Reproducible, not authoritative
How the decision is made.
The lab separates sourced product facts from your planning assumptions. It recommends an operating model, not a universally “best” vendor.
Inputs and outputs
The calculator uses workload, peak concurrency, control requirements, governance, operator capability, recovery evidence, price assumptions, and labor assumptions. It returns a provisional route, 12-month cash and labor totals, responsibility matrix, risks, and a proof plan.
No result is a performance benchmark, availability guarantee, security certification, quote, or professional advice.
VPS sizing gate
The VPS planner does not turn the Project2 fixture into a server recommendation. A passing result requires the visitor to identify the measurement as their own representative workload, cover the planned production concurrency and heaviest applicable data path, complete at least five clean repetitions, remain inside the selected memory budget, meet latency and error targets, restore a backup, prepare monitoring, and assign an operator.
The editable 60%, 70%, or 80% memory budget is a SmallBiz Auto test policy. It is not an official n8n minimum, a hosting-provider guarantee, or a substitute for a provider contract.
Cost model
n8n Cloud
annual cash = published monthly EUR price × editable EUR rate × 12
annual labor = internal oversight hours/month × hourly cost × 12
The tool maps up to 2,500 executions and 5 concurrent executions to Starter, and up to 10,000 and 20 to Pro, using the n8n pricing page checked on 2 September 2026. Workloads outside those public limits require your own Cloud quote.
Managed hosting
annual total = provider monthly quote × 12 + setup fee + your monthly hours × hourly cost × 12
The default is an assumption, not a vendor quote. “Managed” remains conditional until the contract identifies who owns n8n updates, database backup, restore testing, monitoring, and incidents.
Self-managed VPS
annual cash = (server + backup + monitoring) × 12
annual labor = setup hours × hourly cost + operations hours/month × hourly cost × 12
Cash and labor stay separate so a cheap server cannot hide an expensive operating burden.
Recommendation gates
- Governance requirements such as SSO, Git environments, or a formal SLA trigger an Enterprise or licensed self-hosted review before cost comparison.
- Root, custom package, or self-hosted data-control requirements exclude standard n8n Cloud.
- A self-managed VPS is not eligible when no Linux/Docker operator is assigned.
- A business-critical workload without demonstrated recovery favors reduced infrastructure ownership.
- Only after these gates does the engine choose the lowest 12-month cash-plus-labor assumption among eligible routes.
These are transparent planning rules. Change the assumptions and the output changes.
Why there is no universal RAM answer
Workflow payload size, item count, file handling, browser automation, code, AI calls, database behavior, and simultaneous executions can materially change resource use. The lab therefore asks for a representative load test and refuses to convert execution count into a fabricated memory requirement.
Actual-use lab method
A pinned local n8n 2.37.7 runtime on Node.js 24.19.0 imported a deterministic, synthetic workflow and executed 2,500 generated items five times. Each run had to produce exact item, amount, ID-sum, and bucket counts. A separate workflow was made to fail without external actions, corrected, re-imported under the same workflow identity, and executed successfully. The isolated SQLite database was then backed up, restored, and checked for matching workflow material.
Elapsed time covers the end-to-end CLI process, including startup and the task runner. The recorded working-set maximum covers the observed process tree on this machine. Neither value is a provider benchmark, capacity guarantee, or universal production-sizing rule. See the full lab boundary and downloadable safe fixture.
Evidence standard
- Current plan and feature facts use official vendor pages or documentation.
- First-party local measurements are labeled with their runtime, workload, machine-specific boundary, and exclusions.
- Vendor claims are attributed; they are not treated as independent benchmarks.
- Retrieval dates are visible on the main page.
- Unknown contract scope, quotes, conversion rates, and performance remain unknown until measured.
- Affiliate compensation never changes the calculation or route rules.
Known limits
The tool does not evaluate every n8n plan, host, jurisdiction, security requirement, or workload. It does not inspect your workflows. Prices, plan limits, and provider scope can change. Recheck the linked official sources and contract before purchase.