Hetzner Object Storage vs. AWS S3 for EU backup stacks
Cost, rate limits, durability, and data residency: a practical side-by-side for EU backup stacks.
Resistro Cloud stores backups on Hetzner Object Storage in Nuremberg. For EU-hosted backup stacks, AWS S3 in Frankfurt is the obvious comparison point — the APIs are identical, the tooling is everywhere. This is a practical side-by-side based on operating constraints, cost, and one early scaling issue we had to design around.
Cost at 1 TB / month
Storage cost is what everyone looks at first, and it genuinely matters past 500 GB. Net of VAT. Hetzner row: price list of 19 Nov 2026 (6.49 € per TB of storage, 1 TB traffic included). AWS row: eu-central-1 list prices as of April 2026 — check current list prices before you decide.
| Hetzner Object Storage | AWS S3 (eu-central-1) | |
|---|---|---|
| 1 TB stored / month | ~6.49 € (incl. 1 TB traffic) | ~21 € |
| Egress 100 GB / month | 0 € (inside Hetzner) / low external | ~8 € |
| PUT/COPY/POST per million | ~0.55 € (approx) | ~4.40 € |
| GET per million | ~0.055 € | ~0.35 € |
For a typical mid-market backup workload — 200 GB of encrypted dumps per week, fetched rarely — the pure storage bill on AWS runs roughly 3–4× Hetzner. The request-priced APIs (PUT, LIST) push it further if your workload is chunky. A 200–500 GB backup stack moving from S3 eu-central-1 to Hetzner Object Storage typically saves 200–300 € per month at identical retention; the exact delta depends on how often you touch the data.
That said: AWS still wins at fractional-TB scale (below 50 GB) because of the free tier, and is usually cheaper than people expect for hot tiers you access frequently.
Rate limits — what is documented and what is not
This gotcha matters for anyone building multi-tenant, but the honest version is less tidy than a table:
- AWS S3: publishes numbers. 3,500 PUT/COPY/POST/DELETE and 5,500 GET per second per prefix, and the limit scales as you spread load over prefixes.
- Hetzner Object Storage: publishes no per-bucket request rate. What the Object Storage FAQ does state is that under high cluster load the NBG cluster applies temporary limits on upload speed and on the number of concurrent requests, and that the cluster answers HTTP 503 when they are exceeded. S3 clients without retry logic abort there.
So the practical difference is not “750 versus 3,500 per second”. It is: on AWS you can design against a published budget, on Hetzner you cannot — the client has to expect 503s and retry with backoff. We treat 503 on that path as a normal operational event, not as an error.
On key layout, both providers only reward spreading. AWS documents that request rate scales per prefix. Whether Hetzner partitions by prefix the same way is not documented, and we do not claim it. What we do: keys are spread over many prefixes instead of one hot prefix, because it costs nothing and keeps our layout from depending on one provider’s internals.
Correction, 2026-11-19: earlier versions of this post claimed a documented soft limit of ~750 requests/second per bucket at Hetzner and told a sharding story derived from it. Hetzner documents no such number, and the arithmetic in that anecdote did not survive a second look (30 customers with three operations each do not exceed 750 requests/s). The claim is removed, not softened.
Durability and guarantees
AWS publishes eleven nines (99.999999999 %) of durability for S3 objects. Hetzner’s Object Storage documentation publishes no durability figure; if that number matters to your procurement, it is a contract question, not something you can look up. The disclaimer applies either way: a durability figure covers the provider not losing your data, not your data still being correct. Bit rot is what it protects against; a PUT of garbage it does not.
Practical differences:
- AWS has cross-region replication as a product. Hetzner doesn’t; you build it yourself with a second bucket in a second location (NBG and FSN1 are the two).
- Both offer Object Lock for WORM. AWS documents retention modes plus legal hold; Hetzner documents retention and legal hold as how-tos, so check the supported mode per bucket before you promise anything. Neither gives you a SEC 17a-4 attestation.
- AWS has S3 Intelligent-Tiering that auto-moves cold objects to cheaper storage classes. Hetzner has one tier.
Data residency and processing law
The question DACH buyers actually care about:
- Hetzner: GmbH, Gunzenhausen, Germany. Contract, invoice, support — all in German, German jurisdiction. The standard Resistro storage path uses Hetzner data centers in Germany/EU.
- AWS eu-central-1: data resides in Frankfurt, but provider structure, subprocessors, and third-country-transfer assessment remain separate review points.
For some regulated DACH buyers, provider jurisdiction and processor structure can weigh more heavily than pure storage cost or technical capability.
Operational ergonomics
Where AWS still wins:
- CLI tooling.
aws s3 sync, IAM policies, presigned URLs are well-documented and battle-tested. - Documentation. Every edge case has a Stack Overflow thread.
- SDK coverage. Every language has an AWS SDK that works.
Where Hetzner is fine but requires more setup:
- CLI is
mcli(MinIO client). Works, not as polished asaws s3. - No IAM — access is bucket-scoped credentials. For a single-product SaaS, not a problem.
- S3-compatible API works with every SDK; just needs an endpoint override.
The honest summary: if your team already lives in AWS and you’re not cost-sensitive, S3 is fine. If you’re building for EU customers and cost matters, Hetzner is the better default.
When to pick which
| Situation | Recommendation |
|---|---|
| Below 50 GB, already on AWS | Stay on S3. Not worth moving. |
| 100 GB – 5 TB, customers with strict EU-residency preference | Hetzner. Cost and residency both win. |
| > 5 TB, or regulated mid-market | Hetzner for primary + AWS Deep Archive for quarterly off-provider copy. Two-provider strategy is cheap insurance. |
| Need S3 Intelligent-Tiering or CRR | AWS. Hetzner doesn’t play here. |
| Multi-tenant SaaS, need a published rate-limit budget | AWS. On Hetzner you design for 503 + backoff instead. |
In practice, the default depends on two things: whether AWS operational maturity already exists in-house, and how strongly cost and provider jurisdiction matter in procurement. Neither is universally the right choice.