Hetzner Cloud in Late 2026: Plans and Prices for Self-Hosting n8n, Windmill, Twenty and Ollama

Hetzner raised cloud prices on 15 June 2026, and in October 2026 its cheapest shared plans are listed as not available. What the current lineup costs, which plan fits n8n, Windmill, Twenty and Ollama, and what our earlier self-hosting guides assumed at the time.

Why a follow-up instead of a rewrite

Several of our self-hosting guides name a specific Hetzner server: the CX23 in our n8n tutorial on Hetzner Cloud, the CX line in both Windmill guides, a CX53 in our CRM self-hosting post, the CCX42 in the local AI assistant guide and the CCX62 in the multi-tenant Docker stack. Those posts describe the setups as they were when we wrote them, and they stay that way.

Hetzner's catalog has moved since. In October 2025 the cloud lineup in Europe and Singapore was reorganized by hardware generation rather than hardware type. On 15 June 2026 new prices took effect for new orders and rescales. And in October 2026 every Cost-Optimized plan on Hetzner's product page carries the label "not available". This post collects the current state in one place, so the older guides do not have to be rewritten every time a price list changes.

All prices below are monthly caps for Germany and Finland, excluding VAT, as listed on hetzner.com in October 2026. They include a primary IPv4 address, which Hetzner bills separately at €0.50 per month - the price adjustment table on docs.hetzner.com lists the same plans without it.

The three cloud lines

Since October 2025, Hetzner sells its European cloud servers in three lines. The distinction matters more than the plan names, because it decides what kind of CPU time you actually get.

Cost-Optimized (CX, x86, and CAX, Arm64) runs on older hardware generations. CX Gen3 servers can land on Intel or AMD, depending on availability. Hetzner describes the line as suited to test environments and to "workloads that can handle variable CPU performance", and notes that the number of available servers is limited by design.

Plan	vCPU	RAM	Disk	Price
CX23	2	4 GB	40 GB	€5.99
CX33	4	8 GB	80 GB	€8.99
CX43	8	16 GB	160 GB	€16.49
CX53	16	32 GB	320 GB	€29.99
CAX11	2	4 GB	40 GB	€6.49
CAX21	4	8 GB	80 GB	€10.99
CAX31	8	16 GB	160 GB	€21.49
CAX41	16	32 GB	320 GB	€41.49

Regular Performance (CPX Gen2) is shared as well, but runs on the more recent AMD generations available in a location. Two small plans were added this year: the CPX12 in July and the CPX02 in October 2026.

Plan	vCPU	RAM	Disk	Price
CPX02	1	1 GB	20 GB	€6.49
CPX12	1	2 GB	40 GB	€11.99
CPX22	2	4 GB	80 GB	€19.99
CPX32	4	8 GB	160 GB	€35.99
CPX42	8	16 GB	320 GB	€69.99
CPX52	12	24 GB	480 GB	€100.99
CPX62	16	32 GB	640 GB	€130.49

General Purpose (CCX) gives you dedicated vCPUs on the most recent hardware generation in a location.

Plan	vCPU	RAM	Disk	Price
CCX13	2	8 GB	80 GB	€43.49
CCX23	4	16 GB	160 GB	€86.49
CCX33	8	32 GB	240 GB	€138.99
CCX43	16	64 GB	360 GB	€276.49
CCX53	32	128 GB	600 GB	€533.99
CCX63	48	192 GB	960 GB	€853.99

The CCX42 and CCX62 named in our 2025 guides are not part of this list; the CCX63 is the current 48-vCPU, 192 GB plan.

What the June 2026 price adjustment changed

The adjustment was not uniform, and that is the salient point for anyone planning a self-hosted stack. Hetzner's own table shows the Cost-Optimized plans rising moderately - the CX23 went from €3.99 to €5.49 before IPv4, the CX43 from €11.99 to €15.99. The Regular Performance and General Purpose lines rose far more: the CPX22 from €7.99 to €19.49, the CCX13 from €15.99 to €42.99, the CCX63 from €374.49 to €853.49. Hetzner attributes the increase to the "massive increase in procurement costs" for hardware.

Two rules from Hetzner's documentation decide whether this affects servers you already run. Existing servers keep their previous price. But rescaling a server - up or down - moves it to the current pricing, and so does restoring a deleted server or transferring it to a project with a different currency. Rebuilding a server, or moving it to a project with the same currency, does not.

In reality, this turns a routine operation into a pricing decision. A CX-era server that is rescaled from 8 GB to 16 GB in late 2026 is no longer billed at its old rate, and if the Cost-Optimized line is not available at that moment, the comparable plan is a CPX42 at €69.99 rather than a CX43 at €16.49.

Matching the plans to self-hosted tools

The vendors' own sizing guidance is the most reliable starting point, so it comes first here.

n8n. The Hetzner guide in n8n's documentation says the CPX11 "is enough" for most usage levels, and recommends at least 4 GB RAM and 2 vCPU once the n8n Assistant sandbox is added, because it runs Docker-in-Docker. The CPX11 is now only offered at Hetzner's US locations. In Germany and Finland, a CX23 meets the 4 GB / 2 vCPU line for €5.99 when it is available; otherwise the CPX22 does the same for €19.99.

Windmill. Windmill's self-hosting documentation gives a rule of thumb of one worker per vCPU with 1-2 GB of RAM each, on top of the server process and PostgreSQL. A setup with two workers and the bundled database fits a CX33 (4 vCPU, 8 GB) at €8.99, or a CPX32 at €35.99 as the available alternative.

Twenty. The Docker Compose guide for Twenty asks for at least 2 GB of RAM and warns that insufficient memory can cause processes to crash. On its own, Twenty is the lightest of the four. In practice it tends to share a server with other services - which is how we run it - so the deciding factor is the combined memory of everything on that host.

Ollama. Here the model, not the application, sets the size. Ollama's library lists qwen3:8b at 5.2 GB and qwen3:30b at 19 GB, and a model has to fit into memory with room left for the context and the rest of the system. On shared or dedicated CPU plans that points to 32 GB: a CX53 at €29.99 where available, a CPX62 at €130.49 or a CCX33 at €138.99. CPU inference works, but it is slow for larger models. For GPU inference, Hetzner's cloud has no GPU plans; the dedicated GEX45 with an NVIDIA RTX PRO 4000 Blackwell (24 GB VRAM) is listed at €214 per month plus a €209 setup fee.

A combined stack. n8n, Windmill and Twenty behind one reverse proxy on a single host - the pattern several of our guides describe - fits comfortably in 8 vCPU and 16 GB. That is a CX43 at €16.49, a CPX42 at €69.99, or a CCX23 (4 dedicated vCPU, 16 GB) at €86.49.

How we would plan a new server today

The problem is that the cheapest line is also the least predictable one. Cost-Optimized servers are excellent value when Hetzner has capacity, but a deployment plan that depends on them can stall on the day the server is ordered. Three practices help.

First, size for the next twelve months rather than the first week. Rescaling is cheap technically, but since June 2026 it also resets the price, so a slightly larger server at the start can be the more economical choice.

Second, keep the stack portable. A Docker Compose setup with external volumes and tested backups - as described in our disaster recovery routine for self-hosted services - can move between plan lines, or between providers, without a rebuild of the application layer.

Third, separate workloads by their CPU profile. Automation tools with bursty, low average load are a good fit for shared vCPUs. Sustained loads - a busy database, CPU inference, many concurrent workflow executions - are where dedicated CCX vCPUs earn their price.