Dora NodaOn June 15, 2026, Hetzner nearly tripled the price of its cheapest dedicated-vCPU cloud server — from €15.99 to €42.99 a month — and quietly rewrote the cheapest decision in self-hosting. For years the reflex answer to "Contabo or Hetzner?" was Hetzner on performance per euro, Contabo on raw gigabytes per euro. After the hike, one of those answers is wrong: at the dedicated tier, Contabo now undercuts Hetzner by €8.59 a month while shipping three times the RAM. HostAdvice's September 2026 comparison names Contabo the winner for most buyers, and for once the affiliate-review verdict agrees with the spreadsheet.
But "which host wins" is the wrong question for a fleet. Nobody runs a control plane, elastic workers, build machines, and edge nodes on the same plan. The right question is which vendor each role rents from — so here is the verdict up front, role by role, before a single paragraph of price tables:
| Fleet role | Rent it from | Why, in one line |
|---|---|---|
| Control-plane nodes (long-lived, steady) | Hetzner CX23 | Steadier shared cores, private networks, and CAPH-native API control at €5.49 |
| Elastic workers (autoscaled, idle hours) | Hetzner (hourly) | Hourly billing with a monthly cap; a half-idle node costs roughly half |
| Always-on memory-heavy workers | Contabo VPS 30 | 24 GB and 8 vCPU for €14 — Hetzner can't touch that memory per euro |
| Storage-heavy build workers | Contabo | 200 GB NVMe at €14, or swap to 400 GB SSD at the same price |
| APAC / India edge nodes | Contabo | Tokyo, Sydney, and Mumbai DCs Hetzner simply doesn't have |
| Dedicated-core workloads | Contabo VDS S | €34.40 with 24 GB vs Hetzner CCX13 at €42.99 with 8 GB |
That table is the whole post in miniature. Everything below is the evidence: the September 2026 numbers behind each cell, two worked fleet totals, and the four conditions that flip any single row.
The September 2026 price shape, tier by tier
Two sources pin every figure below to a date: HostAdvice's September 2026 head-to-head and HowToHosting.guide's post-hike rerun, which verified pricing against contabo.com and Hetzner's own adjustment notice on July 9, 2026. The June 15 hike is the context that makes September look the way it does — Hetzner's shared CX23 line rose 38% (€3.99 to €5.49), ARM rose 33%, and the dedicated CCX line jumped 169–173%. Pre-June servers keep their old rates, so this is strictly a new-orders comparison.
At entry, both vendors sit near €5.50 — and the spec gap is almost comical:
| Contabo Cloud VPS 10 | Hetzner CX23 | |
|---|---|---|
| Price | €5.50/mo | €5.49/mo |
| RAM | 8 GB | 4 GB |
| vCPU | 4 shared cores | 2 shared cores |
| Disk | 75 GB NVMe (or 150 GB SSD, same price) | 40 GB |
| Port | 200 Mbit/s | 1 Gbit/s-class backbone |
| Billing | Monthly, flat renewal | Hourly with monthly cap |
Same money, double the memory, double the cores, nearly double the disk. Climb the range and Contabo holds the lead: VPS 20 at €7.50 (12 GB), VPS 30 at €14 (24 GB, 8 vCPU, 200 GB NVMe), VPS 40 at €25 (48 GB). The dedicated tier is the headline flip — Contabo's VDS S gives 3 physical AMD cores, 24 GB RAM, and 180 GB NVMe for €34.40, against Hetzner's CCX13 at €42.99 for 2 vCPU and 8 GB. A year ago that matchup went the other way.
Hetzner's one uncontested value pick is ARM: the CAX11 gives 4 GB on 2 Ampere cores for €5.99, and Contabo offers no ARM at all. For ARM-native workloads it's cheap, efficient, and consistent — the exception that still beats Contabo on efficiency per euro.
Worked fleet math: two reference fleets, two monthly totals
Specs per euro only matter once they're assigned to roles. Take a typical small self-hosted PaaS fleet — 3 control-plane nodes plus a worker pool — and cost it both ways, all nodes 24/7, EU region:
Fleet H (all-Hetzner): 3× CX23 control + 4× CX23 workers = €38.43/mo. Seven nodes, 28 GB of RAM total, every node behind hourly billing, private networking, and an API that Cluster API Provider Hetzner (CAPH) speaks natively.
Fleet C (all-Contabo): 3× VPS 10 control + 2× VPS 30 workers = €44.50/mo. Five nodes, 72 GB of RAM total — roughly 2.6× the memory for 16% more spend.
Neither total tells the story alone, because the two fleets buy different things. Fleet C buys memory: two VPS 30 workers hold a tenant pod count that would need twice the Hetzner nodes. Fleet H buys steadiness and optionality: steadier shared cores, and workers that can shrink to zero cost when idle.
That last point deserves its own math, because it's the core of Hetzner's surviving case. Hetzner bills by the hour with a monthly cap, so a node's cost scales with the hours it runs, up to €5.49. Contabo bills the full month whether the node runs 730 hours or 100. At 100% utilization the entry tier is a price wash (€5.49 vs €5.50) that Contabo wins on specs; at 50% utilization — a preview-environment worker that only runs on weekdays, a batch node that sleeps overnight — the Hetzner node costs roughly €2.75 against Contabo's fixed €5.50. The breakeven is blunt: if a worker runs less than about half the month's hours, Hetzner hourly wins outright no matter how fat Contabo's plan is. If it runs 24/7, Contabo wins on capacity and Hetzner has to justify itself on steadiness instead.
When the answer flips: four sensitivity rules
A single worked example is one point; fleet planning needs the range. Four variables flip individual rows of the verdict table, each with a number attached:
1. Idle hours flip workers to Hetzner. The breakeven above is the whole rule. Autoscaled MachineDeployment workers, PR preview nodes, and overnight batch capacity all spend much of the month at zero — exactly the shape hourly billing discounts and monthly billing punishes. Contabo has no hourly tier and no scale-to-zero story; every idle hour is billed at the full monthly rate.
2. Traffic volume — and traffic geography — flips everything to Contabo. Hetzner's headline 20 TB of included traffic applies in Europe only. US servers include 1 TB, Singapore just 0.5 TB, with overage at €1/TB in the US and a steep €7.40/TB in Singapore. Contabo's traffic is effectively unlimited (fair-use throttled; HostAdvice cites a 32 TB figure). A busy site outside the EU can flip the entire comparison on this one line: a Singapore node pushing 5 TB pays over €33 in Hetzner overage alone.
3. Dedicated cores flip to Contabo. This is the June hike's purest effect. VDS S at €34.40 (3 physical cores, 24 GB) undercuts the CCX13 at €42.99 (2 vCPU, 8 GB) on price and triples it on RAM. If your workload needs guaranteed cores — a latency-sensitive API, a database primary — the vendor that used to own "cheap dedicated cores" no longer does. The one carve-out: bare metal reverses the story again, since Hetzner's Server Auction starts near €35/mo against Contabo's cheapest physical box at around USD 110 a month.
4. Declarative tooling flips control planes to Hetzner. Hetzner ships a real cloud API, an official Terraform provider, managed load balancers, private networks, firewalls, floating IPs, and free snapshots — the primitives CAPH turns into declarative MachineDeployments. Contabo offers a REST API and the cntb CLI but no official Terraform provider or SDK, so a Contabo-backed fleet hand-rolls the automation layer Hetzner inherits. That gap is measured in eng-hours, not euros, but it's real money for a small team.
Region math beats vendor math
Once latency enters, the per-euro spreadsheet stops mattering. Contabo operates 9 regions across 11 data centers — Germany, the UK, three US regions, Singapore, Tokyo, Sydney, and Mumbai. Hetzner runs six locations in four countries — two German sites, Helsinki, Ashburn, Hillsboro, and Singapore. For an audience in Japan, Australia, or India, Contabo puts a box nearby and Hetzner cannot.
HostAdvice's ping tests put numbers on the distance: Tokyo at 8.9 ms on Contabo vs 252.7 ms on Hetzner, Mumbai at 1.7 ms vs 131.8 ms, Jakarta at 2.0 ms vs 256.9 ms. No amount of steadier cores recovers a 250 ms round-trip deficit from Falkenstein. And Hetzner's only Asia-Pacific site carries that 0.5 TB traffic cap, so even the Singapore compromise gets expensive for a busy APAC service fast. If your users are in Tokyo or Sydney, the vendor decision is made by the map, not the price list.
The honest counterweight is CPU consistency. Contabo's cheap cores are shared, and user reports put CPU steal at 20–40% during peak hours on busy hosts, with time-to-first-byte swinging through the day. Hetzner's shared cores wobble far less, and its dedicated line removes the noisy-neighbor problem outright. Independent monitoring lands Contabo near 99.91% uptime against a 99.9% SLA — fine for staging, batch, and bursty tenant workloads, but a real tax on a checkout page. Support follows the same split: Contabo is ticket-only with a slow-reply reputation, while neither vendor will mistake you for a managed customer.
The tooling tax nobody puts in the spreadsheet
The deepest fleet-level difference isn't price or regions — it's what each vendor assumes you are. Hetzner assumes you're building a cloud: API-first provisioning, Terraform, private networks, load balancers as API objects, hourly meters you can reconcile against autoscaler decisions. That's why CAPH exists for Hetzner and why a Hetzner-backed fleet's node lifecycle can be a GitOps-managed MachineDeployment with cluster-autoscaler doing the hourly math for you.
Contabo assumes you're renting servers: generous fixed boxes on 1- to 12-month terms, a control panel, a REST API and CLI for the basics, and flat pricing where the signup rate is the renewal rate. There is no hourly meter for an autoscaler to optimize against, no official Terraform provider to declare a fleet in, and no private-network primitive to isolate tenants at the network layer. You can absolutely run Kubernetes on Contabo — full root on every VPS — but you wire the cloud layer yourself, and every hour of that wiring is time not spent on the platform.
So the real fleet question isn't "Contabo or Hetzner" but "where does each role's constraint bind." Elastic, autoscaled, API-driven roles bind on the meter and the tooling: Hetzner. Always-on, memory-hungry, traffic-heavy, or APAC-serving roles bind on capacity and map position: Contabo. Rent by role, not by brand — and re-run the numbers after every price-list change, because June 2026 proved the reflex answer has an expiry date.
Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own. Star the repo on GitHub or deploy your first app today.
