Skip to content

IaaS vs PaaS vs SaaS

5-minute read.

The one-line answer

Three flavors of cloud service, sorted by how much the provider does for you:

  • IaaS - rent the building blocks (servers, disks, networks). You set everything up.
  • PaaS - rent the platform. You bring code; the provider runs it.
  • SaaS - rent the finished software. You just log in and use it.

The pizza analogy

The classic visualization:

You eat... Make from scratch Take and bake Delivery Dine in
Dough You Provider Provider Provider
Sauce You Provider Provider Provider
Toppings You Provider Provider Provider
Oven You You Provider Provider
Plates You You You Provider
Table You You You Provider
On-prem IaaS PaaS SaaS 
flowchart TB
  subgraph ONP[On-prem]
    O1[App + data]:::you
    O2[Runtime]:::you
    O3[OS]:::you
    O4[Hypervisor]:::you
    O5[Hardware]:::you
    O6[Datacenter]:::you
  end
  subgraph IAAS[IaaS]
    I1[App + data]:::you
    I2[Runtime]:::you
    I3[OS]:::you
    I4[Hypervisor]:::them
    I5[Hardware]:::them
    I6[Datacenter]:::them
  end
  subgraph PAAS[PaaS]
    P1[App + data]:::you
    P2[Runtime]:::them
    P3[OS]:::them
    P4[Hypervisor]:::them
    P5[Hardware]:::them
    P6[Datacenter]:::them
  end
  subgraph SAAS[SaaS]
    S1[Your data + config]:::you
    S2[App]:::them
    S3[Runtime]:::them
    S4[OS]:::them
    S5[Hypervisor]:::them
    S6[Hardware]:::them
  end
  classDef you fill:#ffd,stroke:#aa0
  classDef them fill:#dde,stroke:#33a

Yellow = your responsibility. Blue = provider's. The further right you go, the less you manage - but the less you control.

IaaS - Infrastructure as a Service

You get: virtual machines, virtual disks, virtual networks. Raw building blocks.

You manage: OS patches, the runtime, your application, scaling, monitoring.

Provider manages: the physical hardware, hypervisor, datacenter, networking fabric.

Examples: - AWS EC2, EBS, VPC - Azure VMs, Managed Disks, Virtual Networks - GCP Compute Engine, Persistent Disks, VPC

Use when: - You want maximum control - You're lifting-and-shifting an existing app - You have specific OS / kernel / driver requirements - You're running something the PaaS layer doesn't support

PaaS - Platform as a Service

You get: a platform that runs your code. You hand it a container, a Java app, a Node.js project, etc., and it figures out the rest.

You manage: your code and configuration.

Provider manages: OS patching, runtime upgrades, scaling, load balancing, sometimes the database.

Examples: - AWS App Runner, Elastic Beanstalk, Lambda (also serverless) - Azure App Service, Container Apps - GCP App Engine, Cloud Run - Heroku (the OG PaaS) - Vercel, Netlify, Render, Railway

Use when: - You want to ship fast, not manage servers - Your app fits a standard runtime (web app, API, function) - You're a small team without ops bandwidth

SaaS - Software as a Service

You get: finished software, accessed through a browser or API.

You manage: your own data and users inside it.

Provider manages: literally everything else.

Examples: - Gmail, Slack, Notion, Salesforce, GitHub, Figma - DataDog (monitoring), Auth0 (authentication), Stripe (payments)

Use when: - The problem is already solved by software someone built - You'd rather pay $X/month than build and maintain it - (i.e. 99% of business software needs)

Where serverless fits

"Serverless" is fuzzy - it's marketing-speak that overlaps PaaS. The defining traits:

  • You don't think about servers (the platform handles compute capacity)
  • You pay per request or per millisecond, not per hour
  • It scales to zero when idle (no idle cost)

Serverless examples: AWS Lambda, Azure Functions, GCP Cloud Run, Cloudflare Workers.

It's basically "PaaS with very fine-grained billing." Some people call it FaaS (Function as a Service) when it's specifically functions.

A real-world stack

A typical modern startup:

Layer Choice Type
User-facing app Vercel hosting Next.js PaaS
Backend API AWS Lambda Serverless / PaaS
Database Supabase or Neon Postgres SaaS / DBaaS
Auth Clerk or Auth0 SaaS
Email Postmark or Resend SaaS
Payments Stripe SaaS
Analytics PostHog or Mixpanel SaaS
Errors Sentry SaaS
ML OpenAI or Anthropic API SaaS

Notice: barely any IaaS. A modern small team can build serious products without ever provisioning a VM.

When does IaaS still come up? - Specialized workloads (GPU training, custom networking, regulated environments) - Very high scale where managed services are too expensive - Self-hosting because you want control of data / cost / latency

What to look at next