On this page 10 sections
Container as a service (CaaS) is a cloud model where you hand the provider a container image, or the source to build one, and it runs the container for you. The provider owns the servers, the operating system and the scheduling. You choose how much CPU and memory each container gets, how many copies run, and which ports and environment variables it uses.
AWS Fargate, Google Cloud Run and Azure Container Apps are CaaS, and so are developer platforms such as Railway, Render, Fly.io and Ownkube.
This post explains where CaaS sits next to IaaS, PaaS, FaaS and Kubernetes, how the main providers bill, and how to pick one.
TL;DR
- CaaS runs your containers on infrastructure the provider manages. The unit you deploy is a container image.
- It sits between IaaS (you manage virtual machines) and FaaS (you deploy single functions). Modern PaaS products run containers under the hood, so the line between PaaS and CaaS is mostly about how much the platform builds and wires up for you.
- Kubernetes is the engine many CaaS products run on. Using CaaS means someone else operates that engine.
- Most CaaS bills per second or per minute for the CPU and memory a container reserves or uses. Load balancers, data transfer, logs and databases are separate lines.
- Pick raw CaaS (Fargate, Cloud Run) when the rest of your stack lives in that cloud. Pick a developer platform when you want builds, URLs, databases and deploys handled together.
What does container as a service mean?
A container packages an app with everything it needs to run: the runtime, libraries and config. The same image runs on a laptop, in CI and in production. CaaS is the hosting model built around that unit.
With CaaS, you provide:
- A container image, or a repository with a Dockerfile the platform builds.
- The CPU and memory for each container.
- The port it listens on, its environment variables and its secrets.
- How many copies run, or the rules for scaling them.
The provider handles:
- The servers and their operating system, including security patches.
- Placing containers on servers and restarting them when they crash.
- Networking, usually a load balancer or a public URL with TLS.
- Logs and metrics, at least at a basic level.
How is CaaS different from IaaS, PaaS and FaaS?
| Model | You deploy | You manage | Examples |
|---|---|---|---|
| IaaS | Virtual machines | The OS, runtime, scaling and the app | Amazon EC2, Google Compute Engine, DigitalOcean Droplets |
| CaaS | Container images | The image and its resources | AWS Fargate, Google Cloud Run, Azure Container Apps |
| PaaS | Source code | The code and its config | Heroku, Render, Railway |
| FaaS | Single functions | The function code | AWS Lambda, Cloudflare Workers, Vercel Functions |
CaaS vs IaaS
With IaaS you rent a virtual machine and install everything on it. You can run containers there with Docker, but you patch the OS, restart crashed containers and replace a dead machine yourself. CaaS takes all of that off your hands, at a higher price per unit of CPU and memory than the raw machine.
CaaS vs PaaS
The two have converged. Classic PaaS took source code and picked the runtime for you. Most PaaS products now build a container from your code (or your Dockerfile) and run it, which makes them CaaS with a build step and extras such as databases and preview environments. Today the practical difference is how much the platform does around the container.
CaaS vs FaaS
FaaS runs one function per request, starts it on demand and stops it after. It suits short, bursty work. Long-running processes, WebSocket servers, background workers and apps that keep state in memory fit containers better. Function platforms also limit run time, memory and package size, while a container runs as long as you need.
Is Kubernetes container as a service?
Kubernetes is a container orchestrator: the software that decides where containers run, restarts them and wires up networking. Managed Kubernetes (Amazon EKS, Google GKE, Azure AKS) runs the control plane for you, but you still manage nodes, upgrades, networking add-ons and every manifest.
CaaS hides that layer. Many CaaS products run on Kubernetes, and you never see it. If your team has no one whose job is the cluster, CaaS is the closer fit.
What are the main CaaS providers?
Cloud provider CaaS
- AWS Fargate runs containers for Amazon ECS and EKS without servers to manage. You define a task (CPU, memory, image) and pay per second for what it reserves, with a one-minute minimum.
- Google Cloud Run runs a container per request or as an always-on service, scales to zero when idle, and bills for the time containers handle requests or stay allocated.
- Azure Container Apps runs containers on a managed environment with scale to zero and event-driven scaling. Azure Container Instances runs single containers or groups for simpler jobs.
These fit well when the rest of your system already lives in the same cloud. You still set up the networking around them: VPCs, load balancers, IAM roles, registries and a database.
Developer platforms
- Railway, Render, Fly.io, DigitalOcean App Platform and Northflank build and run containers from a repository, give each service a URL, and offer managed databases next to it.
- Ownkube is a developer platform that builds a container from your GitHub or GitLab repository, runs it next to a managed Postgres and Valkey cache, and bills per minute from a prepaid wallet.
These trade some cloud-level control for much less setup. You connect a repo and get a running service with a URL.
How is CaaS priced?
Most CaaS bills for CPU and memory over time. The differences are in what counts:
| Billing model | How it works | Who uses it |
|---|---|---|
| Reserved resources | You pay for the CPU and memory you allocate, per second or minute, busy or idle | AWS Fargate |
| Request time | You pay while a container handles requests, plus a per-request fee | Google Cloud Run (request-based) |
| Measured usage | You pay for the CPU and memory the container actually uses | Railway, Ownkube metered |
| Flat instance | You pay a fixed monthly price for an instance size | Render, Ownkube reserved boxes |
On AWS Fargate in US East (N. Virginia), Linux on x86 is $0.04048 per vCPU-hour and $0.00445 per GB-hour. The smallest task, 0.25 vCPU and 0.5 GB, costs about $9 a month running all the time. A 1 vCPU, 2 GB task is about $36 a month. A load balancer, data transfer, logs and a database come on top. See AWS Fargate pricing explained for worked bills.
For an app that sits idle much of the day, measured usage or scale to zero changes the bill the most. A reserved container bills for every hour it exists.
When should you use CaaS?
CaaS is a good fit when:
- Your app already runs in a container, or can with a short Dockerfile.
- It is a long-running web service, API, worker or bot.
- You want to deploy several times a day without touching servers.
- Nobody on the team wants to own a Kubernetes cluster.
It fits less well when:
- You need GPUs, special kernels or host-level access that the provider does not offer.
- The workload is a handful of short functions that run rarely. FaaS is simpler there.
- You run steady, heavy compute around the clock and have the skills to run your own servers. Raw virtual machines cost less per unit.
How to choose a CaaS provider
Ask these questions in order:
- Where does the rest of your stack live? If your data and other services are in one cloud, its own CaaS avoids cross-cloud traffic and a second set of permissions.
- Do you want to build the image yourself? Some providers take only a prebuilt image from a registry. Others build from your repository on each push.
- What else does the app need? A database, a cache, cron jobs and a worker are separate products on cloud CaaS and part of the platform on most developer platforms.
- How does it bill when idle? Check whether the provider bills reserved resources, measured usage or a flat size, and whether a service can scale to zero.
- Which regions does it run in? Latency and data residency both depend on it.
How Ownkube runs containers
Ownkube is a developer platform built on containers. You connect a GitHub or GitLab repository, and Ownkube builds the image on each push and pushes it to a registry it manages for you. If the repo has a Dockerfile, it builds from that. If not, the automatic build detects the language and builds the image for you.
Each service runs as a web app, a background worker or a cron job. Web apps get an auto-allocated *.ownkube.app hostname with TLS, health checks and automatic sizing. A web app with a public hostname can sleep after an idle window you set and wake on the next request.
Billing is per minute from a prepaid wallet. Metered services bill CPU and memory on actual usage, at about $18 per vCPU and $6 per GiB a month. A service can opt into a flat reserved box instead, from $5 a month. Managed Postgres runs on a database box from $4 a month next to the app, and a Valkey cache on a cache box from $2.
Pulling a prebuilt image from Docker Hub or your own registry is not available yet, so the image is built from a repository. Ownkube runs in the Germany and Mumbai regions. The deploy a Docker image guide walks through a Dockerfile deploy step by step.
FAQ
What is an example of container as a service?
AWS Fargate, Google Cloud Run and Azure Container Apps are CaaS from the big clouds. Developer platforms such as Railway, Render, Fly.io and Ownkube also run your app as containers and add builds, URLs and databases around them.
Is Docker a container as a service?
No. Docker is the tool that builds and runs containers on a machine you manage. A CaaS provider runs Docker-compatible containers on machines it manages for you.
Is CaaS the same as PaaS?
They overlap. CaaS takes a container image and runs it. PaaS takes source code, builds it and runs it, and today most PaaS products do that by building a container. A developer platform is usually both.
Is AWS ECS container as a service?
ECS is AWS’s container orchestrator. With the Fargate launch type, ECS is CaaS: AWS runs the servers and you define tasks. With the EC2 launch type, you manage the instances the containers run on, which is closer to IaaS.
Is container as a service cheaper than a virtual machine?
Per unit of CPU and memory, usually not. A virtual machine running all the time costs less per vCPU. CaaS saves money when the app is idle much of the time, when it scales up and down, or when the time spent patching and running servers is counted.
Where Ownkube fits
If you want the container model without a cloud account, a cluster or a load balancer to set up, Ownkube builds your repo into a container, runs it next to managed Postgres and a cache, and bills per minute from a prepaid wallet you top up. Deploy my first app.