Hetzner Datacenter Locations and Latency: What to Expect

Hetzner Datacenter Locations and Latency: What to Expect

Hetzner Cloud runs in six locations across four network zones: Falkenstein and Nuremberg in Germany, Helsinki in Finland, Ashburn, Virginia and Hillsboro, Oregon in the USA, and Singapore. Hetzner owns and operates its three European data center parks and uses colocation space in the USA and Singapore.

Latency depends mainly on physical distance and the network path between your users and the chosen site. The closest site to your users is usually the best starting point, but the rest of this article shows which products each site supports and how to test the connection before you commit.

Where Hetzner's Datacenters Are

Hetzner's own datacenter page lists the same six sites: Nuremberg, Falkenstein, Helsinki, Ashburn (Virginia), Hillsboro (Oregon) and Singapore. Cloud servers are available in Germany, Finland, Singapore and the USA.

The practical split matters. Nuremberg, Falkenstein and Helsinki are Hetzner-built parks, while Ashburn, Hillsboro and Singapore operate from colocation space. That difference does not automatically change performance for your workload, but it does mean the European sites are the ones Hetzner runs end to end.

Network Zones

Hetzner groups the six locations into four network zones. The European sites share the eu-central zone, while the US sites are split into us-east and us-west, and Singapore sits in ap-southeast. Network zone membership drives pricing and some network features, so it is worth knowing before you pick a site.

Product Availability by Location

Not every product is offered everywhere. Ashburn, Hillsboro and Singapore are cloud-only locations, which means you cannot order dedicated root servers through a Hetzner Robot account there. If your plan depends on a dedicated server, your realistic choices are the three European sites.

ARM-based CAX cloud server types follow a similar pattern. They are offered only in Falkenstein, Helsinki and Nuremberg, while the other three locations provide x86 types only. Hetzner's own locations table confirms that Cloud Shared AMPERE (ARM) is available only in Falkenstein, Nuremberg and Helsinki.

Location Country Network Zone Dedicated Servers ARM (CAX) Cloud
Falkenstein Germany eu-central Yes Yes
Nuremberg Germany eu-central Yes Yes
Helsinki Finland eu-central Yes Yes
Ashburn, VA USA us-east No No
Hillsboro, OR USA us-west No No
Singapore Singapore ap-southeast No No

Two operational details apply across all sites. Cloud instances may occasionally be unavailable for ordering at any location, so stock is not guaranteed. Cloud limits, by contrast, are the same for all locations. DDoS protection is also consistent: Hetzner states its DDoS system is the same in Ashburn as in its other data centers. Support for Cloud servers in Europe, the USA and Singapore comes from Hetzner's own technicians in Germany.

Why Latency Differs Between Locations

Three factors shape the latency you actually experience: distance, routing and peering.

Distance sets the floor. Data travels through fiber at a meaningful fraction of the speed of light, so a user in Frankfurt will always reach Falkenstein faster than a user in Sydney reaches Singapore. No amount of network engineering removes that gap.

Routing determines how close you get to that floor. A short physical distance with a convoluted path can perform worse than a longer distance over a direct route. This is why the quality of a provider's interconnection with your users' networks matters as much as the map.

Peering is where Hetzner's network design becomes relevant. Hetzner describes its connections as multi-redundant and linked to important internet exchanges. Its published peering tables show heavy concentration in Europe, with DE-CIX Frankfurt listed at 2800 Gbit/s as the largest single entry. Amsterdam, Helsinki, London and Paris also appear among the European exchange points.

Private peerings reinforce the European advantage. Hetzner lists direct connections with Cloudflare across Amsterdam, Frankfurt, Helsinki, London and Paris, including 1200 Gbit/s at Frankfurt. Google peerings are listed at Amsterdam, Frankfurt, Helsinki and Stockholm, with Frankfurt at 1400 Gbit/s. Transit providers including Telia, Lumen, NTT, Cogent, Deutsche Telekom and TATA appear across the site list, and Telia alone is listed at 900 Gbit/s at Nuremberg.

The US and Singapore sites have their own exchange presence but a thinner published footprint. MegaIX Ashburn and NWAX Hillsboro appear among the US peering points, and Equinix Singapore and SGIX appear for Singapore. Treat these tables as a snapshot: peering relationships change, and the published figures describe capacity, not the latency any individual user will see.

Cross-Zone Traffic and Pricing

Network zone boundaries have billing consequences. Prices are identical across all eu-central locations, while us-east, us-west and ap-southeast are priced separately. Traffic between two different network zones is billed like normal internet traffic. A design that splits load between Helsinki and Ashburn therefore pays cross-zone rates for that traffic rather than enjoying internal transfer.

Cloud Networks add a structural constraint. All locations within a Network must belong to the same network zone as the subnet, and all subnets in a Network must come from the same network zone. You cannot stretch a single private Network across continents, so multi-region designs need separate Networks and explicit routing between them.

How to Measure Latency Before You Deploy

Published peering tables tell you about capacity, not about the round-trip time your users will get. Measure it yourself.

  1. Ping the target site. Send ICMP echo requests from a machine in your users' region to a server or test endpoint at the candidate location. Repeat over several minutes and at different times of day, because congestion varies.
  2. Trace the route. Use traceroute or mtr to see which networks the packets cross. A route that detours through an unexpected transit provider explains latency that distance alone does not.
  3. Test from both ends. Run measurements from your users' network toward Hetzner and from a Hetzner server back toward your users. Asymmetric routing is common and only shows up when you check both directions.
  4. Repeat from multiple user regions. If your audience is spread out, test from each significant region rather than extrapolating from one city.
  5. Test the actual application path. ICMP latency is a starting signal. HTTP request timing, TLS handshake time and database round trips ultimately determine perceived speed.

Public looking-glass servers and third-party measurement networks can supplement your own tests, especially when you cannot place a probe in every region. Whatever method you use, run the test before you commit and again after deployment, since routing can change.

Choosing a Location

A simple decision order works for most deployments: pick the closest site to your users, then check product availability, then confirm pricing, then verify with a measurement.

For a European audience, Falkenstein, Nuremberg and Helsinki all sit in eu-central and share the same pricing, so the choice often comes down to server type and stock rather than cost. If you need ARM CAX instances, only those three European sites offer them. If you need a dedicated root server, the same three sites are your options.

For a North American audience, Ashburn serves the East Coast and Hillsboro the West Coast, both cloud-only. For audiences in Southeast Asia, Singapore is the regional option, also cloud-only. In each case, confirm that the specific server type you want is orderable at that location, since instances can be temporarily unavailable.

Redundancy changes the calculus. Because a Cloud Network cannot span network zones and cross-zone traffic is billed as normal internet traffic, a multi-region failover setup costs more and requires more architecture work than a single-zone deployment. Decide whether that resilience is worth the added complexity and transfer cost before you build for it.

Limitations to Keep in Mind

Several things this article cannot tell you. Hetzner does not publish latency figures or latency guarantees for specific user regions, so no source here supports a claim like "Frankfurt to Ashburn takes X milliseconds." Your own measurements are the only reliable input for that decision.

Product availability and pricing also change. The official Cloud locations documentation carries an older last-changed date, so verify current per-location availability and pricing in the Hetzner console before you finalize a design. The peering and transit figures on the datacenter page are a snapshot of a moving network and should be read as indicators of where Hetzner concentrates capacity, not as fixed performance numbers.

Finally, one billing detail catches people out: Hetzner bills all servers that have finished the creation process, including powered-off servers, until they are deleted. Deleting a test server is the only way to stop paying for it, which is worth remembering when you spin up instances in several locations to compare latency.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top