EU datacenter · GDPR compliant
What Hetzner actually sells
Hetzner is a German hosting company founded in 1997 and one of Europe's most trusted infrastructure names. Its GPU products are the GEX line of dedicated root servers. Hetzner Cloud, the VM product people usually mean by "Hetzner", has no GPU instances at all; the cloud pages simply link out to the dedicated server matrix.
That distinction changes everything about how you use it. You rent a physical machine with a workstation-class NVIDIA card bolted into it, on a monthly contract, and it stays powered and billed until you cancel. There is no stop-to-save-money button and no per-second meter. Any hourly figure you see for Hetzner, including the €1.42/h on this page, is €889 divided across a month.
The current line-up is two models, and one of them is out of stock. The GEX131 pairs an RTX PRO 6000 Blackwell Max-Q (96 GB GDDR7 ECC, 3511 TFLOPS tensor) with a 24-core Xeon Gold 5412U and 256 GB of DDR5 ECC, for €889/month with no setup fee; it launched in December 2025. The GEX44 pairs an RTX 4000 SFF Ada (20 GB, 306.8 TFLOPS tensor) with a Core i5-13500 and 64 GB of DDR4, and the matrix currently shows it as not available. An older GEX130 with a 48 GB RTX 6000 Ada has dropped off the matrix entirely and should be treated as discontinued.
On price, Hetzner is not the answer. €1.42/h is above OVH at €0.36/h and Scaleway at €0.79/h, and far above marketplace rates like RunPod Community at $0.16/h. What you get instead is a machine nobody else is sharing, in German or Finnish jurisdiction, on 100% green electricity, with 96 GB of VRAM in one card and no minimum contract term.
EU Data Residency and GDPR
Hetzner's GPU servers run exclusively in datacenters in Germany (Nuremberg, Falkenstein) and Finland (Helsinki). All data processing stays within the European Economic Area, making Hetzner straightforwardly GDPR-compliant by design rather than through contractual gymnastics.
For sectors like healthcare (patient data), finance (financial records), and public sector projects (government data), this is often a hard requirement that eliminates US-headquartered providers regardless of their contractual commitments. Hetzner offers standard Data Processing Agreements for enterprise customers.
Hetzner GPU Pricing (July 2026)
| GPU | VRAM | Approx Hourly | Monthly | Best For |
|---|---|---|---|---|
| GEX131 — RTX PRO 6000 Blackwell Max-Q | 96 GB GDDR7 ECC | €1.4247/h | €889/m, no setup fee | Large models that need 96 GB on one card |
| GEX44 — RTX 4000 SFF Ada | 20 GB GDDR6 | Currently not available | Currently not available | Smaller inference jobs, when stock returns |
| GEX130 — RTX 6000 Ada (previously offered) | 48 GB | €1.3429/h | €838/m + €79 setup | No longer listed; treat as discontinued |
| A100 / H100 / any SXM part | — | Not offered | — | Hetzner has never sold these |
Prices in EUR, excluding VAT. The monthly price is what you are actually billed; the hourly column is that figure divided across a month, for comparison against per-hour providers only. These are dedicated servers, so powering one off does not stop the charge. There is no minimum term and you can cancel immediately. Check the Hetzner GPU server matrix for current pricing and stock.
Hetzner GPU Pros & Cons
- Dedicated hardware, not a shared or virtualised slice of a GPU
- 96 GB of VRAM on a single card in the GEX131
- German and Finnish datacenters, GDPR by default
- 100% green electricity
- No minimum term, cancel immediately
- Not the cheap option: €1.42/h entry price vs €0.36/h at OVH
- Two SKUs in the whole line-up, and one of them is out of stock
- No A100, no H100, no SXM datacenter parts, and none planned
- Monthly dedicated servers, so you cannot stop one to pause billing
- One GPU per server, no InfiniBand, no multi-node clusters
Who Should Use Hetzner GPU?
Hetzner GPU is ideal for: teams with a steady, always-on inference or fine-tuning workload that will keep a card busy for weeks; organizations where EU jurisdiction and dedicated (not shared) hardware are hard compliance requirements rather than preferences; anyone who needs 96 GB of VRAM addressable from one card without renting a multi-GPU node; and existing Hetzner customers who want the GPU on the same invoice and the same network as the rest of their infrastructure.
Hetzner GPU is not ideal for: bursty or experimental work, where paying €889 for a month to use a card for six hours makes no sense; anyone who needs A100, H100 or SXM parts; multi-node distributed training; and price-driven buyers, who will do better at OVH, Scaleway or a marketplace.
Hetzner GPU Alternatives
- OVH GPU — French, EU-jurisdiction, and cheaper at the entry point (from €0.36/h) with a wider GPU catalogue and enterprise SLA contracts. The obvious first alternative if EU data residency is the reason you were looking at Hetzner.
- Scaleway — Also EU, from €0.79/h, with genuine per-hour cloud instances you can stop. More flexible than Hetzner for anything that is not running continuously.
- RunPod — Far cheaper on Community Cloud ($0.16/h) and much wider GPU variety including H100. US-headquartered, so the compliance story is weaker, but the right choice for short or spiky jobs.
- Lambda Labs — H100 access and reliable dedicated hardware, but US-headquartered with US/AU datacenters only, which rules it out for EU data residency requirements.
Verdict
Hetzner GPU is a narrow but honest product: one good server, the GEX131, with 96 GB of VRAM, dedicated hardware, EU jurisdiction and green power, at €889/month. It is not the cheap option and we no longer describe it as one; OVH and Scaleway both undercut it, and marketplaces undercut all three. It is also not a cloud, so the "spin one up for an afternoon" workflow does not exist here.
Buy it when you have a workload that will saturate a card for weeks and a compliance requirement that rules out US providers. If either half of that is missing, rent by the hour somewhere else.
Hetzner GPU FAQ
Does Hetzner have H100 or A100?
No, and it never has. As of July 2026 the entire Hetzner GPU line-up is two dedicated server models: the GEX44 with an RTX 4000 SFF Ada (20 GB), currently listed as not available, and the GEX131 with an RTX PRO 6000 Blackwell Max-Q (96 GB) at €889/month. Both use workstation cards in a single-socket server. Hetzner sells no A100, no H100 and no SXM datacenter parts, and its own documentation states you cannot install additional graphics cards in the servers. For H100 in Europe, Lyceum is $2.79/h, OVH €2.80/h and Nebius $3.85/h. For A100, Lyceum at $1.59/h is the clearest EU option — OVH has de-listed A100 from its public GPU page.
Is Hetzner GDPR compliant?
Yes, Hetzner is a German company with datacenters in Germany and Finland, both fully within the EU. All data processing is subject to GDPR by default, and Hetzner offers Data Processing Agreements (DPA) for enterprise customers. Its EU-only infrastructure makes it one of the strongest choices for European organizations with strict data residency requirements, including healthcare, finance, and public sector projects. The whole estate also runs on 100% green electricity, which matters if you report on procurement emissions.
How does Hetzner compare to RunPod for EU users?
Not on price. Hetzner's cheapest GPU server is €1.42/h equivalent, while RunPod Community starts at $0.16/h and its A100 80GB is $1.19/h on Community, $1.39/h on Secure Cloud. Hetzner wins on jurisdiction and on hardware model: a German company, EU-only datacenters, euro invoices, and a physical server nobody else is sharing. RunPod wins on price, on GPU variety including H100, and on being able to spin a pod up for ten minutes. Pick Hetzner when EU jurisdiction or dedicated hardware is a requirement rather than a preference.
How does Hetzner billing work for GPU servers?
These are dedicated servers on a monthly contract, not cloud VMs. The GEX131 costs €889/month with no setup fee; the hourly figures quoted here and elsewhere are that monthly price divided across the month, not a rate you can actually be billed at. There is no stop button that pauses the bill: the server is yours, and powering it off saves nothing. Hetzner does apply no minimum term and lets you cancel immediately, so a two-week project costs roughly half a month. But you cannot rent one for twenty minutes, and if that is what you need, a per-second provider is the right tool.
Can you build a multi-GPU or multi-node setup on Hetzner?
Only in a limited way. Each GEX server ships with one GPU, and Hetzner's documentation states you cannot install additional graphics cards in the servers, so a single box tops out at one card. There is no InfiniBand and no GPU fabric between machines, which rules out the tightly coupled distributed training that multi-node jobs depend on. The GEX131's 96 GB of VRAM is the practical answer here: it fits models on one card that would otherwise need several. For anything larger, CoreWeave, Lambda Labs or a hyperscaler are the realistic options.