Hetzner pricing: GPU servers and cloud VPS compared
We may earn commissions when you shop through the links below. Learn more
German hosting provider with EU cloud VPS and single-GPU dedicated servers (RTX PRO Blackwell).
- From
- ≈ $2.74 (converted from the EUR list price, €2.40)€2.40
- GPU families
- 1
- Regions
- 6
- Billing
- Hourly
In short
Hetzner Online (Gunzenhausen, Germany; founded 1997) operates its own data-center parks in Falkenstein, Nuremberg and Helsinki and sells Hetzner Cloud VPS (also in Ashburn, Hillsboro and Singapore) plus dedicated GPU servers.
Read full summaryShow less
The GPU line consists of the GEX45 (NVIDIA RTX PRO 4000 Blackwell SFF, 24 GB, Helsinki) and GEX131 (RTX PRO 6000 Blackwell Max-Q, 96 GB, Falkenstein/Helsinki), each with exactly one GPU, an hourly rate capped at a monthly price, a one-off setup fee and unlimited traffic on the default 1 Gbit uplink. A free, experimental OpenAI-compatible inference API and a hosted OpenClaw private beta run on the Hetzner Experiments platform without SLA.
Pricing
Billing: hourly, monthly, usage-based. Lowest pay-as-you-go GPU price: ≈ $2.74 (converted from the EUR list price, €2.40)€2.40 per GPU-hour (RTX PRO 6000 Blackwell Max-Q Workstation Edition 96GB, on-demand). Lowest plan price: ≈ $13.67 (converted from the EUR list price, €11.99)€11.99 per month (VPS).
GPU offers
| Price type and configuration | Billing | Price per GPU-hour | Checked · source |
|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 96GB | |||
| On-demand | Hourly | ≈ $2.74 (converted from the EUR list price, €2.40)€2.40 | |
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 96GB
- On-demand≈ $2.74 (converted from the EUR list price, €2.40)€2.40
Research notes for these offers
Plans
VPS
- Cloud CPX12 (Regular Performance, shared AMD vCPU)≈ $13.67 (converted from the EUR list price, €11.99)€11.99
- Cloud CPX22 (Regular Performance, shared AMD vCPU)≈ $22.79 (converted from the EUR list price, €19.99)€19.99
- Cloud CPX32 (Regular Performance, shared AMD vCPU)≈ $41.04 (converted from the EUR list price, €35.99)€35.99
- Cloud CCX13 (General Purpose, dedicated AMD vCPU)≈ $49.59 (converted from the EUR list price, €43.49)€43.49
- Cloud CPX42 (Regular Performance, shared AMD vCPU)≈ $79.81 (converted from the EUR list price, €69.99)€69.99
- Cloud CCX23 (General Purpose, dedicated AMD vCPU)≈ $98.62 (converted from the EUR list price, €86.49)€86.49
- Cloud CCX33 (General Purpose, dedicated AMD vCPU)≈ $158.49 (converted from the EUR list price, €138.99)€138.99
Show all 11 plansShow fewer plans
Dedicated server
- GEX45-1 dedicated GPU server (NVIDIA RTX PRO 4000 Blackwell SFF, 24 GB)≈ $244.02 (converted from the EUR list price, €214.00)€214.00
- GEX131-1 dedicated GPU server (RTX PRO 6000 Blackwell Max-Q, 96 GB)≈ $1,709.31 (converted from the EUR list price, €1,499.00)€1,499.00
- GEX131-2 dedicated GPU server (RTX PRO 6000 Blackwell Max-Q, 96 GB; 512 GB RAM)≈ $2,393.49 (converted from the EUR list price, €2,099.00)€2,099.00
- GEX131-3 dedicated GPU server (RTX PRO 6000 Blackwell Max-Q, 96 GB; 768 GB RAM)≈ $3,077.67 (converted from the EUR list price, €2,699.00)€2,699.00
Research notes for these plans
Storage and data transfer
- Cloud Volumes (SSD block storage)
- ≈ $0.05 (converted from the EUR list price, €0.044)€0.044
- Object Storage (S3-compatible)
- $0.0071≈ €0.0064 (estimated conversion from $0.0071)
- Storage Box BX11 (1 TB backup storage)
- $0.0037≈ €0.0033 (estimated conversion from $0.0037)
- Local NVMe included with GEX GPU servers
- $0.00≈ €0.00 (estimated conversion from $0.00)
- Egress (data out)
- $0.00≈ €0.00 (estimated conversion from $0.00)
Before you compare
- Suited to
- Private LLM inference on a single-tenant GPU server in Germany or Finland
- Self-hosting AI agents and automation tools (n8n, OpenClaw, Dify) on EU cloud VPS
- Main limitation
- Single-GPU servers only: GEX models cannot be configured with multiple GPUs, and no H100/H200/B200 or NVLink systems are offered
- Price basis
- GPU prices per GPU-hour, billed hourly; egress included; 4 storage options listed
Who it fits
Suited to
- Private LLM inference on a single-tenant GPU server in Germany or Finland
- Self-hosting AI agents and automation tools (n8n, OpenClaw, Dify) on EU cloud VPS
- Teams that need EU data residency and a GDPR Art. 28 DPA for AI workloads
- Steady, month-long GPU workloads where a capped monthly price and unlimited traffic matter more than per-second billing
Good fit for
- Private LLM hostingSingle-tenant dedicated GPU servers in Hetzner-owned ISO/IEC 27001-certified data centers in Germany and Finland, an Art. 28 GDPR DPA, unlimited traffic, and 96 GB VRAM on GEX131 (Hetzner's Ollama tutorial cites up to 70B quantized models) fit private LLM hosting.
- Fine-tuning96 GB of VRAM on a single-tenant GEX131 accommodates parameter-efficient fine-tuning of mid-size open models, and Hetzner links a community tutorial on fine-tuning embedding models. The hourly rate with monthly cap favours multi-week projects because of the EUR 1,499 setup fee (GEX131-1).
- Batch inferenceA dedicated GPU with unlimited traffic on the default 1 Gbit uplink and a fixed monthly cap gives predictable cost for steady batch jobs; there is no spot pricing, and scaling means ordering additional single-GPU servers.
- Real-time inferenceSingle-tenant GPUs in Germany and Finland avoid noisy neighbours, and Hetzner reports serving small LLMs internally on GEX131 with vLLM. There are no autoscaling or serverless GPU endpoints, and new capacity is provisioned server by server (GEX131-1 estimate 10-12 workdays in Germany).
Show 3 moreShow fewer
- AI agentsCloud VPS with hourly billing, a Docker CE one-click app and Hetzner-specific guides in the official n8n and OpenClaw docs suit self-hosted agents. Since 26 Jun 2026 Hetzner restricts creation of new cloud servers for new and some existing customers, and the 15 Jun 2026 price adjustment raised shared and dedicated vCPU plan prices.
- RAG applicationsCloud VPS, S3-compatible Object Storage, block Volumes and GPU servers cover self-hosted RAG stacks (a Hetzner community tutorial covers Dify); there is no managed vector database or paid managed embedding/LLM service.
- RenderingHetzner markets GEX45 and GEX131 for 3D rendering, CAD and video production (RTX GPUs with RT cores); Windows Server 2025 licences are available as paid add-ons on GEX45.
Look elsewhere for
- TrainingEvery GEX server has exactly one GPU (RTX PRO 6000 Blackwell Max-Q 96 GB or RTX PRO 4000 Blackwell SFF 24 GB) and cannot be configured with multiple GPUs, so there is no multi-GPU or NVLink scaling. Suitable only for small-to-mid-size training runs; setup fees and multi-week provisioning estimates add overhead.
Not sure Hetzner fits what you want to build?Help me choose
Strengths and limitations
Strengths
- Dedicated RTX PRO 6000 Blackwell Max-Q (96 GB) server at EUR 2.4022/hour, capped at EUR 1,499/month incl. IPv4 (about USD 2.7392/hour or USD 1709.31/month at ECB 1.1403)
- Unlimited traffic on the default 1 Gbit/s uplink of dedicated/GPU servers; at least 20 TB included per EU cloud server
- Hetzner-owned data centers in Germany and Finland with ISO/IEC 27001:2022, a BSI C5 Type 2 attestation for Hetzner's cloud services (per Hetzner) and a self-service GDPR DPA
- Hourly billing with a monthly price cap across cloud and dedicated products; no minimum contract term on GEX servers
Show 2 moreShow fewer
- Cloud REST API, CLI, Terraform/Ansible integrations and one-click Docker CE and Coolify apps
- The official n8n and OpenClaw documentation both include Hetzner-specific deployment guides
Limitations
- Single-GPU servers only: GEX models cannot be configured with multiple GPUs, and no H100/H200/B200 or NVLink systems are offered
- One-off setup fees (EUR 209 for GEX45, EUR 1,499 for GEX131-1) make short GPU rentals expensive
- GPU servers only in Helsinki (GEX45) and Falkenstein/Helsinki (GEX131); provisioning estimates range from minutes to several weeks depending on type and location
- Cloud capacity is constrained: since 26 Jun 2026 new cloud servers are restricted for new and some existing customers, and on 26 Sep 2026 every Cost-Optimized (CX and CAX) type was shown as not available, so none is listed as a plan here (the CX23 was listed at EUR 5.99/month, about USD 6.83)
Show 4 moreShow fewer
- 15 Jun 2026 price adjustment raised cloud prices substantially for new orders (e.g. CCX13 from EUR 15.99 to EUR 42.99/month excl. IPv4); existing servers keep legacy prices until a rescale, cross-currency transfer or restore
- No serverless GPU, spot or reserved pricing, no native managed Kubernetes and no SOC 2 report
- Unmanaged servers: OS, security and backups of dedicated/GPU servers are the customer's responsibility
- The Experiments inference API and hosted OpenClaw are free but best-effort, without SLA and not intended for production
Infrastructure
Regions
6 regions across 4 countries; 3 are recorded as EU regions.
- Europe (3)
- Falkenstein · Germany · fsn1 (EU); Helsinki · Finland · hel1 (EU); Nuremberg · Germany · nbg1 (EU)
- North America (2)
- Ashburn, VA · United States; Hillsboro, OR · United States
- Asia (1)
- Singapore
Region notes
Frameworks
How Hetzner works with common AI serving, agent and training frameworks. Support levels, strongest first: an official template, documentation, a community guide, or manual setup on a general-purpose machine.
Hosting
- n8n
- Documented setup
- Ollama hosting providers
- Documented setup
- OpenClaw hosting providers
- Documented setup
- Dify
- Community guide
- Flowise
- Manual setup
- Hermes Agent
- Manual setup
- PyTorch
- Manual setup
- vLLM
- Manual setup
Operations
Managed service: no. Each feature below states what the provider documents, with the source it was checked against. “Not verified” means no reliable source was found.
Features
Compute and access
- Custom containers (Docker)
- Yes
- Root access
- Yes
- SSH access
- Yes
- API access
- Yes
- Jupyter notebooks
- No
- Templates
- Partial
- Multi-GPU instances
- No
- Bare metal
- Yes
- Kubernetes
- Partial
Show all 21 featuresShow fewer features
Management
- Managed service
- No
- Serverless
- No
Storage and data
- Persistent storage
- Yes
- Snapshots
- Partial
- Managed backups
- Partial
Billing
- Per-second billing
- No
- Spot pricing
- No
- Reserved pricing
- No
Regions and compliance
- EU region
- Yes
- SOC 2 report
- No
- GDPR data processing agreement
- Yes
- Customer-managed keys (BYOK)
- Partial
SLA and support
- Uptime SLA
- 99.9% uptime · Cloud servers only (no GPUs)
- Support channels
- 5 listed
Sources and verification
How we verify
Prices, regions, storage, features, framework support, SLA and support channels link to the page they were read from, and every price shows when it was checked. Fit labels, strengths and limitations summarise the same research record. The Hetzner record was last checked on 26 Sep 2026. A check date older than 7 days is marked “Stale” with an amber dot, and one older than 30 days “Needs refresh”.
- Last verified
- 26 Sep 2026, 18:40 UTC
- Method
- Source-by-source review by AI agents, independently re-checked
- Confidence
- High
- Checked by
- AI research agent; every value re-checked against its source by a separate AI agent
Verification notes
On 26 September 2026 we checked every price in this listing against Hetzner's live price lists, configurators, documentation and service agreement, converting EUR at the ECB rate of 25 September 2026 (EUR 1 = USD 1.1403). Where official pages disagree, the listing follows the product page and configurator for GEX131 locations (FSN1, HEL1) and the billing docs for hourly billing up to a monthly cap. Live GEX stock was not verified. Later corrections are listed on the methodology page.
Commercial relationship
AIHostCompare has no active affiliate link with Hetzner; the Visit provider link goes to its official website. Commissions never change ordering, alternatives, fit labels or conclusions. Read the affiliate disclosure.
Pricing
Show all 42 sourcesShow fewer sources
Pricing
Documentation
Terms & policies
Status
Secondary sources
Alternatives to Hetzner
| Fact | This providerHetzner | DigitalOcean profile | OVHcloud profile | Scaleway profile |
|---|---|---|---|---|
| From | ≈ $2.74 (converted from the EUR list price, €2.40)€2.40 | $1.57≈ €1.40 (estimated conversion from $1.57) | ≈ $0.798 (converted from the EUR list price, €0.70)€0.70 | ≈ $0.898 (converted from the EUR list price, €0.788)€0.788 |
| GPU families | RTX PRO 6000 | MI300X · H200 · H100 · +2 more | H100 · L40S · A10 · +2 more | H100 · L40S · L4 |
| EU regions | 3 in the EU | 2 in the EU | 4 in the EU | 10 in the EU |
| Billing increment | Hourly | Per second | Per second | Per minute |
| Egress | No egress feesSource | $0.0093≈ €0.0083 (estimated conversion from $0.0093) Source | No egress feesSource | No egress feesSource |
- DigitalOcean profile$1.57≈ €1.40 (estimated conversion from $1.57)
- OVHcloud profile≈ $0.798 (converted from the EUR list price, €0.70)€0.70
- Scaleway profile≈ $0.898 (converted from the EUR list price, €0.788)€0.788
Related comparisons and guides
- Compare AI hosting and GPU cloud providers
- GPU Cost Calculator
- VRAM / Model Fit
- Agent Hosting Selector
- GDPR AI hosting: an EU data-residency checklist
- How much VRAM do you need to run an LLM?