PaaS, VPS or Kubernetes: what each one asks of a small team.
For a team of one to ten developers running a few web apps, a PaaS usually costs the least once you count hours as well as the invoice. A VPS is cheapest on paper but makes you the sysadmin. Kubernetes pays off when you run many services, need to survive losing a zone, or already have someone who runs clusters.
Facts checked
The work each option leaves with you
The server bill is the small number. The large one is the recurring work: patches, certificates, backups, deploys and being the person who gets the alert. Here is who does each job, with Canner standing in for the PaaS column because those are the facts we can vouch for.
| Job | PaaS (Canner) | VPS | Managed Kubernetes |
|---|---|---|---|
| Operating system patches † | Canner runs and patches the servers. You update your own dependencies. | Yours: security updates, reboots, and a new release when yours ages out. Ubuntu LTS gets 5 years of standard security maintenance; 22.04 ends in May 2027. | The provider patches the control plane. Node images are yours to roll. |
| TLS certificates † | Automatic Let’s Encrypt certificates for your-project.canner.app and for custom domains. | Yours: Caddy or certbot, plus watching renewals. Let’s Encrypt certificates shrink from 90 days to 64 in February 2027 and to 45 in February 2028. | Yours: an ingress controller and cert-manager, kept upgraded. |
| Database backups | Scheduled, encrypted backups on every plan; point-in-time recovery with the $9/mo dedicated instance. | Yours: a pg_dump cron, a copy off the box, and restore tests. | Usually a managed database outside the cluster, or a Postgres operator you run. |
| Deploys and rollback | Git push, canner deploy or MCP. The new version must answer on / within 30 seconds before it takes traffic; a failed GitHub deploy rolls back to the last good commit. | Yours: a script or a tool such as Coolify or Dokku that pulls, builds, restarts and copes with a failed start. | CI builds an image, pushes it to a registry and applies manifests or Helm charts. Rollouts and rollbacks are built in once that pipeline exists. |
| On-call | Canner watches the servers. Your app’s own bugs are still yours. | Everything on the box is yours, including a full disk at 3 a.m. | Nodes, ingress, DNS add-ons and storage drivers are yours; the control plane is the provider’s. |
| Version upgrades † † | None to schedule. | None beyond the OS. | Kubernetes keeps only the three newest minor versions, each with about a year of patches. On EKS, a cluster past standard support costs US$0.60 an hour instead of US$0.10. |
† Competitor details come from each vendor’s own published pages on the dates shown; prices and plans change, so check the vendor before you decide. Operating system patches (2026-10-29) · TLS certificates (2026-10-29) · Version upgrades (2026-10-29) · Version upgrades (2026-10-29)
Cost of ownership for one app and its database
The bills below are the vendors’ listed prices. The hours are our assumptions, not measurements: change them to match your team. They are meant to show the shape of the comparison, which is that the hours decide it.
- Workload: one web app with a small Postgres database and modest traffic, in production, with backups.
- Hours a month: 1 on a PaaS (dependency updates, reading the occasional failed build), 4 on a VPS (patches, renewals, backup checks, one incident every few months), 10 on Kubernetes (node and add-on upgrades, manifests, the pipeline, incidents).
- An hour of a developer’s time: $100. If your time is free because you enjoy running servers, drop that column and the VPS wins.
- Currency: each bill is in the currency the vendor lists. Canner bills in Canadian dollars; DigitalOcean and AWS list US dollars.
| Option | Monthly bill as listed | Hours a month (assumed) | Time at $100/h (assumed) | Rough monthly total |
|---|---|---|---|---|
| Canner Live | $9 CAD: 2 always-on apps, Postgres with scheduled backups. +$9 for point-in-time recovery. | 1 | $100 | About $110 to $120 |
| DigitalOcean Droplet, Toronto † † | US$12 for 2 GiB / 1 vCPU, plus US$3.60 for daily backups (30%). Postgres runs on the same box. | 4 | $400 | About $415 |
| DigitalOcean Kubernetes, Toronto † † † | Control plane free; two US$12 nodes = US$24. High-availability control plane +US$40. | 10 | $1,000 | About $1,025 to $1,065 |
| Amazon EKS, Montréal † † | US$0.10 per cluster-hour, about US$73 a month before nodes, load balancers and storage. | 10 | $1,000 | Over $1,075, plus nodes |
† Competitor details come from each vendor’s own published pages on the dates shown; prices and plans change, so check the vendor before you decide. DigitalOcean Droplet, Toronto (2026-10-29) · DigitalOcean Droplet, Toronto (2026-10-29) · DigitalOcean Kubernetes, Toronto (2026-10-29) · DigitalOcean Kubernetes, Toronto (2026-10-29) · DigitalOcean Kubernetes, Toronto (2026-10-29) · Amazon EKS, Montréal (2026-10-29) · Amazon EKS, Montréal (2026-10-29)
Managed Kubernetes with a Canadian location
If you do need Kubernetes and want it in Canada, these are the managed options we could confirm from the vendors’ own pages. A Canadian region does not make a US provider Canadian: the US CLOUD Act reaches data a US company controls wherever it is stored. Our sovereignty pages go through it.
| Service | Canadian location | Owner | Control plane price as listed |
|---|---|---|---|
| DigitalOcean Kubernetes (DOKS) † † | Toronto (TOR1) | DigitalOcean (US) | Free; high-availability control plane US$40/mo |
| Amazon EKS † † | Montréal (ca-central-1), Calgary (ca-west-1) | Amazon (US) | US$0.10 per cluster-hour; US$0.60 on extended support |
| OVHcloud Managed Kubernetes † † | Beauharnois, Quebec (BHS5 only) | OVHcloud (France) | Free plan; Standard plan $0.15 per cluster-hour (currency not stated) |
| Canner | Montréal | Canner (Canada) | Not Kubernetes. Live is $9/mo flat |
† Competitor details come from each vendor’s own published pages on the dates shown; prices and plans change, so check the vendor before you decide. DigitalOcean Kubernetes (DOKS) (2026-10-29) · DigitalOcean Kubernetes (DOKS) (2026-10-29) · Amazon EKS (2026-10-29) · Amazon EKS (2026-10-29) · OVHcloud Managed Kubernetes (2026-10-29) · OVHcloud Managed Kubernetes (2026-10-29)
When Kubernetes is the right call
If none of these apply, Kubernetes mostly adds the work in the last row of the first table. Start simpler and move when one of them becomes true; a twelve-factor app with its config in environment variables moves to a container later without much rewriting.
- You run many services that deploy on their own schedules, and the cost of the platform is spread across all of them.
- You must survive losing a zone or a server. Node pools across zones and several replicas per service give you that; a single-region PaaS without failover does not.
- Your workloads don’t fit a PaaS: custom images, sidecars, GPU node pools, operators, or a fleet of workers next to the web tier.
- Someone on the team already runs clusters and wants to. Kubernetes without that person is a part-time job nobody signed up for.
When a VPS is the right call
A VPS makes sense when cash is tighter than time, the app is a single process with steady traffic, and you like owning the box. Tools such as Coolify or Dokku take care of the deploy step. They do not patch the OS, test your restores or answer the pager. If you want Coolify without running the server, MapleDeploy sells a managed instance in Toronto.
Where Canner fits, and where it doesn’t
Canner suits a small team with a handful of web apps, Postgres next to each app, data that should stay in Quebec, and a bill in Canadian dollars. Starter is free and Live is $9 a month; the pricing page has the add-ons. It is not the right choice if any of these matter to you:
- One region. Everything runs in Montréal. There is no second region to fail over to.
- No high availability or automatic failover. If the server has a problem, your app is down until it is fixed or restored. Enterprise adds a 99.9% uptime SLA with service credits, which is a refund, not a standby copy.
- One process per project. There is no separate worker process yet; background jobs run inside the web process (background jobs). There is no Redis.
- No Dockerfile builds. Canner detects the framework and builds from source. .NET is not supported.
Questions
Does a small startup need Kubernetes?
Rarely. With one to five services and no requirement to survive a zone outage, a PaaS or a well-run VPS covers it with far less recurring work. Kubernetes starts paying for itself when many services share the platform or failover across zones is a hard requirement.
Is a VPS cheaper than a PaaS?
On the invoice, usually yes: a 2 GiB DigitalOcean Droplet in Toronto is US$12 a month. Once you count patching, certificate renewals, backup checks and incidents, the VPS costs more for most teams unless that time is free to you.
Is there managed Kubernetes in Canada?
Yes. DigitalOcean Kubernetes runs in Toronto, Amazon EKS in Montréal and Calgary, and OVHcloud Managed Kubernetes in Beauharnois, Quebec. The first two are US-owned and OVHcloud is French.
Is Canner a Kubernetes platform?
No. Canner is a PaaS: it builds your app from source and runs it on its own servers in Montréal, with Postgres alongside. It has no high availability or failover, so it is not a substitute for a multi-zone cluster.
Try the PaaS column for free.
Starter costs nothing and needs no card. Deploy one app and see how much of the first table goes away.