DigitalOcean managed kubernetes: free control plane, predictable per-second billing

If you've ever tried to set up Kubernetes on your own, you already know the dirty little secret of the cloud native world: the cluster itself isn't the hard part. The hard part is keeping the control plane alive, patching it at 2 a.m., making sure etcd doesn't quietly corrupt itself, and convincing yourself that running three masters in three regions is a reasonable use of a Saturday afternoon.

That's the whole reason managed Kubernetes exists. And when people go searching for "digitalocean managed kubernetes," they're usually not asking an abstract question about architecture. They're asking a very practical one: can I get a real Kubernetes cluster, with a real managed control plane, without signing up for a six-page pricing spreadsheet and a sales call that ends with "let's talk about your enterprise commitment"?

The short answer, after spending some time with DigitalOcean's offering, is yes — and the way they handle the billing is actually one of the more honest things in this corner of the industry.

What DOKS actually is, in plain terms

DigitalOcean Kubernetes — DOKS to its friends — is a fully managed Kubernetes service. DigitalOcean runs the control plane for you: the API server, scheduler, controller manager, etcd, all of it. You don't SSH into it, you don't patch it, you don't lose sleep over it. You just create node pools, drop your workloads on them, and go back to whatever you were actually trying to ship.

It speaks standard Kubernetes, so anything you've written for kubectl, Helm, Argo, or whatever your toolchain of choice happens to be will work without translation. It also plugs into DigitalOcean's own load balancers, block storage volumes, Container Registry, and — increasingly relevant — their GPU Droplets if you're doing inference workloads alongside the regular stuff.

The thing that catches people's attention, though, is the pricing model. Or rather, the absence of a separate pricing line item for the control plane.

The control plane situation

Here's where DOKS diverges from the bigger-name hyperscalers. The managed control plane is included at no cost. Not "free tier with caveats," not "free for the first N clusters" — it's just part of the service. You pay for the worker nodes, the storage, the bandwidth you use, and that's the deal.

If you need high availability on the control plane itself — multiple replicas spread across availability zones, so a single control plane hiccup doesn't take your cluster API down — that's a flat $40 per month, prorated hourly. So you can run a dev cluster on the free single-replica control plane during the week, and it's genuinely free on the control plane side, then flip on HA for production without re-architecting anything.

For a lot of small teams and side projects, that free control plane is the difference between "I'll just use Docker Compose" and "fine, let's actually use Kubernetes." You can 👉 spin up a DigitalOcean Kubernetes cluster and see this for yourself; the math gets friendly fast when there's no per-cluster management fee eating into your month.

Worker nodes: billed per second, capped at 28 days

This is the part I genuinely like, and I don't say that about many billing pages.

Worker nodes in DOKS are just Droplets under the hood, and they're billed per second at the same rate as regular Droplets. The entry point is $12 per month for a basic node, and it scales up through General Purpose, CPU-Optimized, Memory-Optimized, and GPU plans depending on what you're running.

Two details make this more than a footnote:

First, there's a 28-day billing cap per month. If your cluster runs continuously through a 31-day month, you only pay for 28 days — the longest 672 hours, specifically. Those extra three days are effectively free. It's a small thing, but it's the kind of small thing that compounds across a year of autoscaling.

Second, the per-second billing means autoscaling actually saves you money rather than just feeling like it should. If your traffic drops at midnight and your node pool scales from 15 nodes down to 3, you stop paying for those 12 nodes the moment they're destroyed — not at the end of the hour, not at the end of the day. For workloads with a real day/night pattern (and a lot of SaaS traffic has one), this is where the savings hide.

Bandwidth is bundled too: each worker pool accrues free outbound transfer based on its size, so you're not paying metered egress on every byte that leaves the cluster.

The plan landscape, without the spreadsheet

Since the node cost is really just Droplet cost, the "plans" are really just Droplet families. Here's how they shake out for DOKS use:

• Basic Droplets as worker nodes — starting at $12/month per node. Regular or Premium CPUs, shared tenancy. This is where most people start: dev clusters, small production workloads, anything where you'd otherwise be eyeballing a $5 VPS and wishing it could run k8s. Best for getting off the ground, validating an idea, or running a low-traffic service where you don't need guaranteed CPU. You can 👉 start a Basic-node DOKS cluster here and have a real cluster running for the price of a couple of coffees.
  
  

• General Purpose Droplets as worker nodes — dedicated CPU and RAM, NVMe SSDs, and on the Premium variant, up to 10 Gbps outbound. This is the workhorse tier: production web apps, databases-in-containers, anything where noisy neighbors would ruin your week. Pricing scales with the size, and the Premium General Purpose line is what DigitalOcean has been pushing hard lately.
  
  

• CPU-Optimized and Memory-Optimized Droplets as worker nodes — for the obvious cases. Big data pipelines, batch processing, in-memory caches, ML inference that's CPU-bound. You pay more per node, and you get a machine that doesn't fold under the load.
  
  

• GPU worker nodes — yes, you can put GPUs in your DOKS cluster. Standard GPU Droplets are billed per second like everything else. If you want H100s reserved or contract pricing on a serious AI workload, you go through the sales form rather than the self-serve panel, which is reasonable given that nobody wants to accidentally spin up a five-figure GPU bill on a Friday afternoon.
  
  

Across all of these, the control plane stays free (or $40/month with HA), the surge upgrades during cluster updates are free, and the 28-day cap applies. It's the same service — you're just picking the metal your pods land on.

High Availability: when to flip the $40 switch

The free single-replica control plane is fine for a lot of things, but it's a single point of failure by definition. If your cluster API goes down, you can't schedule, you can't deploy, you can't autoscale — your workloads keep running if they're already up, but you're frozen on the control side.

The HA control plane runs multiple replicas across availability zones for that flat $40/month. It's prorated hourly, so if you only need HA during a launch week or a traffic spike, you're not committing to a full month. For anything you'd call "production" with a straight face — customer-facing apps, anything with an SLA, anything where 3 a.m. pages are not acceptable — the $40 is one of the cheaper insurance policies in the cloud. You can 👉 enable HA on a DigitalOcean Kubernetes cluster and treat it as a line item rather than a project.

The upgrade story, which is more important than people admit

A managed Kubernetes service is only as good as its upgrade path. Self-managed clusters die slow deaths because nobody wants to do the version bumps, and then you end up three minor versions behind, and then the jump becomes terrifying, and then you're stuck.

DOKS handles this with automatic updates, maintenance windows, and surge upgrades. Surge upgrades spin up replacement nodes before tearing down the old ones, which means your pods get rescheduled onto fresh nodes before the old ones drain — no gap, no downtime dance. And surge upgrades are included at no extra cost; you're not paying a premium to avoid shooting yourself in the foot.

As of the latest changelog, DOKS is tracking Kubernetes v1.35.5, with the Node Readiness Controller and a fresh kernel underneath. They publish the changelog openly, so you can see exactly what's landing before you click update.

The 1K nodes thing

Earlier in 2025 DigitalOcean rolled out what they called the "next evolution" of DOKS, pushing single clusters toward supporting on the order of a thousand nodes. The practical version of this: if you started on DOKS because it was cheap and simple, and your workload grew into something genuinely large, you don't have to migrate to a different provider to keep scaling. The same cluster that ran your MVP can run your Series B infrastructure. That's the theory, anyway, and the customer logos they put up — Nitropack going from 15,000 sites to 120,000 on DOKS, ScraperAPI consolidating 100+ Droplets into a DOKS cluster — suggest it's not just theory.

The new-user credit

One more thing worth knowing before you commit any money: DigitalOcean runs a referral program for new accounts that gives $200 in credit valid for 60 days. There's no coupon code to type in — it's applied automatically when you sign up through a referral link. That credit covers DOKS, Droplets, storage, the whole catalog, so you can run an actual cluster — with HA, with a few General Purpose nodes, with real traffic — for two months without touching your card.

A heads-up: I saw one source mention a change in mid-2026 to a smaller $5/90-day credit for new accounts, but the majority of current sources still describe the $200/60-day offer as active. If you're signing up today, check the offer page directly to confirm what's live — referral terms get adjusted, and the only authority on what you'll actually receive is the signup page itself.

Either way, the credit is real, it's auto-applied, and it's enough to do a serious evaluation rather than a "hello world" pod. You can 👉 claim the DigitalOcean new-user credit and try DOKS without committing to anything.

Who this is actually for

If you're a solo developer or a small team that's been circling Kubernetes but been scared off by hyperscaler pricing and procurement, DOKS is built for you. The free control plane, the per-second billing, the 28-day cap, the flat $40 HA — all of it is structured around the assumption that you'd like to use Kubernetes without needing a FinOps person on staff.

If you're running GPU inference workloads and want them in a cluster rather than loose Droplets, the GPU worker node support matters.

If you're already on DigitalOcean for Droplets or App Platform, putting your containerized workloads into DOKS keeps everything in one bill, one API, one CLI, one identity layer. That's not a trivial convenience.

And if you're a larger shop that's been quietly resenting your current managed k8s bill, the math is at least worth running. A free control plane and per-second worker billing will beat most alternatives on a workload with any kind of variance to it.

You can 👉 get started with DigitalOcean managed Kubernetes here — sign up, apply the new-user credit, and have a cluster with a free control plane running before your coffee gets cold. Whether it stays your cluster after the 60 days is up is a question for your actual workload, but at least the trial answers it honestly.