Is Hetzner Reliable Enough for Production? Yes — But Check Your Volumes

Yes — Hetzner owns its data centres in Falkenstein, Nuremberg and Helsinki, runs sites in Ashburn, Hillsboro and Singapore, and states its own performance promise plainly:

“Our performance promise: 99.9% uptime and 24/7 support”

99.9% allows about 8 hours 45 minutes of downtime a year. That’s a normal figure for this market and it’s not the interesting part.

The interesting part is that almost nobody who loses data on Hetzner loses it to a hardware failure. They lose it to a default.

The default that actually bites: backups are opt-in, and exclude Volumes

Two facts from Hetzner’s own documentation, and the second is the one people don’t know.

Backups are opt-in and cost extra. The control panel button is literally “Enable & Buy now”. A server you launched and never configured has no backups at all — daily backups exist, with seven slots per server, but only once you’ve turned them on and paid.

And here’s the one that ruins weekends:

“Backups and Snapshots do not include any Volumes which are attached to your server.”

Read that again if you keep your database on a Volume. The standard advice for production is to put your data on an attached Volume so it survives a server rebuild — and that’s exactly the storage that Hetzner’s server backups don’t cover.

So the failure mode is: you did the responsible thing, put your database on a Volume, enabled backups, and your database is not in them.

This is documented, not hidden. But it inverts the intuition — the more carefully you architected your storage, the less your backups cover. If you take one thing from this article, make it this: verify what’s actually inside your backup, don’t assume.

What the status page shows right now

We read Hetzner’s status page on 18 August 2026. The banner said:

“All systems are operational.”

Underneath it, these notices were open:

The first row is the one to weigh. Limited availability of cloud instances has been an open notice for roughly seven and a half weeks — and Hetzner’s own cloud page confirms it, showing its Shared Resources line as “Currently not available.”

That’s a reliability question of a different kind. Not “will my server go down” but “will I be able to get a server, or scale, or rebuild in the region I need?” For production planning that matters at least as much, and it’s connected to the 2026 repricing.

The Object Storage timeout notice has been open since January — seven months. If Object Storage is in your critical path, that’s a documented known issue, not a rumour.

And two regional Object Storage degradations — in Nuremberg and Helsinki — were resolved the day before we looked. Real, recent, region-scoped. Same pattern as DigitalOcean’s status page: the summary line stays green while specific components have real problems. Read the notices, not the banner.

The maintenance is scheduled, published, and long

This is genuinely a point in Hetzner’s favour, and it’s also a production constraint people miss.

Upcoming maintenance published at the time we looked:

• Emergency power system at HEL1-DC3 — a 12-hour window, 05:00 to 17:00 UTC
• Load balancer work in Singapore, Hillsboro, Ashburn and Helsinki
• Backbone maintenance between Nuremberg and Frankfurt, late August
• Backbone maintenance between Frankfurt and Helsinki, early September

Publishing a 12-hour maintenance window weeks ahead is good practice. It’s also a reminder that you need to be reading these, because a backbone maintenance between two sites is exactly the event that partitions a two-region setup you thought was redundant.

Support: the recovery time caveat

Telephone support for SSL and domain matters has been restricted since April 2024, per an open notice on the status page, and processing times for domain and billing inquiries have carried a notice since April 2026.

The practical implication for production: your recovery from an account-level or billing-level problem is measured in hours through a ticket queue, not minutes on a phone. For a server-level fault you have the panel, the API and the console, which is what you’ll actually use. But an account lockout or a payment failure is a slower path than at a US hyperscaler.

Budget accordingly: keep payment details current, and don’t let a card expire silently.

The production-ready checklist

Everything here follows from the verified facts above:

1. Turn backups on, then verify what’s in them. They’re opt-in and paid. Then confirm your Volumes are covered — they aren’t, by default. Back up Volume data separately, at the application level.

2. Do a restore drill before you need one. An untested backup is a belief, not a backup. Restore into a fresh server and confirm the data is actually there.

3. Put DNS and monitoring at a different provider. Not because Hetzner is unreliable, but because monitoring that shares infrastructure with the thing it monitors tells you nothing during an incident.

4. Read the planned maintenance list. Particularly backbone windows if you span regions.

5. Assume capacity isn’t guaranteed. With cloud instance availability limited since June, “I’ll just spin up another one in that region” is not currently a safe assumption. If your recovery plan depends on provisioning on demand, test that it works.

6. Keep billing current. The slowest support path is the account-level one.

What to run it on

Hetzner — 99.9% uptime promise, own data centres, and the clearest product labelling in the market — its dedicated CCX line is sold explicitly for “constantly high CPU usage” while the shared line is not. Production-ready for stateless and self-hosted workloads if you do the six things above. Two live caveats: shared cloud instances are currently limited, and backups exclude Volumes.

DigitalOcean — $24 for 4GB, more expensive, with the most granular status reporting of the three (256 components with per-region breakdown and a public API) and the best documentation in the category. The pick if you want incident visibility you can query programmatically, or if you need capacity you can rely on provisioning.

Hostinger — $11.99 renewal for 4GB, one-click templates, the least infrastructure to reason about. Reasonable for smaller production workloads, though its terms carry no support commitment, so plan to be self-sufficient.

The honest summary: all three are reliable enough for most production workloads. What separates them for production isn’t uptime — it’s what happens on the bad day, and that’s mostly determined by choices you make on day one.

How we checked this

The uptime promise and infrastructure details are from Hetzner’s own cloud page, read on 18 August 2026 — the “Our performance promise: 99.9% uptime and 24/7 support” line, the data centre locations, and the “Currently not available” label on its Shared Resources line.

The backup facts are from Hetzner’s own documentation, read the same day: daily backups with seven slots per server, snapshots taken manually with a default limit of 30 across all projects, the “Enable & Buy now” opt-in, and the explicit statement that “Backups and Snapshots do not include any Volumes which are attached to your server.”

The status notices, their start dates, the planned maintenance windows and the two resolved Object Storage degradations in Nuremberg and Helsinki are all read from status.hetzner.com on 18 August 2026.

A limitation worth stating: Hetzner’s status page has no public API, unlike DigitalOcean’s. So we could not do the quantified incident analysis we did there — no counts, no impact distribution, no historical percentages. What’s above is what the page displayed on one day, and it is not a reliability measurement. We have not tracked Hetzner’s uptime over any period, and nothing here should be read as verifying or disputing the 99.9% figure.

We also did not find a contractual SLA with service credits. The 99.9% figure is a performance promise on a marketing page; we did not locate a document specifying compensation for missing it, and absence from what we read isn’t proof none exists. If an SLA with credits matters to your business, ask for it in writing before committing — that’s a real difference from US hyperscalers and we can’t resolve it for you.

On backup pricing: we confirmed backups are opt-in and paid via the “Enable & Buy now” control. We did not verify the current surcharge percentage, so there’s no figure here.

On the support caveat: the telephone restriction and processing-time notices are quoted from the status page. We have not contacted Hetzner support, opened a ticket, or measured a response time, so the “hours not minutes” characterisation is inference from those notices rather than something we observed.

What we did not do: hold a Hetzner account, run a server, trigger a failure, restore a backup, or experience any of the incidents described.

The host links above are affiliate links. The most valuable advice in this article — check whether your Volumes are in your backups — costs nothing and applies whether or not you use any of these providers.

FAQ

Is Hetzner reliable enough for production?

Yes for most workloads. It owns its data centres and promises 99.9% uptime. The risks that actually materialise are configuration defaults, not hardware — chiefly that backups are opt-in and exclude attached Volumes.

Does Hetzner back up my server automatically?

No. Backups are opt-in and paid — the control is labelled “Enable & Buy now”. A server you never configured has none.

Do Hetzner backups include my Volumes?

No. Its documentation states plainly that backups and snapshots do not include any Volume attached to your server. If your database lives on a Volume, back it up separately.

What is Hetzner’s SLA?

Its cloud page promises 99.9% uptime and 24/7 support — roughly 8 hours 45 minutes of permitted downtime per year. We did not find a contractual document specifying service credits, so ask in writing if that matters.

Why can’t I create a Hetzner cloud server?

A “Limited availability of cloud instances” notice has been open on its status page since 26 June 2026, and its own cloud page shows the Shared Resources line as currently not available.

Is Hetzner good for business use?

For stateless and self-hosted workloads, yes. Weigh two things: capacity has been constrained since June, and account-level support runs through a ticket queue rather than a phone.

What should I set up on day one?

Backups enabled and verified, Volume data backed up separately, a tested restore, DNS and monitoring at a different provider, and the planned maintenance list on your reading list.