Hetzner Cloud Review 2026: Benchmarks, Pricing, and the Real Trade-offs

Hetzner has been running datacenters in Germany since 1997, but for most of that history it was primarily known as a European dedicated server provider. Its cloud product, Hetzner Cloud, launched in 2017 and changed the conversation around what budget VPS pricing could look like.

The headline numbers are hard to ignore. A 2 vCPU / 4 GB AMD instance currently runs at €5.99/month, a fraction of what comparable plans cost on DigitalOcean or Vultr. European developers adopted it quickly. Then American developers started noticing. The question worth asking in 2026 is what exactly you're giving up at that price, and whether the trade-offs matter for real workloads.

Pricing notice: Hetzner has announced a price adjustment taking effect April 1, 2026 for all new and existing products. The CPX22 in Germany/Finland will increase from €5.99/month to €7.99/month. All prices in this article show both current and post-April figures. See Hetzner's full price adjustment page for the complete breakdown.

This review answers the performance question with real benchmark data. We provisioned a 2 vCPU / 4 GB RAM CPX22 instance in Hetzner's Helsinki (hel1) datacenter and ran YABS twice to measure disk I/O, network throughput, and CPU performance, then went beyond the numbers to cover support, reliability, security, billing, and the scenarios where Hetzner works well and where it doesn't.

Quick verdict

What Hetzner Cloud offers

Hetzner Cloud's compute lineup is organized around Cloud Servers (their term for virtual machines) in two broad categories: Shared vCPU for general-purpose workloads and Dedicated vCPU for production and compute-intensive jobs.

Under Shared vCPU, Hetzner splits plans by processor type:

• Shared Cost-Optimized (CX-series): general-purpose shared instances on Intel or AMD hardware with NVMe SSD storage in a local RAID10. The cheapest entry point on the platform, labeled "Cost-Optimized" and available only in EU regions with limited capacity.
• Shared Regular Performance (CPX-series): AMD EPYC shared instances, labeled "Regular Performance" and available globally including US and Singapore. Delivers meaningfully stronger single-core performance than the CX tier. This is what we benchmarked.
• ARM64 (CAX-series): Ampere Altra-based instances at the lowest prices in the lineup, suitable for workloads compiled for ARM or containerized applications running multi-architecture images.

Under Dedicated vCPU, the CCX-series offers AMD EPYC cores where no neighboring tenant can contend for your CPU time, designed for production databases, CI/CD pipelines, and anything where performance consistency under load matters more than raw price.

Beyond compute, Hetzner Cloud includes load balancers, managed Kubernetes, object storage (S3-compatible), volumes (attachable block storage), private networks, firewalls, and DNS management. Backups and snapshots are built in at predictable per-GB pricing.

It is a more focused product than DigitalOcean or Vultr. There is no managed database service, no container registry, and no PaaS app deployment layer. What Hetzner offers, it does well and cheaply. What it doesn't offer, you'll need to build or source elsewhere.

Account registration and verification

One thing worth knowing before you commit: Hetzner sometimes requires identity verification for new accounts, particularly for accounts being created outside of Germany or paying with certain methods. This can mean submitting a passport scan or government ID before your first server can be provisioned.

This is not universal. Many accounts are approved automatically, but it's common enough to trip up developers who want to spin up a server immediately. If you're evaluating Hetzner under a deadline, register a few days early. The process is generally completed within a business day once documentation is submitted, but it can be surprising if you're not expecting it.

Payment methods accepted include credit/debit cards and PayPal. Hetzner does not currently accept cryptocurrency.

Setup and first impressions

Hetzner Cloud's provisioning interface, the Cloud Console, is the simplest in this benchmark series. Everything fits on a single page with no second screen, no upsell funnel, and no clutter. The form flows top to bottom: plan type, location, OS image, networking, SSH keys, and optional extras.

For this review we provisioned a CPX22: 2 vCPU / 4 GB RAM / 80 GB SSD in the Helsinki (hel1) datacenter. Hetzner operates this from Tuusula, Finland, about 30 minutes from central Helsinki. The instance ran Ubuntu 24.04 LTS. Monthly cost at time of benchmarking: €5.99/month (~$6.50). This rises to €7.99/month from April 1, 2026.

The first choice is the plan. Shared Resources and Dedicated Resources are split into tabs, with Shared further divided by processor type: Cost-Optimized (CX-series, Intel/AMD), Regular Performance (CPX-series, AMD EPYC), and ARM64 (CAX-series, Ampere Altra). Each card shows vCPU count, RAM, disk, included traffic, hourly rate, and monthly cap. The CPX22 sits in the Regular Performance tab. All plans use NVMe SSDs in a local RAID10 on the host system, providing fault tolerance and high I/O performance across every tier and region.

Location comes next. It covers five regions: Nuremberg (Germany), Falkenstein (Germany), Helsinki (Finland), Ashburn (US East), and Hillsboro (US West). The selector notes that private networks, load balancers, primary IPs, and volumes must share the same location or network zone as the server — a useful reminder when building multi-resource stacks. For European workloads, the German and Finnish options are geographically dense and well-peered at major exchange points. Note that US regions include only 1 TB of monthly transfer and Singapore only 0.5 TB, compared to 20 TB for European locations. Overage is charged at €1.00/TB in EU and US, and €7.40/TB in Singapore.

Image selection follows location. OS support covers Ubuntu, Debian, Fedora, CentOS Stream, AlmaLinux, and Rocky Linux, with the most current release of each available. Ubuntu 24.04 LTS is the default. An Apps tab offers pre-configured images including WordPress, Docker, LAMP, and Nextcloud, pre-installed and ready to use on first boot.

Networking is next, with three options: public IPv4 and IPv6, IPv6 only, or private network only (which requires selecting a private network first). IPv4 primary IPs cost €0.00097/hour whether or not they are attached to a running server. IPv6 is free. SSH keys follow networking: keys saved to your account appear as checkboxes, and adding one means no root credentials are sent by email. After SSH keys, the form exposes optional extras in order: Volumes (up to 10 TB SSD block storage), Firewalls, Backups (automated daily snapshots at 20% of the server price), Placement groups, Labels, and Cloud config for cloud-init scripts up to 32 KiB.

Benchmarks (YABS)

I ran Yet Another Bench Script (YABS), the standard benchmark used across the LowEndTalk community, twice on fresh instances to verify consistency.

To run it on your own Hetzner instance:

curl -sL yabs.sh | bash

To save results as JSON for cross-provider comparison:

curl -sL https://yabs.sh | bash -s -- -w results.json

Both runs were conducted on Ubuntu 24.04.3 LTS, kernel 6.8.0-90-generic, KVM virtualization, Helsinki, Finland. Disk figures are averaged across both runs; network figures are from the second run, which had fewer busy endpoint collisions.

CPU

Processor  : AMD EPYC-Rome Processor
CPU cores  : 2 @ 2445.404 MHz
AES-NI     : ✔ Enabled
VM-x/AMD-V : ❌ Disabled
RAM        : 3.7 GiB

Geekbench 6 scores:

The two runs show normal variance for a shared-CPU instance. Single-core scores of 964 and 914 reflect host-level load variation rather than instability. The working average of ~939 single-core is the more useful planning number.

That average sits comfortably above DigitalOcean's Basic Droplet (772 single-core at the same price tier), and at roughly half of what Vultr's AMD EPYC-Genoa instance delivers at $24/month. The gap to Vultr is explained by silicon generation: the CPX22 runs EPYC-Rome at 2.4 GHz, while Vultr's High Performance AMD tier uses newer EPYC-Genoa with higher clock speeds and better IPC. At roughly one-fifth the price, the CPU performance is strong, but this is the one benchmark where the price-per-dollar story gets more nuanced. Teams running consistently CPU-saturating workloads should consider the CCX Dedicated vCPU tier within the same platform.

The VM-x/AMD-V flag shows disabled, which is standard for shared-CPU instances where nested virtualization is not exposed to the guest. This is irrelevant for typical application workloads.

Disk I/O

fio results averaged across both runs:

The 4k random IOPS figure of ~40.9k is solid and consistent across both runs. It lands below Vultr's 118k IOPS result at the same price point, though direct comparison is complicated by workload conditions and host-level variance on shared storage. Sequential throughput scales well: 1.54 GB/s combined at 64k block sizes, approaching 3 GB/s at 1m, consistent with the NVMe RAID10 local storage that Hetzner uses across all plans and regions.

The YABS fio test runs mixed read/write at 50/50, so absolute throughput numbers will be lower than single-direction benchmarks. For most web application workloads, these numbers are sufficient. Teams where 4k random IOPS is a hard bottleneck (high-write databases, dense logging pipelines) should evaluate the CCX Dedicated vCPU series, which guarantees CPU and storage resources without noisy-neighbor contention.

Network

The server is in Helsinki, and the results show Hetzner's home advantage clearly. IPv4 results from the second run:

*The NYC receive anomaly (115 Kbits/sec) was a test endpoint issue. Run 1 returned 1.42 Gbits/sec to the same location and is more representative.

The European results are the standout. Amsterdam clocked 8.08 Gbits/sec send at 28.8 ms, the highest throughput in this benchmark series for that direction. London returned 5.02/5.22 Gbits/sec send/receive at 38 ms. Hetzner's network peers directly at DE-CIX (Frankfurt) and AMS-IX (Amsterdam), and for workloads serving European users, these numbers represent a genuine infrastructure advantage that no comparable budget provider can match.

The picture changes across the Atlantic and into APAC. NYC sits at 131 ms, Los Angeles at 157 ms, Singapore at 184 ms. These figures are the honest cost of running from Helsinki when your users are not in Europe. For US-facing workloads, Hetzner's Ashburn or Hillsboro regions close this gap entirely. Hetzner has no Asian or Australian datacenters, and that absence is felt from any region.

IPv6 performance matches IPv4 across European endpoints: 7.41–7.93 Gbits/sec to Amsterdam, 4.20+ Gbits/sec receive to London. IPv6 is first-class here, not an afterthought.

Control panel and management

Hetzner Cloud's Cloud Console is fast, minimal, and visually uncluttered. It loads instantly, the navigation is obvious, and every common operation is within two or three clicks of the server list. No dashboard banners, no upsell prompts in the UI, no navigation elements that exist primarily to surface other products.

After provisioning, the server detail page shows status, public and private IP addresses, usage graphs (CPU, disk, network, traffic), and a browser-based console. Action menus expose power controls, networking configuration, snapshot management, and resize options.

Notable management features:

Firewalls. Stateful, free, and applied at the infrastructure level, not as an iptables wrapper inside the VM. Define inbound and outbound rules, assign a named firewall set to one or more servers, and changes propagate immediately. Reusable across a fleet without per-server rule management.

Placement groups. Configure servers to always land on different physical hosts, reducing the blast radius of a hardware failure. Available at no extra cost and set at provisioning time. This is an operational resilience feature that neither DigitalOcean nor Vultr offer as a first-class provisioning option.

Volumes. Attachable block storage at €0.044/GB/month (€0.057/GB/month from April 1). Detachable and reattachable, useful for separating persistent data from stateless compute.

Snapshots. On-demand full disk images at €0.011/GB/month (€0.014/GB/month from April 1). Useful for recovery points, staging copies, and pre-upgrade checkpoints.

Backups. Automated snapshots at 20% of the instance cost, retaining up to 7 backups. For the CPX22 at €5.99/month (€7.99/month from April 1), that's approximately €1.20–1.60/month for automatic recovery coverage.

Private Networks. Layer 3 virtual networks connecting servers within a project, for isolating backend services from public internet exposure without firewall rule complexity.

Cloud-init support. Pass user data at provisioning to automate first-boot setup. Combined with the Hetzner API or Terraform provider (both actively maintained), reproducible infrastructure is straightforward.

Browser Console. In-browser terminal access that bypasses SSH entirely. No additional configuration needed when a firewall rule locks you out.

Support

Hetzner offers 24/7 ticket-based support for cloud products, available around the clock every day of the week. Their own cloud page advertises "expert assistance for your cloud servers and services around the clock." Phone support is available but restricted to dedicated server customers who have activated a telephone password; it does not extend to cloud VPS products.

There is no live chat. For cloud products, your channel is the ticket system via the Hetzner Console, and response quality is generally considered competent when they engage. The platform's stability means most teams rarely need to open a ticket, and Hetzner's 350+ community tutorials cover the majority of common issues.

For teams running production systems with tight on-call SLAs, the absence of phone support for cloud products is worth factoring in. Hetzner's philosophy leans toward stable, self-service infrastructure rather than managed hand-holding, and the pricing reflects that trade-off.

Uptime and reliability

Hetzner does not publish a formal uptime SLA for cloud products. There is no guaranteed uptime percentage in their terms of service and no credit-backed compensation mechanism for downtime. This is a meaningful distinction from DigitalOcean (99.99% SLA) and worth stating clearly for teams with contractual uptime requirements.

In practice, independent monitoring and community reports show Hetzner Cloud performing reliably. The platform's architecture is KVM-based virtualization on their own hardware in their own datacenters, which gives them direct control over maintenance windows and hardware replacement. Hetzner is vertically integrated in a way most providers are not: they own their physical infrastructure rather than reselling from a hyperscaler.

For genuinely high-availability deployments, spreading instances across Nuremberg and Helsinki (or adding an Ashburn instance) provides meaningful redundancy. Placement groups help with physical host diversity within a single datacenter.

Security and data privacy

Hetzner is a German company subject to strict EU data protection law and GDPR-compliant by default across all datacenter regions. European data residency is the platform's baseline. Your data does not leave the EU unless you explicitly provision in Ashburn or Hillsboro. For companies with data residency requirements or EU-facing compliance obligations, this is a meaningful structural advantage over US-headquartered providers.

Built-in security features included at no extra cost:

DDoS protection. Basic DDoS mitigation is included free of charge on all cloud servers with no opt-in required. Hetzner's network infrastructure absorbs volumetric attacks before they reach your instance. This is handled at the network level, without any configuration step on your part.

Firewalls. As covered above: stateful, infrastructure-level, free, and applied before traffic reaches the VM.

Private networks. Isolate backend services from public internet exposure with no additional cost.

SSH key authentication. The default authentication method. Password-based root login is disabled by default on clean installs.

For teams with deeper compliance requirements (ISO 27001, SOC 2, PCI DSS), Hetzner does not publish third-party audit certifications for cloud products. IONOS and larger cloud providers may be more appropriate for regulated industries with formal certification requirements.

Billing model

Hetzner bills hourly, capped at the monthly rate. You pay by the hour for as long as a server exists (including while it's powered off), but the bill is capped at the monthly price. There are no surprise overages at month end for compute, and no termination fees.

The hourly cap model means short-lived instances (for CI/CD jobs, batch processing, or spinning up a staging environment for a day) are billed at their actual usage rather than a full month. Combined with sub-30-second provisioning, this makes Hetzner practical for ephemeral workloads.

Servers accrue charges from creation until deletion. Powering off a server stops compute activity but keeps the disk and IP reserved, and billing continues. To stop accruing charges completely, you need to delete the instance (snapshot it first if you need to restore it later).

Transfer is included generously in European regions (20 TB/month). Overage rates are modest and clearly documented. The US and Singapore regions include only 1 TB/month, so factor that into your region selection if bandwidth volume matters.

The ecosystem

Hetzner Cloud's managed service layer is intentionally narrower than DigitalOcean or Vultr. No hosted database product, no container registry, no PaaS deployment platform, and no built-in CDN. What Hetzner does offer around its compute core:

Load Balancers. Currently starting at €5.39/month for the LB11 (rising to €7.49/month from April 1), handling up to 25 services and 10,000 connections, with HTTP, HTTPS, and TCP support and TLS termination included.

Managed Kubernetes (HKE). Worker nodes are provisioned as Cloud Server instances, so platform familiarity carries over. The Kubernetes control plane is free; you pay only for worker nodes and associated resources. Auto-scaling node groups and integrated load balancers are supported.

Object Storage. S3-compatible storage, currently starting at €4.99/month base (rising to €6.49/month from April 1), available in Nuremberg and Falkenstein. Not yet available in US datacenter regions.

Dedicated Servers. Accessible under the same account as Hetzner Cloud, and one of the cheapest sources of bare-metal compute available anywhere. For teams that grow beyond VM-appropriate workloads, the path forward stays in the same billing account.

The absence of managed databases is the most significant gap for teams wanting a single-provider stack. PostgreSQL, MySQL, Redis, and message queue services all require either self-hosting on Hetzner Cloud instances or introducing a third-party provider like Supabase, PlanetScale, or Upstash. That's a real operational cost, and for teams without database administration experience it partially offsets the compute savings.

Documentation and community

Hetzner's official documentation covers the essentials accurately: server provisioning, the API, networking, Kubernetes, and CLI usage. It is not as comprehensive or tutorial-rich as DigitalOcean's documentation library, which remains the industry standard. Hetzner doesn't invest in SEO-driven community content at the same scale, and it shows for less common configuration scenarios.

What the official docs don't cover, community resources often do. The LowEndTalk forums have active Hetzner threads, the Hetzner subreddit is well-trafficked, and the platform's popularity among European developers means third-party guides are easy to find. The Hetzner Community site hosts user-contributed tutorials, though quality varies more than on DigitalOcean's curated library.

For debugging or advanced configuration, expect to rely on general Linux and cloud administration resources rather than Hetzner-specific documentation. This is not a blocker for experienced teams, but it's worth naming for developers new to self-managed infrastructure.

Pricing

The CPX22 (2 vCPU / 4 GB, Germany/Finland) currently costs €5.99/month ($6.99/month). On April 1, 2026, prices increase across the entire cloud lineup. Benchmarks in this article were run before the price change; figures below show both current and post-April pricing.

Shared Regular Performance AMD (CPX-series) Germany/Finland:

Shared Cost-Optimized (CX-series) Germany/Finland only (lower price, EU-only availability):

Dedicated vCPU (CCX-series) Germany/Finland:

US and Singapore regions: The CPX-series is available in the US at slightly higher prices and with 1 TB transfer included (not 20 TB); Singapore includes only 0.5 TB. The CPX31 (4 vCPU / 8 GB) runs $17.99/month currently, rising to $24.99/month from April 1. Note also that CX and CAX series are EU-only: the Cost-Optimized and ARM64 tiers are not available in US or Singapore regions.

Additional resources (all locations):

The most striking number in Hetzner's pricing, even after the April increase, is the included 20 TB monthly transfer in European regions. That's five times DigitalOcean's 4 TB and four times Vultr's 5 TB at equivalent pricing.

Pricing context. After the April 1 adjustment, the CPX22 rises to €7.99/month (~$9.49/month). Compared to the $24/month DigitalOcean Basic or $24/month Vultr High Performance AMD instance benchmarked in this series, that's still roughly a 60% cost reduction for the same vCPU/RAM spec. The "absurdly cheap" framing that has followed Hetzner for years is becoming less absolute, but the gap remains wide, and the 20 TB transfer advantage is unaffected by the pricing change.

Final thoughts

Hetzner Cloud remains the reference point for price-per-spec in the VPS market in 2026. Even after the April 1 price increase, few providers come close at the same spec. CPU performance is strong for the price, disk performance is solid with NVMe RAID10 local storage, and the European network is exceptional, with very strong intra-EU throughput. The 20 TB transfer allowance in EU regions remains one of the clearest differentiators in the market.

The trade-offs are real. There is no managed database, no formal uptime SLA, and the ecosystem is relatively lean. Teams that require managed services, compliance certifications, or strict uptime guarantees may see these as meaningful gaps.

For European teams, the value proposition is straightforward: EU data residency, excellent connectivity, generous transfer limits, and aggressive pricing.

For US or global audiences, Hetzner’s US regions reduce latency, but the included transfer drops to 1 TB, which should be considered when estimating total cost.

If raw compute and storage performance per dollar is the primary constraint and you can manage your own database layer, Hetzner remains one of the strongest value options in the VPS market in 2026.

Got an article suggestion? Let us know

Next article

DigitalOcean Review 2026: Benchmarks, Pricing, and the Real Trade-offs

We benchmarked DigitalOcean's $24/month 2vCPU/4GB Droplet with real YABS results. Strong disk and network performance, but weaker CPU scores and higher pricing compared to European rivals

→

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.