self-hosted n8n made sense initially because we were building custom workflows that needed heavy customization. But as we scaled and added more users, the operational burden grew faster than we expected.
What changed the equation was when we started experimenting with AI agents and needed access to multiple model providers. Managing separate subscriptions and API keys became a compliance nightmare. We were essentially paying for platform maintenance, infrastructure, AI subscriptions, and custom development all separately.
The unified model we looked at was consolidating all of that—less to manage, predictable costs, built-in scaling. It took three months to migrate our workflows, but the labor savings in the first year alone covered the platform costs several times over.
When evaluating total cost of ownership, you need to account for several dimensions beyond just platform fees. Infrastructure costs scale unevenly with self-hosted solutions—you’re paying for peak capacity most of the time. Human overhead for platform management, security patching, and compliance monitoring typically runs 1-2 full-time equivalents in enterprise settings.
AI model fragmentation introduces another cost layer. Maintaining multiple vendor relationships, separate billing streams, and authentication mechanisms creates operational friction and accounting complexity. A unified subscription model eliminates this friction.
For high-execution-volume scenarios, time-based pricing models significantly outperform per-task or per-workflow pricing because you can handle substantial data transformations and multiple API calls within a single execution window. The cost per operation drops substantially compared to usage-based competitors.
Self-hosted costs are deceptive. Server fees are just the tip. DevOps labor, security updates, scaling—that’s where money drains. Most teams undercalculate this by 40-50%.
I did this exact analysis last year. Here’s what nobody mentions: the real money drain with self-hosted is the invisible operational overhead plus managing multiple AI subscriptions separately.
We were paying for servers we didn’t always need, keeping someone on-call for infrastructure issues, and juggling OpenAI, Anthropic, and other API keys across different projects. It was a cost management nightmare.
When we consolidated to an execution-based platform with unified AI model access, the change was dramatic. One subscription covers everything. No more scattered AI keys to track. Scaling is automatic so you’re never paying for idle capacity. We actually spend less per month and get way more capacity.
The migration took a week because our workflows translated cleanly. The real win was reclaiming that DevOps overhead and turning it into actual automation work.
If you want to do this math properly, calculate your platform costs plus infrastructure costs plus 0.5-1 FTE of operations labor. That’s your real self-hosted TCO. Compare it to a managed platform and the picture usually clears up fast.