Both Hetzner and Contabo sit in the EU budget cloud tier. Both offer dramatically more resources per euro than DigitalOcean or Vultr. Neither is promotional — the pricing reflects deliberate infrastructure decisions that each provider made.

The comparison is not about which is more affordable. It is about one specific trade-off: CPU performance consistency in exchange for RAM and storage density. Contabo maximizes RAM and storage per euro through deliberate CPU oversubscription. Hetzner delivers dedicated vCPU performance at a higher resource cost. The workload determines which trade is acceptable.

This is one of the few infrastructure comparisons where the workload profile — not the budget, not the team's technical capability — is the determining factor.

Hetzner tends to suit production workloads where CPU consistency is a product requirement: API services, interactive web applications, CI/CD pipelines, any application with concurrent users.

Contabo tends to suit RAM-bound or storage-bound workloads where CPU variability is acceptable: development and staging environments, backup targets, media servers, batch processing jobs without latency requirements.

The determining variable is simple: does your workload have users who notice when the server's CPU is slow? If yes, Hetzner. If no, Contabo's resource density becomes relevant.

Hetzner's product is honest hardware at honest prices. Own data centers, own hardware fleet, no managed services ambition, no marketing overhead. The efficiency gains go to customers as compute density and pricing that independent benchmarks confirm. A CCX32 — 4 dedicated AMD EPYC vCPUs, 16 GB RAM — at €16/month performs comparably to configurations that cost $80–100/month from cloud providers with larger managed services investments. The product reputation is built almost entirely on benchmark results and developer community trust.

Contabo's product model is structurally different. High CPU oversubscription ratios enable extreme resource density at the entry tier. 8 GB RAM and 200 GB NVMe SSD for under $7/month is not a promotional offer — it is the product. Memory and storage are hard-allocated and consistent. CPU is shared at higher density. Under load, CPU performance varies based on neighboring VM activity in ways that Hetzner's dedicated vCPU architecture does not.

Contabo is not hiding this structure. The oversubscription model is a deliberate engineering choice designed for specific workload profiles — not a constraint imposed by budget limits. The product is excellent for RAM-bound and storage-bound workloads. It is not designed for CPU-sensitive production applications, and using it as a general-purpose VPS because the resource numbers look compelling is where the model creates problems.

Hetzner aligns with production workloads where response time consistency is a product characteristic users experience. API services where response time variability affects user perception. Interactive web applications under concurrent load. CI/CD pipelines where build time predictability matters for engineering workflows. At Hetzner's EU pricing, there is rarely a cost reason to accept the CPU variability that Contabo's model produces for these workloads.

Contabo's genuine use cases are narrow and real. Development and staging environments where CPU behavior doesn't need to match production. Backup infrastructure where storage density is the binding resource. Media servers and file storage where RAM and disk allocation drive the architecture. Batch processing jobs that run on a schedule where total throughput matters more than individual task latency. In these scenarios, Contabo's resource density exceeds what Hetzner provides at equivalent cost.

The failure pattern with Contabo is consistent: teams choose it for production web applications based on the RAM-per-dollar comparison, encounter CPU variability under concurrent load, and attribute the problem to other causes before tracing it to the oversubscription model. The workload profile assessment should happen before provisioning, not after a production incident. CPU oversubscription tends to become a limiting factor under production load for applications where users observe response times.

Hetzner's CCX dedicated cloud series allocates specific AMD EPYC cores without oversubscription. What you provision is what the workload receives, regardless of neighboring VM activity. Independent benchmarks confirm consistent CPU performance across time and load conditions. For compute-intensive or latency-sensitive workloads, this predictability is the product characteristic that matters most.

Contabo's CPU performance is variable by design. Under light overall server load, CPU availability can be adequate. Under heavy load conditions — when neighboring VMs compete for the oversubscribed CPU pool — individual VM CPU performance degrades. The degree of degradation depends on server-level conditions that are not visible to the customer. This variability is not a bug. It is the mechanism that makes the RAM and storage pricing possible.

For storage-heavy workloads, both providers offer NVMe SSD. Contabo's storage density per dollar is higher. Hetzner's storage performance is consistent with its dedicated compute model. For backup targets, media storage, and database workloads where disk throughput is the binding resource, both perform adequately — and Contabo's capacity per dollar becomes the relevant differentiator.

Contabo's headline resource numbers are more impressive than Hetzner's at every comparable price tier. More RAM, more storage, lower monthly cost. If a feature comparison table determined infrastructure decisions, Contabo would appear to lead this matchup consistently.

Hetzner's pricing reflects consistent dedicated vCPU performance. The premium over Contabo's entry plans is the premium for CPU predictability. For production workloads where response time consistency is a product requirement, that premium is not an infrastructure cost — it is an insurance cost against the CPU variability that Contabo's model introduces.

There is a meaningful operational asymmetry between the two providers: Hetzner supports hourly billing and immediate cancellation. Contabo applies setup fees on some plans and operates without a money-back guarantee. For teams that benchmark infrastructure before committing, this asymmetry makes parallel evaluation more costly with Contabo. The evaluation should happen with workload testing before provisioning, not after.

Hetzner aligns with production workloads where CPU consistency affects user experience: API services, web applications with concurrent users, CI/CD pipelines. You gain predictable dedicated vCPU performance at EU pricing that remains competitive by global cloud market standards. You give up Contabo's RAM-per-dollar density.

Contabo aligns with RAM-bound or storage-bound workloads where CPU variability is acceptable: development environments, backup targets, media servers, scheduled batch jobs. You gain resource density — the highest RAM and storage per euro in the VPS market. You give up performance predictability.

A practical diagnostic: identify whether your workload has concurrent users or time-sensitive CPU tasks. If yes, test on Hetzner. If the workload is primarily storage or RAM-driven with flexible latency, benchmark Contabo's resource allocation against the actual application requirements.

The decisive question is: does your workload have users or systems that observe CPU slowness?

If the answer is yes — because the application serves concurrent HTTP requests, runs interactive user sessions, or has response time SLAs — Hetzner's dedicated vCPU model aligns with that requirement. Contabo's oversubscription model introduces variability that will surface under those conditions.

If the answer is no — because the server stores files, runs overnight jobs, hosts a staging environment, or processes batch workloads without latency requirements — Contabo's resource density aligns with that profile. The CPU variability is irrelevant when no one is waiting for the CPU to respond.

The comparison becomes clear once the workload is profiled honestly. The uncertainty typically comes from evaluating infrastructure on resource numbers before assessing what the workload actually demands from the CPU.

You gain performance predictability with Hetzner. You give up resource density. With Contabo, the trade runs in reverse.