Hetzner expands cloud and AI infrastructure with new GPU server options

German cloud provider Hetzner has announced expanded cloud offerings with GPU server options targeting European AI training and inference needs.

Details

Hetzner's GPU servers are based on NVIDIA A100 and L40S GPUs, offered in both dedicated server and cloud instance form factors. Known for price-performance leadership among European cloud providers, Hetzner's GPU pricing is expected to be 30-40% lower than major cloud providers.

On specific models, Hetzner's dedicated GPU servers typically come in a few common configurations (prices are public reference prices at launch; check the website for current figures):

Configuration GPU VRAM CPU / RAM Form Reference Price
Entry GPU 1 × L40S 48 GB 12 cores / 64 GB Cloud instance ~€0.9-1.3/hour
Single GPU 1 × A100 80GB 80 GB 16 cores / 128 GB Dedicated server ~€500-700/month
Dual GPU 2 × A100 80GB 160 GB 32 cores / 256 GB Dedicated server ~€1,100-1,500/month

Compared with an equivalent AWS instance (such as the p4d.24xlarge at about $32/hour), Hetzner's advantage shows most clearly for 24x7 inference services: the same budget buys a single on-demand instance on the big clouds, but a dedicated server on Hetzner — possibly with a spare GPU card.

European advantage

For websites and AI applications operating in Europe, Hetzner offers GDPR-compliant data localization. Its data centers in Germany, Finland, and Austria benefit from lower electricity costs, supporting competitive pricing.

16IDC Takeaway

Hetzner's entry into the GPU server market is a significant signal for European cloud computing. For AI applications targeting European users, Hetzner offers a balance between data compliance and cost control.

Background: European Cloud's AI Breakout

Hetzner is a veteran German cloud provider known for "extreme price-performance" in dedicated servers and cloud hosting. But it has been on the sidelines of the AI infrastructure wave — until now.

Key context for Hetzner's GPU server entry:

  1. European AI demand surge: European startups and enterprises are accelerating AI adoption, but data sovereignty and GDPR compliance push them toward local cloud providers
  2. Major cloud pricing premium: AWS, Azure, GCP GPU instances in European regions typically cost 20-30% more than US regions, leaving room for European providers
  3. Hetzner's cost advantage: Data centers in Germany, Finland, and Austria benefit from lower electricity costs (especially Finland's green energy), giving Hetzner a natural cost structure advantage

Practical Impact for Site Builders

Hetzner GPU Value Positioning

Comparison Hetzner GPU AWS/GCP GPU Difference
Price expectation 30-40% lower Baseline Hetzner cheaper
GPU models A100, L40S H100, H200, B200 Hetzner older gen
Form factor Dedicated server / cloud instance Cloud instance Hetzner more flexible
Network 1-10 Gbps 100-800 Gbps Major clouds faster
Data compliance ✅ Native GDPR ✅ Global compliance Comparable

Suitable AI Workloads

  1. Model inference: L40S is inference-optimized, more economical than H100
  2. Small-to-medium fine-tuning: A100 80GB covers most open-source model fine-tuning
  3. European-localized AI apps: Data stays within Europe for EU users
  4. Batch inference: Low-cost GPU advantage for async batch processing

Less Suitable

  • Large-scale distributed training (lacks InfiniBand interconnect)
  • Workloads needing latest GPU models (H200/B200)
  • Global multi-region deployment needs

Getting Started

Setting up a GPU on Hetzner is straightforward: log into the Cloud Console, create a server, pick the GPU series (A100 or L40S), location, and data center; choose Ubuntu 22.04 or 24.04 and pair it with NVIDIA's official driver install script; upload your SSH key in advance to avoid password logins. Once up, run nvidia-smi to confirm the driver and VRAM are recognized, then deploy an inference service directly or through an AI platform tool. Note that dedicated servers are billed monthly while cloud instances are hourly — pick hourly for short experiments and monthly for long-running workloads.

Actionable Recommendations

  1. Compare TCO: Look beyond GPU hourly price — include network transfer, storage, and operational costs
  2. Leverage dedicated server model: For long-running inference, Hetzner's dedicated GPU servers may be more economical
  3. Watch network throughput: If your model loads data from object storage frequently, assess if bandwidth meets needs
  4. Assess data sovereignty needs: For EU users with GDPR requirements, Hetzner is a strong candidate
  5. Consider hybrid cloud: Train on AWS/Azure, deploy inference on Hetzner for cost-performance balance

Deeper Perspective

Hetzner's GPU entry reflects a key European cloud trend — "sovereign AI" demand is nurturing a local GPU cloud ecosystem. Not just Hetzner — OVHcloud, Scaleway, and other European providers are also actively investing in AI infrastructure.

For websites and AI applications targeting European users, these local GPU cloud providers offer a better balance between cost and compliance. As data protection regulations tighten globally, data localization will become a hard requirement for more businesses.

Source: Hetzner

Reference: Hetzner GPU server docs: https://docs.hetzner.com/cloud/advanced-topics/accelerated-computing/