September 2, 2026
What Docker Swarm Actually Costs to Run in 2026, Priced and Compared with Managed Kubernetes
The cost of a Docker Swarm is the cost of its nodes; there is no control-plane fee. Worked examples for 3, 5 and 10 nodes on published VPS prices, the EKS, GKE and AKS cluster fees you do not pay, electricity for self-hosted hardware at the Eurostat rate, and where the money actually goes.

Every cost article about container orchestration seems to know that teams save "40 percent" by switching, and none of them says from what. This one has no percentage in it. It has the published prices of the things you pay for, as of 10 September 2026 with a link to each, and the arithmetic laid out so you can redo it with your own numbers.
1. What a swarm costs: its nodes, and nothing else
Swarm mode is part of Docker Engine. The managers are ordinary machines that also run the control plane, the routing mesh is in the engine, secrets and configs are in the managers' Raft store, and load balancing is a virtual IP the engine maintains. There is no cluster fee, no managed control plane, no separately billed load balancer unless you choose one, and no per-node agent to license.
So the monthly cost of a swarm is:
nodes × node price
+ block storage beyond what the nodes include
+ egress beyond what the nodes include
+ any managed service you choose to add (a load balancer, a registry, backups)
For small and mid-size swarms the first line is usually the whole bill.
2. What managed Kubernetes adds: the control-plane fee
The three hyperscalers charge for the control plane by the cluster-hour, before any node runs. As of 10 September 2026:
| Provider | Control plane fee | Per month at 730 h | Source |
|---|---|---|---|
| Amazon EKS | $0.10 per cluster per hour; $0.60 per hour once the Kubernetes version is in extended support | $73, or $438 in extended support | EKS pricing |
| Google GKE | $0.10 per cluster per hour, Standard and Autopilot alike; a $74.40 monthly credit per billing account covers one zonal or Autopilot cluster | $73 for each cluster after the first zonal one; regional clusters get no credit | GKE pricing |
| Azure AKS | Free tier: no cluster fee, no financially backed SLA, not recommended above 10 nodes. Standard tier: $0.10 per cluster per hour. Long-term support: $0.60 per hour | $0, $73, or $438 | AKS pricing tiers and the Azure retail price API |
Two things are easy to miss in those rows. The extended-support price is what you pay for not upgrading Kubernetes on schedule, six times the base fee, and it is the number a team on an old version is actually paying. And the free tiers are for one cluster: a staging cluster and a production cluster on GKE means one is paid.
What the fee does not include is the share of every node that Kubernetes takes for itself: kubelet, kube-proxy, the CNI agent, the metrics and logging agents that a managed offering installs. The providers publish reservation formulas rather than a single number, so it is not tabulated here; it is real, it grows with node count, and on a 2 GB node it is not small.
3. Worked examples on published node prices
The examples use OVHcloud's published VPS prices, from the VPS page as of 10 September 2026, in USD, traffic unlimited, daily backup included:
| Plan | vCores | RAM | NVMe | Per month |
|---|---|---|---|---|
| VPS-1 | 2 | 4 GB | 40 GB | $4.54 |
| VPS-2 | 4 | 8 GB | 75 GB | $8.50 |
| VPS-3 | 6 | 12 GB | 100 GB | $12.32 |
| VPS-4 | 8 | 24 GB | 200 GB | $23.37 |
Hetzner Cloud is the other common European choice. These are its regular-performance and general-purpose plans as of 10 September 2026, EU locations (Falkenstein, Nuremberg, Helsinki), in EUR, including an IPv4 address and excluding VAT, with a monthly price cap per server:
| Plan | vCPU | RAM | NVMe | Per month |
|---|---|---|---|---|
| CPX12, shared | 1 | 2 GB | 40 GB | €11.99 |
| CPX22, shared | 2 | 4 GB | 80 GB | €19.99 |
| CPX32, shared | 4 | 8 GB | 160 GB | €35.99 |
| CPX42, shared | 8 | 16 GB | 320 GB | €69.99 |
| CCX13, dedicated | 2 | 8 GB | 80 GB | €43.49 |
| CCX23, dedicated | 4 | 16 GB | 160 GB | €86.49 |
| CCX33, dedicated | 8 | 32 GB | 240 GB | €138.99 |
The two providers are not priced alike: at the 4-vCPU, 8 GB size the OVH plan is a third of the Hetzner one. Hetzner's list is dedicated NVMe on newer hardware with a per-server monthly cap, OVH's is a VPS line with unlimited traffic and a daily backup included, and the right one depends on what you run. The point of showing both is that the arithmetic is the same on either.
| Swarm | Nodes | At OVHcloud | At Hetzner | The fee the same nodes carry under EKS or GKE |
|---|---|---|---|---|
| Homelab or a small product | 3 nodes at 4 vCPU and 8 GB (one manager, two workers) | 3 × VPS-2, $25.50 | 3 × CPX32, €107.97 | plus $73 control plane |
| Small production | 3 managers at 2 vCPU and 4 GB, 2 workers at 4 to 6 vCPU | 3 × VPS-1 + 2 × VPS-3, $38.26 | 3 × CPX22 + 2 × CPX32, €131.95 | plus $73 |
| Mid-size production | 3 managers at 2 vCPU and 4 GB, 7 workers at 8 vCPU | 3 × VPS-1 + 7 × VPS-4, $177.21 | 3 × CPX22 + 7 × CPX42, €549.90 | plus $73, or $438 on an old version |
The last column is deliberately not "the price of EKS": you cannot attach an OVH or Hetzner node to EKS, and EC2 or GCE nodes are priced differently. It isolates one thing, the fee that exists on one side and not the other. For the three-node swarm at OVH, that fee is nearly three times the whole bill; at Hetzner it is about two thirds of it. For the ten-node swarm it is 40 percent on top at OVH and about 13 percent at Hetzner. At a hundred nodes it is noise on either, which is the honest version of "Kubernetes makes sense at scale".
4. Self-hosted hardware: the electricity line
On hardware you own, the recurring cost is power, and the arithmetic is:
watts × 24 h × 365 / 12 ÷ 1000 × price per kWh = monthly cost
Eurostat's most recent figure for household electricity is €0.2896 per kWh, the EU average for the second half of 2025, ranging from €0.1082 in Hungary to €0.4042 in Ireland (Eurostat). At that average:
| Cluster | Draw | Per month |
|---|---|---|
| Three Raspberry Pi 5 boards with NVMe and PoE, as built in our Pi guide | under 40 W total | under €8.45 |
| Three mini PCs at an assumed 10 W average each | 30 W | €6.34 |
| One repurposed tower at an assumed 100 W average | 100 W | €21.14 |
The wattages for the mini PCs and the tower are assumptions to plug your own meter reading into, not measurements; the Pi figure is the one from our own build. What the table shows is the shape: a three-node swarm of small boards costs less per month to power than one old server, and both cost less than the control-plane fee above. Hardware amortisation is your number to add, and a €120 mini PC over three years is €3.33 a month.
5. Where the money actually goes on a swarm
Because the bill is the nodes, waste on a swarm is nodes you are paying for and not using. The five places it hides:
- Reservations sized by fear. A
reservations.memory: 4Gon a service that uses 600 MB blocks 4 GB on a node for nothing, and it is the usual reason a task sits in Pending while the node reports free memory. Size reservations from a week of metrics, not from the Compose file's first draft. - Managers doing nothing. Three drained managers on the same plan as the workers are three workers' worth of money for a control plane that needs a fraction of it. Put managers on the smallest plan that keeps Raft comfortable; VPS-1 at $4.54 or CPX22 at €19.99 is fine for a swarm of this size, and a dedicated-vCPU CCX13 at €43.49 is the upgrade if manager latency ever shows in
docker node ls. - Images and build cache on every node. A node with a full disk is a node that fails tasks, and image layers are what fill it.
docker image prune -a --filter "until=168h"on a schedule keeps a week; do not add--volumes, which on a swarm node deletes the data of any service that is not running there at that moment. - Batch work on always-on nodes. A nightly job does not need a node all day.
docker service scale encoder=0from a scheduled service and=1when the window opens; in the cloud, that node can be a smaller plan or none. - Storage you replicate twice. Backups on the same provider as the nodes are one invoice away from being lost with them; backups off-site are a second bill, and the right one. What is not worth paying for is a distributed filesystem under a database: a pinned local volume is faster and cheaper.
None of these is a Swarm trick. They are what "the bill is the nodes" makes visible, because there is no other line to hide behind.
6. The tooling line
SwarmCLI's Starter tier is free on swarms of up to three nodes with every feature; larger swarms are priced per node on the pricing page, which quotes in the currency you pay in. It is the one line in this article we set, so it is the one we will not put a number on here: read it there, where it is current.
Common concerns, answered
Does Docker Swarm have a control-plane fee? No. Swarm mode is part of Docker Engine and the managers are ordinary nodes you already pay for. Managed Kubernetes charges for the control plane separately: EKS and GKE at $0.10 per cluster per hour, about $73 a month, and AKS at $0.10 per hour on its Standard tier, with a Free tier that carries no SLA.
How much does a 3-node Docker Swarm cost per month? On published VPS prices, three 4-vCPU, 8 GB nodes cost about $25.50 a month at OVHcloud or €107.97 at Hetzner as of September 2026, and that is the whole bill: no cluster fee and no load-balancer fee. On self-hosted hardware drawing 10 W per node, the electricity is about €6.30 a month at the EU average rate.
Is Docker Swarm cheaper than Kubernetes? For the same nodes, yes, by the control-plane fee and by the resources the Kubernetes system components take on every node. Whether that matters depends on scale: $73 a month is the entire cost of a small swarm and a rounding error on a hundred-node one. The honest comparison is in the worked examples in this article, with sources.
What to do with this
Count your nodes and multiply. If you are choosing between a swarm and a managed Kubernetes cluster for a handful of services, the control-plane fee alone is the size of the swarm's whole bill, and the comparison post covers the rest of the trade. If you already run a swarm, the five items in section 5 are the audit, and the best-practices checklist is where they live alongside everything else.
All prices are as published by the providers on 10 September 2026. We check them whenever we revise this page, and the date at the top tells you how current they are.