Self-hosting n8n and AI automation on an Indian VM
Automation SaaS is priced per execution. Self-hosting is priced per month. The crossover arrives faster than most teams expect, and for anything touching personal data it arrives immediately, because a workflow that posts customer records to a third-party runner is a data-residency problem before it is a cost problem.
What the workloads need
Workflow runners are memory-bound far more than CPU-bound. A node that waits on an HTTP call costs almost nothing; a node that holds a 50,000-row payload in memory while it transforms it costs a lot.
| Workload | What drives the size |
|---|---|
| n8n, a few scheduled flows | Baseline runtime plus the largest single payload you transform |
| n8n with queue mode and workers | One runtime per worker, so memory scales with concurrency |
| A self-hosted vector store | The index, which wants to sit in RAM to stay fast |
| An LLM agent calling hosted models | Small — the model runs elsewhere; you are paying for orchestration |
| An LLM running locally | Dominated by the model weights, and wants a GPU, not a VM |
The last two are worth separating properly. "AI automation" is usually orchestration: your workflow calls a hosted model and waits. That is a small, cheap, network-bound workload and a modest VM runs it comfortably. Running the model yourself is a different purchase entirely.
The catalogue
| Tier | Plan | vCPU | RAM | Disk | Monthly |
|---|---|---|---|---|---|
| General Purpose | GP Small | 2 | 4 GB | 30 GB | ₹1,860/mo |
| General Purpose | GP Medium | 2 | 8 GB | 30 GB | ₹3,410/mo |
| General Purpose | GP Large | 4 | 16 GB | 30 GB | ₹6,500/mo |
| General Purpose | GP XL | 8 | 32 GB | 30 GB | ₹12,530/mo |
| Compute Cloud | Compute 4 | 2 | 4 GB | 30 GB | ₹3,020/mo |
| Compute Cloud | Compute 8 | 4 | 8 GB | 30 GB | ₹5,820/mo |
| Compute Cloud | Compute 16 | 8 | 16 GB | 30 GB | ₹11,330/mo |
| Compute Cloud | Compute 32 | 16 | 32 GB | 30 GB | ₹22,500/mo |
| Compute Cloud | Compute 64 | 24 | 64 GB | 30 GB | ₹33,000/mo |
| Budget Shared | Budget 4 | 2 | 4 GB | 30 GB | ₹1,230/mo |
| Budget Shared | Budget 8 | 2 | 8 GB | 30 GB | ₹2,160/mo |
| Budget Shared | Budget 16 | 4 | 16 GB | 30 GB | ₹3,840/mo |
| Starter / Dev VPS | Starter 4 | 2 | 4 GB | 30 GB | ₹1,150/mo |
| Starter / Dev VPS | Starter 8 | 2 | 8 GB | 30 GB | ₹2,080/mo |
The smallest instance is Starter 4 at ₹1,150/mo (₹2.53/hr), which is enough for n8n with a handful of scheduled workflows and no large payloads. Add memory before you add vCPU.
Why run it in India
Three reasons, in the order they usually matter:
- Data residency. Under the DPDP Act, a workflow that moves personal data through infrastructure outside Indian jurisdiction is your compliance problem, not your vendor's. Self-hosting in Noida keeps the whole path inside the country, by default rather than by configuration.
- No per-execution ceiling. A flow that runs every minute costs the same as one that runs every hour. Pricing stops shaping your architecture.
- Cost at steady state. Against AWS Mumbai on equivalent instances, our catalogue runs up to ~53% lower — the matched comparison for Compute 16 is in the table on our pricing page.
What you are responsible for
Self-hosting is not free of work, and the honest list is short but real: upgrades, backups, and the webhook endpoint being reachable. We cover the parts below the application — L3/L4 DDoS protection on every plan, snapshots, and the network — and you cover n8n itself.
If that trade is wrong for your team, the managed-app catalogue runs the same stack with the upgrades handled.
Start from self-hosting n8n or the VM catalogue.
