SaaS vs PaaS vs IaaS, with examples you'll recognize
The short version
All three are ways of renting computing instead of owning it. The difference is how much of the stack you rent ready-made.
- SaaS (software as a service) rents you a finished application. You log in and use it. Examples: web-based email, a CRM (customer relationship management system), online accounting.
- PaaS (platform as a service) rents you a ready-made place to run software your own developers write. You bring the code; the provider handles servers, operating systems and scaling. Examples: Heroku, Google App Engine, Azure App Service, AWS Elastic Beanstalk.
- IaaS (infrastructure as a service) rents you raw building blocks: virtual servers, storage and networks. You install and manage everything on top. Examples: Amazon EC2, Azure Virtual Machines, Google Compute Engine.
The further down the list, the more control you get and the more work you take on.
These three terms come from the US National Institute of Standards and Technology (NIST), whose definition of cloud computing, SP 800-145, describes each model by what the customer does and doesn’t manage. That’s still the clearest way to tell them apart.
Side by side
| SaaS | PaaS | IaaS | |
|---|---|---|---|
| What you rent | A working application | A managed runtime for your own code | Virtual machines, storage, networking |
| What you manage | Your data, users and settings | Your code and data | Operating system, software, data, security settings |
| Who it’s for | Everyone in the business | Developers | IT teams, developers, managed service providers |
| How it’s priced | Per user, per month or year | By app size, running time or resources | By resource: compute hours, storage, data transfer |
| Time to value | Minutes to days | Days to weeks | Weeks |
| Control | Least | Medium | Most |
| Everyday examples | Microsoft 365, Google Workspace, Salesforce, Slack | Heroku, Google App Engine, Azure App Service, AWS Elastic Beanstalk | Amazon EC2, Azure Virtual Machines, Google Compute Engine, DigitalOcean Droplets |
The analogy: getting to work
Think about how you get to the office.
- IaaS is leasing a car. You get the vehicle, but you choose the route, fill the tank, book the servicing and drive it yourself. Lots of freedom, lots of responsibility.
- PaaS is a taxi. You say where you want to go (your code) and the driver handles the car, the fuel and the route. You can’t choose the engine, but you don’t have to maintain it.
- SaaS is the bus. It runs a fixed route on a fixed timetable. You just get on. It’s the cheapest and easiest option per trip, as long as the route goes where you need.
Most companies use all three at once, just as most people walk, take the bus and occasionally hire a car.
Who manages which layer
Layer Own servers IaaS PaaS SaaS ------------------ ----------- ------- ------- ------- Data & users you you you you Application you you you vendor Runtime / database you you vendor vendor Operating system you you vendor vendor Virtual machines you vendor vendor vendor Physical servers you vendor vendor vendor Building & power you vendor vendor vendorCloud providers call this split the shared responsibility model. Amazon’s shared responsibility page and Microsoft’s equivalent for Azure both make the same point: the provider looks after the cloud itself, and what you put in it stays your job. The top line never moves. Whatever you rent, your data and who can reach it remain your responsibility.
Examples in more detail
SaaS examples
Your team’s email and documents (Microsoft 365 or Google Workspace), your CRM, your help desk, your payroll tool, your project management board. Nobody in your company installs or patches these. You configure them, add users and connect them to each other. Our explainer on what SaaS means in business covers the buying checklist, and an ERP system is one of the larger things you might buy this way.
PaaS examples
Heroku describes itself as a PaaS for deploying and scaling apps; you can see how it charges on its pricing page. Google App Engine, Azure App Service and AWS Elastic Beanstalk play similar roles inside the big clouds. A typical small-business use: your agency builds a customer portal and deploys it on a PaaS, so nobody on your side has to manage a server. Managed databases also sit in this layer.
IaaS examples
Amazon EC2, Azure Virtual Machines and Google Compute Engine rent you virtual servers by the hour or second. A typical use: moving an old Windows application that only runs on a server out of your office closet and into a data center you don’t have to look after. You still patch the operating system and back it up, unless an IT partner does that for you. Our AWS vs Azure vs Google Cloud comparison covers how the three big IaaS providers differ.
One company, all three models
Let’s follow a made-up 30-person furniture wholesaler to see how the models sit side by side.
SaaS, for nearly everything staff touch. Email and documents, the CRM the sales team uses, the help desk, payroll and the online accounting system are all subscriptions. The office manager adds and removes users; nobody installs anything. The monthly bill is predictable: number of users times price per user.
PaaS, for one custom tool. The wholesaler’s biggest customers wanted a portal to check stock and reorder. An agency built it and deployed it on a PaaS. The agency pushes updates; the platform handles servers and scaling when a big customer runs a large order. The bill rises a little in busy months.
IaaS, for one stubborn old system. The warehouse still runs an older stock-control program that needs its own Windows server. Rather than keep a machine in a closet, the company’s IT partner moved it to a virtual machine on a big cloud provider. The partner patches it and checks the backups each month, and the company pays both the cloud bill and the partner’s fee.
Three models, three different owners of the work, and three different kinds of bill. The useful habit is knowing which is which, so that when something breaks, everyone knows whose phone should ring.
The gray areas
Real products blur the lines.
- Serverless functions (small pieces of code that run only when triggered) are often called FaaS (function as a service), sitting between PaaS and SaaS.
- Integration platforms are sold as SaaS but let you build workflows, so they feel a bit like PaaS. We cover them under middleware.
- Low-code platforms let non-developers build apps on a managed platform, which is PaaS with a friendlier front end.
- Managed IaaS, where an IT firm runs your virtual servers for you, feels like SaaS from your side of the desk even though it’s IaaS underneath.
Don’t worry about the label. Ask the practical question: which layers will my team have to look after?
Decision factors
- Is there an off-the-shelf app that does the job? If yes, SaaS is almost always the fastest and cheapest route.
- Do you have developers, or an agency, building something custom? PaaS lets them focus on the app instead of servers.
- Do you need to run software that expects its own server? Older line-of-business applications often do. That’s IaaS territory.
- Who will be on call? With IaaS, someone has to patch, monitor and back up. Name that person or firm before you start.
- How predictable must the bill be? SaaS per-user pricing is easy to forecast. IaaS usage pricing moves with traffic and needs watching.
What this means for a 20-person company
- Most of your software should be SaaS. That’s normal, not a lack of ambition.
- If an agency builds you a custom app, ask whether it will run on PaaS or IaaS, and who maintains it after launch.
- Use IaaS mainly to retire on-site servers, ideally with an IT partner managing it.
- Whatever the model, keep a list of where your data lives and who has admin access.
Also consider
- Managed service providers (MSPs): IT firms that run IaaS or on-site systems for you on a monthly contract.
- Private cloud and colocation: renting space for your own hardware in someone else’s data center.
- Staying on-premise for specific workloads: sometimes the right answer for latency, regulation or old software.