Hetzner Cloud: server limits, locations and why the price is what it is

Hetzner sits at a price point that makes people suspicious. A cloud server with dedicated resources and generous included traffic frequently costs a fraction of the equivalent from a hyperscaler, and the quality is not the problem people expect it to be.

The reasons are structural. Hetzner owns its data centres, operates in locations with cheap power and cheap land, sells a deliberately narrow product range, and does not fund a catalogue of two hundred managed services. You are buying servers and network, priced close to cost.

The server families

• CX. Shared Intel vCPU. The entry line, and entirely adequate for web servers and small services.
• CPX. Shared AMD vCPU, generally faster per core than the CX equivalent at a small premium.
• CAX. Shared Arm vCPU on Ampere. The best value in the catalogue where your software runs on ARM64, which today covers most Linux workloads and container images built for multiple architectures.
• CCX. Dedicated vCPU. No noisy neighbour effects, predictable performance, priced accordingly. This is what you move to when a shared instance stops being consistent.

Alongside cloud servers, Hetzner sells dedicated root servers, including an auction market for older hardware that is remarkable value if you can work with fixed specifications and a longer provisioning time.

Locations

The European footprint is the strength: sites in Germany and Finland, with Nuremberg, Falkenstein and Helsinki being the long standing ones. North American locations were added later, and there is Asian capacity too, but the pricing and availability are most attractive in Europe.

If your users are European, this is close to ideal: low latency, EU data residency by default, and prices that make redundancy affordable. If your users are elsewhere, check the regional pricing and the included traffic allowance for that location, because they differ.

Traffic

Hetzner includes a large monthly traffic allowance with each server, historically twenty terabytes on European locations, with lower allowances outside Europe. Overage is charged per terabyte at a rate that is a rounding error next to hyperscaler egress pricing.

For bandwidth heavy workloads this is often the single largest reason to be here. A media service that would generate a five figure egress bill on a hyperscaler can sit inside the included allowance. The comparison across providers is in data transfer costs.

The account level settings

Two properties belong to the account rather than to any server, and both decide what you can actually do.

The server limit. A cap on how many cloud servers the account may run at once. New accounts start low. Increases are granted on request once there is a payment history. If you are planning a Kubernetes cluster or a horizontally scaled service, this is the number to check first, because it applies across every project in the account.

Outbound SMTP. Port 25 is blocked by default on new accounts, as it is nearly everywhere. It is unblocked on request. If your application delivers mail directly rather than through an API, this needs resolving before anything works.

What you give up

Honest accounting matters here. Compared to AWS, Azure or Google Cloud you lose:

• The managed service catalogue. There is object storage, load balancing, volumes, firewalls and private networking, and that is close to the list. No managed Kafka, no managed data warehouse, no serverless functions platform.
• Global region coverage. Fine for Europe, thin elsewhere.
• Enterprise procurement machinery. Support is competent and direct rather than contractual.
• Deep identity and compliance tooling. Access control is simple, which is a virtue until it is not.

For a team that runs containers on servers and manages its own database, none of that is a loss. For a team that has built on managed services, the migration cost is the real number to weigh.

Who it suits

European workloads with a real bandwidth component, self managed infrastructure, cost sensitive projects, and anybody running Kubernetes on plain servers. It is also an excellent second provider for redundancy, precisely because it is cheap enough to keep a warm standby running.

The direct comparisons are in Hetzner vs DigitalOcean and Hetzner vs OVHcloud, which is the other large European operator with its own data centres.