Best VPS for n8n in 2026: picks by use case from our load tests

The best VPS for n8n, for most people, is a 4 vCPU, 8 GB shared server. In our load tests the Hetzner CX33 (10.27 € a month) ran 31.93 webhook executions per second with n8n 2.46.6 and PostgreSQL, more than twice the 15.04 of a 2 vCPU server, and twice the cores again added only 13% more. For a hobby setup, the 2 vCPU CX23 at 6.64 € is enough.

Why this matters now: Hetzner repriced its cloud plans in 2026, promo VPS prices often double at renewal, and n8n 2.x changed how Code nodes run. Most "best VPS for n8n" lists rank providers by specs and affiliate payouts. We rented five servers, ran the same n8n stack and workflow on each, and measured. This page is part of our n8n hosting section.

Tested in the Lab Hetzner CX23, CX33, CX43, CAX11, CAX21 · n8n 2.46.6 · Nov 2026

How we picked

We ran n8n 2.46.6 with PostgreSQL 16 in Docker Compose, the same stack we run in production, on five Hetzner Cloud servers in Helsinki on 5 November 2026. The test workflow is a webhook that runs a Code node and returns the result. ApacheBench fired 300 to 2000 requests at 1, 10, 25 and 50 concurrent. We recorded executions per second, median and p95 latency, failed requests and the n8n container's peak RAM. Executions/s means completed workflow runs per second; p95 is the response time 95% of requests beat. The method and the script are in our n8n webhook performance report.

Two caveats up front. The load came from the same server over localhost, so these are ceilings of the n8n stack, not internet latency. And the workflow is light: real workflows that call APIs will run fewer executions per second on every server.

The comparison

"€ per execution/s" is the monthly price divided by the best measured executions per second: what you pay each month for each unit of webhook capacity. Lower is better. Prices are gross (incl. VAT) from the Hetzner Cloud API, checked 5 November 2026. Every server answered every request: 0 failed across all runs.

The full table with every concurrency level:

Measured by AI Hosting Lab on 2026-11-05 · 5 servers · method · raw data

The pick by use case

Hobby and personal automation: Hetzner CX23

A few scheduled workflows, some webhooks from forms or chat apps, a few hundred to a few thousand executions a day. That is a tiny fraction of the 15.04 executions/s the CX23 reached. RAM is not the limit either: on our production CX23 the whole server used 1036 MiB idle with n8n and PostgreSQL running, out of 4 GB. A single request took a median 104 ms.

Small business: Hetzner CX33

Client-facing webhooks, a shop or CRM sending events, AI agent workflows that should answer fast during busy minutes. Here p95 under bursts matters. At 10 concurrent requests the CX33 kept p95 at 438 ms, against 893 ms on the CX23. Its 8 GB also leaves room for a growing execution history in PostgreSQL, a second app, or a small local model. And it was the cheapest per unit of throughput of all five.

Heavy webhooks: CX33 now, queue mode next

If you expect sustained traffic above roughly 30 executions/s of a light workflow, a bigger server is not the answer we measured. The CX43 has twice the cores of the CX33 and reached 36.14 executions/s, 13% more for 88% more money. In regular mode one n8n main process receives webhooks and runs executions, so it cannot use all eight cores. The n8n docs describe queue mode for this: Redis plus worker processes, and optional webhook processors behind a load balancer. We have not tested queue mode yet; it is scheduled. Until then our advice for heavy webhooks is a CX33 and a plan to move to queue mode, not a CX43.

Each option in detail

Hetzner CX33: 4 vCPU, 8 GB, 10.27 € a month

• Best for: business n8n with real webhook traffic, and anyone who wants headroom.
• Main drawback: shared vCPU. A busy neighbour on the same host can slow you down; we measured 0.0 % CPU steal on test day, but that can change.
• Skip it if: you only run personal automations. The CX23 does that for 3.63 € less.

The CX33 doubled the CX23's throughput (31.93 vs 15.04) and halved latency under load (p95 1846 ms vs 3954 ms at 50 concurrent). It is the server we would rent for a client today. See Hetzner Cloud plans.

Hetzner CX23: 2 vCPU, 4 GB, 6.64 € a month

• Best for: personal n8n, side projects, learning.
• Main drawback: older Intel Skylake CPU, and the ceiling arrives early. Above 10 concurrent requests throughput stayed near 15 and latency climbed: median 3445 ms at 50 concurrent.
• Skip it if: webhook callers have short timeouts and arrive in bursts.

We run our own production n8n on a CX23. n8n idled at 379.7 MiB and PostgreSQL at 69.6 MiB there, so 4 GB is comfortable for n8n alone. How we set it up is in install n8n on Hetzner.

Hetzner CX43: 8 vCPU, 16 GB, 19.35 € a month

• Best for: n8n plus other heavy apps on one server, such as Ollama or Supabase.
• Main drawback: for n8n alone, the extra cores added little throughput in regular mode.
• Skip it if: n8n is the only thing on the server.

It had the best latency of all (p95 341 ms at 10 concurrent) and the highest peak (36.14 executions/s), but the worst price per unit of throughput (0.54 €).

Hetzner CAX11 and CAX21 (ARM)

• Best for: stacks that are already on ARM.
• Main drawback: more expensive and slower for n8n than the x86 server of the same size.
• Skip it if: you are choosing a server for n8n today.

The official n8n image ran on both without changes. But the CAX11 peaked at 13.9 executions/s against 15.04 on the cheaper CX23, and the CAX21 at 25.77 against 31.93 on the CX33, which costs 2.42 € less. Details in n8n on an ARM server.

Other providers: prices only, not yet tested

Prices from the official pricing pages, opened on 5 November 2026. Currencies as shown by the provider.

What this means against our data, without inventing results:

• Hostinger KVM 2 gives 2 vCPU and 8 GB. Our tests found n8n needs far less than 8 GB, and 2 vCPU was the throughput limit on Hetzner. At renewal it costs $14.99, more than the 4 vCPU CX33. Hostinger VPS plans.
• DigitalOcean has no promo trap, but 2 vCPU and 4 GiB costs $24 a month, several times the CX23. You pay for its regions, managed services and documentation, not for n8n speed. DigitalOcean Droplets.
• InterServer is cheap on RAM, but 2 slices bring 1 core. We have not measured n8n on 1 vCPU; on our servers throughput grew with vCPU count up to four, so expect a lower ceiling than the 2 vCPU results above. InterServer VPS.
• Vultr: its pricing page refused our automated check on 5 November 2026, so we show no price rather than a remembered one. Vultr pricing.

More options outside Hetzner are in Hetzner alternatives.

And n8n Cloud?

n8n Cloud Starter is 20 € a month billed annually with 2,500 executions and 5 concurrent executions; Pro is 50 € a month billed annually with 10,000 executions (n8n pricing page, 5 November 2026). You get updates, backups and support handled for you. A self-hosted CX23 has no execution limit, but you do the maintenance. The trade-offs are in n8n Cloud vs self-hosted. n8n Cloud plans.

All Hetzner prices

From the Hetzner Cloud API, location hel1, checked 2026-11-05. "Incl. VAT" is what our Czech (EU consumer) account pays; businesses and buyers outside the EU usually pay the price excl. VAT. Primary IPv4 is billed separately.

The CX line is shared x86, the CAX line shared ARM, the CCX line dedicated vCPU. Hetzner backups cost a flat 20% of the server price per month (Hetzner billing FAQ), so a CX33 with backups is 12.32 €, a CX23 7.97 €.

What to check before you buy any VPS for n8n

• vCPU count before RAM. Going from 2 to 4 vCPU doubled our throughput. The extra RAM on the bigger servers went unused by n8n: its container never passed 1049.6 MiB.
• Renewal price, not promo price. A 24-month prepay that renews at nearly double changes the ranking.
• PostgreSQL, not SQLite, if you might ever need queue mode: n8n says a distributed setup on SQLite isn't supported.
• Disk: our install used 4.7 GB of 38 GB. Any plan above 40 GB is fine at the start; execution history grows with use, so keep pruning on.
• Commissions: none of the provider links on this page pays us a commission at the time of writing, and the ranking above comes only from our measurements and the listed prices.

Next step: if you picked a Hetzner server, follow our timed setup in install n8n on Hetzner.