Skip to main content

Coolify Cloud at $5 for 2 Servers vs a Raw Hetzner CX22 at €4.35: What That Subscription Actually Buys

14 min readDora NodaDora Noda
Share

Coolify Cloud will manage two of your servers for $5 a month. The cheapest Hetzner box it usually manages costs about €4.35 a month and ships with 20TB of traffic. One is a dashboard subscription. The other is the computer underneath it.

If you are comparing them as two prices for the same thing, you are already miscategorizing one of them. The $5 does not buy you a server — it rents you the control plane that talks to a server you already pay for somewhere else. Get that distinction wrong and every "Coolify vs. Hetzner" spreadsheet you build will quietly double-count or under-count the same line.

This post is the line-by-line teardown that fixes the math: what $5 actually covers, what €4.35 actually covers, and what the real monthly total is at the fleet sizes teams actually hit.

Why it matters now: Coolify just shipped v4.0.0 on April 27, 2026 and sits at roughly 57,000 GitHub stars, the clear leader among open-source PaaS panels. Its hosted offer — Coolify Cloud — stayed flat at $5/month for two managed servers ($3 per extra) while the hardware it usually fronts repriced around it. Any team weighing "just use Coolify" against "just own the box" needs the same two numbers on one page.

TL;DR: The real monthly total at 1, 2, 5, and 10 servers

This is the table the title promised. All totals are monthly. Hetzner CX22 is taken as €4.35 (~$4.70 at recent EUR/USD ~1.08) with 20TB included traffic; Coolify Cloud is $5 for the first two servers, then $3 per server after that.

Fleet sizeHetzner CX22 hardware (× servers)Coolify Cloud feeCombined total / monthCombined annualCloud fee as % of total
1 server€4.35 (~$4.70)$5.00~$9.70~$11652%
2 servers€8.70 (~$9.40)$5.00~$14.40~$17335%
5 servers€21.75 (~$23.50)$14.00 ($5 + 3×$3)~$37.50~$45037%
10 servers€43.50 (~$47.00)$29.00 ($5 + 8×$3)~$76.00~$91238%

Raw box alone for comparison: 10 Hetzner CX22s with no management layer is ~$47/month. The Cloud subscription adds ~$29 on top at that size — about 62% overhead on the hardware line, before you count the operator time you still supply yourself.

Takeaway before the detail: at one server, the management fee is the bill. At ten, it is a steady ~38% surcharge on top of cheap hardware, not a rounding error — and it still does not include auto-provisioning a new machine for you.

What $5 actually buys — and what it explicitly does not

Coolify Cloud's pricing page is short enough to quote in full: base fee $5/month includes up to two connected servers, then $3/month per additional server. There is no per-seat charge, no per-app charge, and no bundled compute. You bring the servers — Hetzner, Contabo, a homelab box behind Tailscale, whatever has Docker and SSH.

What that fee covers:

  • A managed control plane. You do not run the Coolify panel itself. Coolify hosts it, updates it, and keeps the dashboard, API, and build workers online. Self-hosting Coolify free on your own VPS is the alternative, and then you own the panel's uptime, backups, and upgrade from v3 to v4 yourself.
  • Automatic panel updates. v4.0.0 landed in April 2026 with breaking changes from v3; Cloud tenants got it without running a migration script on their own box at 2 a.m. Self-hosters ran the updater themselves and dealt with any stuck deploy.
  • Dashboard, git-push deploys, and built-in primitives. Git integration, Docker builds, preview deploys, managed Postgres/Redis via one-click services (280+ templates), Traefik with Let's Encrypt, and Docker Compose support. This is the Heroku-like experience that earned Coolify ~57,000 stars and ~575 contributors.
  • Support and multi-server routing. One dashboard can push to multiple destination servers, so two $5-included boxes can each host a different subset of apps and databases behind the same UI.
  • No infrastructure. This is the line people misread. Coolify Cloud does not give you a server, disk, bandwidth, or IP. The €4.35 box, the snapshot storage, the floating IP — every euro of hardware is a separate Hetzner invoice.

What it does not buy:

  • No machine provisioning. Coolify connects to a server you already created via Hetzner Console, Terraform, or hcloud CLI. It does not call the Hetzner API to create a CX22 for you and does not replace a missing one if nbg1 says no stock.
  • No fleet lifecycle. There is no declarative Machine or MachineDeployment — no "desired state is 5 nodes, go make it so" loop. Scaling from two to five servers is you opening Hetzner, creating three CX22s, and clicking Connect in Cloud for each one.
  • No autoscaling of servers. App-level replicas and Docker Swarm placement are manual. Kubernetes-style autoscaling with Cluster Autoscaler or Karpenter is out of scope — Coolify targets Docker, not an orchestrator that owns the node pool.

What €4.35 actually buys — the box underneath

The Hetzner CX22 is the entry shared-vCPU line that most Coolify tutorials default to because it is cheap enough to leave on 24/7 while idle apps spin down to zero:

  • 2 vCPU, 4GB RAM, 40GB NVMe SSD, 1 IPv4, 20TB monthly traffic — billed hourly (~€0.006/hour) or monthly at a flat €4.35 (some price tables still show the pre-hike €3.79; the April 2026 repricing moved the entry CX line from €3.49–€3.79 to ~€4.15–€4.51 depending on the catalog snapshot — the delivery here uses the €4.35 mid-point the TODO anchored to).
  • 20TB included traffic is the pricing story. At $0.15/GB on Vercel, 20TB metered would be ~$3,000. On Cloudflare Workers, stateless egress is free but storage is now metered since January 2026. On Hetzner it is just there — same price whether you push 200GB or 12TB of container images and preview deploys in a month. The tradeoff is you own the machine that sends those bytes, including the 2 a.m. page if the NVMe fills up.
  • Hourly billing + API + Terraform. CX22s are real IaaS primitives: create, snapshot, firewall, floating IP, placement group — all via hcloud or Terraform. That API surface is exactly what a Cluster API provider like CAPH drives to make "add a server" a kubectl apply instead of a dashboard click.

The fleet math, worked line by line

The TL;DR table above is the summary. Here is how each row is built and what it implies if you read the invoices literally.

Assumptions: All servers are CX22 at €4.35. EUR/USD 1.08 is used only to put both invoices in one currency for comparison; Hetzner bills in euros, Coolify Cloud in dollars, so a European tenant sees ~€4.64 vs $5 at today's FX as a wash rather than a discount. No snapshot, backup, or floating-IP extras are included — those add €0.0119/GB for snapshots on either side.

  • 1 server. You pay $5 to manage €4.35 of hardware. The management fee is 52% of the combined bill and larger than the computer. This is why Coolify Cloud's own pitch favors self-hosting the panel if you truly have one box and are comfortable updating it — free, plus you keep the $60/year. The hosted value at one server is "I never want to update a panel," not "I saved money."

  • 2 servers. The sweet spot the $5 is priced for. Two CX22s are €8.70 of hardware; the Cloud fee is $5 total, or $2.50 per managed server. Combined ~$14.40/month is still cheaper than a single Railway Pro workspace ($20 base before usage) or a Vercel Pro seat ($20/seat before Active CPU and bandwidth), but the saving comes from cheap hardware, not from a cheap manager.

  • 5 servers. Three extra servers at $3 each pushes the Cloud fee to $14. Combined ~$37.50/month, ~$450/year. The fee is now 37% of the total. You have five boxes to patch, five Docker daemons to keep healthy, and five SSH keys to rotate — all still manual. The $14 bought you one pane of glass, not five fewer sysadmin tasks.

  • 10 servers. Eight extra at $3 = $24 + $5 base = $29 Cloud fee on ~$47 of hardware. Combined ~$76/month, ~$912/year. At this size the per-server Cloud cost ($2.90) is finally approaching "background noise" — but the per-server operational cost is not. Ten CX22s at 20TB each is 200TB of included traffic you will never use; what you will use is ten Traefik configs, ten Let's Encrypt renewals, and ten places to debug a failed deploy. That work is not on the invoice but it is on the calendar.

Sensitivity check. If you followed an old tutorial pricing CX22 at €3.79, every total drops $0.60 per server ($6 at ten servers). If Hetzner reprices again — the June 15, 2026 hike already pushed CCX/CPX dedicated lines 113–175% — CX22's shared-vCPU line has so far moved far less than the dedicated lines, but "entry tier escaped the big hike this time" is not a price guarantee. Re-run the sheet at €5.50 and the Cloud fee's percentage shrinks while the absolute bill still rises.

What still breaks at server three

Coolify's ~57,000 stars measure how many teams love the single-server experience. That popularity does not answer the question that arrives the afternoon you need a third machine — because the second machine was not the hard part; the third is where you wanted a fleet, not a longer list of hosts.

Three concrete gaps that show up right there:

  • No declarative machine lifecycle. Coolify knows about servers you registered. It does not own a desired-state loop for them. Cluster API does: a MachineDeployment that says replicas: 5 will keep trying to make five Machines exist, replacing a dead one and rolling one at a time on update, without a human clicking Connect. A workshop that abstracts away machine creation entirely — bex scale --nodes 5 — can do the same behind a simpler CLI, but the invariant is the same: the platform, not the operator, closes the loop.

  • No fleet-wide scheduling or autoscaling. Coolify deploys a service to a destination server you pick. Docker Swarm mode is available but remains manual clustering. There is no Cluster Autoscaler watching Pending pods and adding a CX22, no Karpenter picking the cheapest fitting instance type, no bin-packing of preview environments onto spare capacity. When traffic spikes, you add a server. When it drops, you remove one. The platform watches; it does not react.

  • No provider-level fallback. Hetzner's June 26, 2026 capacity restriction — new CX22 orders refused at specific locations for a random subset of accounts with no exception queue — hit Coolify-managed fleets exactly as hard as bare Hetzner users, because the API call that fails is the Hetzner one, not Coolify's. A CAPH-backed fleet hedges this with multi-location MachineDeployments (say fsn1 + nbg1 + hel1) and multi-type templates so a scale-up can land on CPX21 if CX22 is out of stock. Coolify has no such fallback because it never owned the provisioning call.

None of this is a critique of Coolify's design goal. "Manage the box you already have, beautifully" is a real job and Coolify is the best open-source answer for it — Dokploy (~34,000 stars, still pre-1.0 at v0.29.x), CapRover (2017-era, stable but dated UI), and Deploynix all orbit the same single-server gravity well for a reason: most teams really do start with one box. The gap is not that the tool is missing a feature. It is that the job changes at three boxes, and the tool optimizes for the job before the change.

Where a Cluster-API fleet differs — and why bex bundles it

Bex.co's bet is the opposite seam: assume the fleet, not the single box, is the unit you will eventually need, and build the control plane so the first server is already a fleet of one.

CapabilityCoolify Cloud ($5 + $3)Raw Hetzner CX22 (€4.35)Bex / Cluster API fleet
Provision a new nodeYou create it in Hetzner, then Connect ithcloud server create or Terraformbex scale --nodes 5 / MachineDeployment declarative
Lifecycle (replace, roll, heal)Manual: you notice, you replaceManual: you script itController loop: dead Machine → new Machine automatically
Multi-location fallbackNone (operator picks location)Manual (you retry elsewhere)Built-in: fallback locations/types in provider config
AI-agent operationMCP server exists (community, 42 tools) but drives the panel, not the fleetAPI + Terraform (agent can call both)Render-compatible API + open MCP server on day one; agents call the same API humans do

The pricing implication is structural: Coolify Cloud is a subscription on top of the hardware bill, forever, because the hardware lifecycle stays outside the panel. A Cluster API fleet folds that lifecycle into the platform — the Hetzner bill is still there, but there is no second meter that grows per server for the privilege of adding a server. Bex's own cost pitch is not "cheaper than €4.35" — nothing is — but "one bill for the hardware plus a platform that already knows there will be a second, third, and tenth box."

That does not make Coolify the wrong pick. If you run a homelab, a side project, or a small agency that will credibly stay on one or two boxes for the next year, $5/month to never update a panel is a better deal than rebuilding your platform around a fleet you do not yet have. If you can already see the ticket where someone asks "can we add a staging fleet in hel1 without touching prod in nbg1?" — that ticket is cheaper to answer when the answer is already a MachineDeployment and not a new Hetzner project plus three Connect clicks.

Which should you pick? A one-paragraph decision guide

You areBest fitWhy
Solo or 1–2 boxes, side projects and client sitesSelf-host Coolify free or Coolify Cloud $5Fewest moving parts; Docker + Traefik + git push covers the job; Cloud is worth it if you hate panel updates
5 apps on 2–3 boxes, small team, no on-call rotationCoolify Cloud today, plan the fleet migrationCloud's $5–$14 is trivial vs. a Vercel seat; start sketching where MachineDeployments will live when you outgrow manual Connect
Growing fleet, or you already got a "no capacity" errorBex / Cluster API fleetDeclarative nodes, autoscaling, and provider fallback are not features on a roadmap — they are the premise

One closing number to anchor the choice: 10 CX22s alone are ~$47/month. The same 10 behind Coolify Cloud are ~$76. The same 10 behind a platform that provisions them declaratively is ~$47 plus whatever the platform itself costs to run — typically one small management node you already had. The $29 difference is not large in absolute dollars — it is large as a signal of who owns the hard part. If the hard part for your team is clicking Deploy, pay the $29 and ship. If the hard part is keeping ten machines the right ten machines, own the loop.

Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own. Cluster API owns the fleet so you don't manually Connect every new box. Star the repo on GitHub or deploy your first app today.

Related articles

Run this on infrastructure you own

bex is the open-source, AI-native Render alternative — push a git repo and get a running HTTPS service on your own machines.

Get started with bex