What's the actual difference between IaaS, PaaS, and SaaS, and how would you decide which one is the right fit for a given workload?

The three sit on a spectrum of how much of the stack the provider manages versus you: IaaS (e.g. raw EC2 VMs) gives you compute/storage/networking primitives and you manage the OS up; PaaS (e.g. Elastic Beanstalk, Heroku, App Engine) manages the OS and runtime, leaving you to manage just your application code and its configuration; SaaS (e.g. Salesforce, Datadog) is a complete application you consume, managing none of the underlying stack at all. The right choice depends on how much control versus operational burden you're willing to trade — custom infrastructure needs or unusual runtime requirements push toward IaaS, standard web-app patterns push toward PaaS, and a solved business problem you don't need to build yourself points to SaaS.

It helps to think of the trade-off as a single sliding scale of "who manages what":

• IaaS gives you the lowest-level building blocks — virtual machines, block storage, virtual networks — and you're responsible for everything above that: OS patching, runtime installation, scaling logic, deployment tooling. Maximum control and flexibility, maximum operational responsibility. This is the right call when you have non-standard infrastructure requirements (custom kernel modules, specific compliance-driven OS hardening, unusual networking topology) or you're building the platform other things run on top of (e.g. you're the team building the Kubernetes cluster itself).
• PaaS takes the OS and runtime off your plate — you deploy application code and configuration, and the platform handles provisioning, scaling, and often load balancing and zero-downtime deploys for you. The trade-off is reduced control: you're constrained to what the platform supports (specific runtimes, specific scaling knobs), and debugging sometimes requires working around the abstraction rather than through it. This is the right call for standard application workloads where the operational overhead of managing infrastructure yourself isn't buying you anything — most CRUD web apps and APIs fit here.
• SaaS is a finished application, and "your job" reduces to configuration and integration, not building or operating anything. The trade-off is the least control of all three — you're bound by the vendor's feature set, data model, and roadmap. This is the right call whenever the problem being solved isn't your differentiator: nobody should be building their own CRM or their own APM platform from IaaS primitives when mature SaaS options already solve it well.

The decision in practice usually isn't "pick one for the whole company" — most real organizations run a mix: IaaS or Kubernetes for their core differentiated services, PaaS for internal tools and quickly-shipped features where operational overhead isn't worth it, and SaaS for well-solved horizontal problems (observability, CRM, identity). The interview-relevant skill isn't reciting the definitions, it's being able to justify why a specific workload belongs at a specific point on that spectrum given its actual requirements.

• Where would you place "serverless" (e.g. AWS Lambda) on this spectrum, and does it fit cleanly into IaaS/PaaS/SaaS at all?
• How would you evaluate whether to move an existing IaaS-hosted application to a PaaS, given the migration cost?
• What operational risks does heavy SaaS reliance introduce that IaaS/PaaS don't (e.g. vendor lock-in, data portability)?

• IaaS, PaaS, and SaaS differ in how much of the stack the provider manages for you, not in what "cloud" means generically.
• More managed (PaaS/SaaS) means less operational burden but less control and more constraint to the platform's supported patterns.
• The right model is a per-workload decision based on how non-standard the requirements are and whether the problem is a differentiator worth owning.
• Most real organizations use a deliberate mix across all three rather than standardizing on just one.

• NIST: The NIST Definition of Cloud Computing (SP 800-145)
• AWS: Types of cloud computing
• Google Cloud: IaaS vs PaaS vs SaaS