In managed vs unmanaged Kubernetes, a managed service runs the control plane for you, while a self-managed cluster leaves every layer to your team. On either model your team still owns workloads, access, backups and compliance. So team size and on-call cover decide the fit.
Last reviewed 2026-11-16 by Manish Sharma.
On a managed service, the cloud provider runs the control plane and keeps it available. That means the API server, etcd, the scheduler and the controllers. Amazon EKS on AWS and AKS on Azure work this way. An unmanaged cluster, also called self-managed, has a team that installs and runs all of it. That can happen in its own rooms or on rented cloud machines. Weighing managed vs self managed Kubernetes is a choice about who carries the run load. Also, help picking and running clusters on AWS or Azure is available if the team wants a second view.
Managed vs Unmanaged Kubernetes: Who Owns Each Layer
The table marks each layer as team, provider or shared. Shared means the provider supplies the part and the team still acts on it.
| Layer | Self-managed | Amazon EKS | AKS |
|---|---|---|---|
| 1. Physical hosts and data center | Team | Provider | Provider |
| 2. Control plane and etcd | Team | Provider | Provider, with etcd backed up automatically |
| 3. Control plane operation and availability | Team | Provider runs the control plane | Provider |
| 4. Version upgrades | Team plans and runs | Team starts control plane and node upgrades | Shared, team triggers and schedules |
| 5. Node operating system and patching | Team | Team | Shared, Microsoft ships node images |
| 6. Node provisioning and scaling | Team | Team, or shared with managed node groups | Shared |
| 7. Add-ons such as DNS and storage drivers | Team | Team upgrades | Shared |
| 8. Cluster networking plugin | Team | Shared, installed by Amazon EKS and upgraded by the team | Shared |
| 9. Access roles, secrets and network policies | Team | Team | Team |
| 10. Workload monitoring and alerts | Team | Team, who enables control plane logging | Team, with control plane logs and metrics from Microsoft |
| 11. Cluster backup and disaster recovery | Team | Team | Team |
| 12. Customer data and compliance | Team | Team | Team |
The Amazon EKS rows for control plane, upgrades, nodes, add-ons, networking and backup follow the AWS guide to cluster upgrades (no page date shown). Also, the EKS rows for hosts, node operating system, access and data follow the Security in Amazon EKS page (no page date shown). Finally, the AKS column follows Microsoft Learn on AKS support policies (page dated 27 August 2026). We last checked all three pages on 16 November 2026. Rows 9 to 12 do not move on either model.
Key Takeaway The provider takes the control plane, and the team keeps access, backups, data and compliance.
Managed vs Unmanaged Kubernetes Compared on Upgrades
A minor version on Amazon EKS gets 14 months of standard support, then 12 months of extended support. After the 26-month lifecycle, Amazon EKS upgrades a cluster still on that version automatically, as the AWS guide describes. The guide also says the team initiates upgrades for the control plane and for the data plane. The data plane covers the nodes and add-ons.
On AKS the supported window is N-2: the latest minor version and the two before it. AKS also offers Long Term Support, made up of one year of Community Support and one year of LTS. The Microsoft Learn page on supported Kubernetes versions in AKS sets this out (page dated 15 September 2026, read on 16 November 2026). From 1 June 2027, a cluster left on an unsupported version is upgraded automatically once a 60-day grace period ends.
On both services the team still picks the moment, reads the release notes and tests its workloads before and after. Under the self-managed model the team also tracks the upstream releases itself. A separate article lays out what running the control plane yourself takes week to week.
Managed vs Unmanaged Kubernetes Below the Control Plane
Worker nodes are where a managed service still hands the team a large share of the work. However, the two services differ in shape there.
On Amazon EKS, managed node groups automate the provisioning and lifecycle management of nodes. The AWS guide also points to them for simpler node upgrades. The team still starts each node upgrade and upgrades add-ons such as DNS and storage drivers. Amazon EKS installs the networking plugin on every cluster, so that row is shared. However, the guide says the service does not upgrade add-ons automatically, and the team starts those upgrades. Also, keeping the node operating system patched stays with the team, as the Security in Amazon EKS page lists it.
On AKS, Microsoft ships node images and repairs nodes, while the team sets node pool size and counts. Microsoft shares the add-on work, and the team applies a node image or picks an auto-upgrade channel.
Self-managed Kubernetes leaves all of this with the team. It builds, patches and replaces every node, and it also upgrades every add-on. That work is steady rather than dramatic, and it falls on the same few people each month.
What Stays With Your Team on Either Model
A managed service narrows the run load but does not hand over the security settings. The Microsoft table keeps access roles, secrets, pod security, network policies, application code, customer data and compliance with the customer on AKS. The Security in Amazon EKS page likewise names data sensitivity, access management and network controls as the customer’s side.

Six areas stay with the team on every model:
- access roles, secrets and pod security
- network policies
- application images
- persistent storage settings
- cluster backup and recovery
- workload alerts, customer data, compliance and audit
Each one is a setting or a record that a provider cannot choose for you. For a scored look at them, a separate article covers the access, workload and image marks for any cluster.
A managed service does not choose your access roles, secrets or network policies, so check them on day one.
When to Choose Each Model: A Decision Guide
Four inputs settle managed vs unmanaged Kubernetes, each with its own heading below. Signisys’s working view is that a managed service fits first where the team is small or on-call cover is thin. Self-managed fits where a rule or a site needs control of the control plane. The basis is that rows 1 to 3 of the table move to the provider and rows 4 to 8 become shared. So the choice between managed or self managed Kubernetes comes down to whether the team wants those rows back.

Advantages
Limitations
Team Size and On-Call Cover
Ask who answers a control plane alert at night today. If the answer is one engineer, a managed service removes that alert from the list. The provider carries rows 1 to 3.
Ask next how many people can run an upgrade without the others stopping work. A change that needs two or three people for a day is a signal about team size instead of the platform itself.
The article on running clusters yourself covers the people side of this. Here the test is simple. Count the named people who can carry each row marked team, and note where one name appears on many rows.
Write one name beside each row marked team. A single name on many rows shows where on-call cover is thin.
Compliance and Data Location in India and the GCC
Take four questions to the compliance owner before choosing. Which data must stay in which country? Does the managed service run in a region there when the decision is made? Check the provider’s own region list before you decide. If an audit needs control plane logs, are they available on the service? Does an auditor ask for host access, although a managed control plane does not give it?
We name no law or region here, because the rules differ by sector and change. Then the answers go into the record in the mixed estate section, next to the model chosen.
Rows 9 to 12 stay with the team on either model, so a compliance duty stays with the team.
Control Over the Control Plane
Some needs point to self-managed, and they show as signs rather than rules.
- The team needs custom control plane settings that a managed service does not expose.
- A site has no cloud region nearby, or sits on-premises or at the edge.
- A cluster is air-gapped, with no route to the provider.
- Vendor products ask for host access.
One sign is a reason to look closer. However, several signs together favor running the control plane in the team’s own hands, with the on-call cover to match.
The Cloud and Skills Already in Place
If the firm already runs on AWS or Azure, look first at the managed service on that cloud. Staff also know the console, the access model and the support route already.
The eks vs aks question then follows from the cloud in use, not from this choice. The table shows matching depth on both.
Skills count as much as the cloud. A team with strong systems and networking skills can run self-managed clusters well. In contrast, a team with strong application skills gains more from a managed service.
Teams still deciding whether the app needs a cluster at all should settle that first.
Key Takeaway Choose managed for a small team or thin on-call cover, and self-managed for a rule or site that needs the control plane.
Managed vs Unmanaged Kubernetes in a Mixed Estate
Signisys’s working view is that a firm can run cloud clusters on a managed service and keep on-premises or edge clusters self-managed. The basis is that the four inputs above differ from site to site, so one answer for every cluster is rare.
Keep one record per cluster with four lines:
- the model in use
- why that model was chosen
- who agreed to it
- what would reopen the choice, such as a team change or a new rule
Because it is written down, the record turns a one-time opinion into something a new engineer or an auditor can read. When a cluster changes model, the move of a first application, in order gives the sequence for the application side.
Signisys offers DevOps services for cluster builds and upkeep and uses the solution we use to modernize and watch applications alongside it. The firm also partners with AWS and Microsoft. In a working session, we then review your clusters against the layer table and the four inputs. No charge, with a fixed scope.
Managed vs Unmanaged Kubernetes: FAQ
Can a team move from self-managed to managed clusters later?
Does the provider upgrade a managed cluster without being asked?
Is a cluster on cloud virtual machines still self-managed?
Who gets the call at night on a managed cluster?
Security Insights, Delivered.
Practical, vendor-neutral analysis of threats, tools, and architecture decisions.
