AWS CDK vs Terraform for SaaS: A Pragmatic Comparison

Every SaaS that outlives its prototype stage has the same early fork: how do you define and manage infrastructure? The two serious options for most teams are AWS CDK and Terraform. We've shipped with both. Here's how we think about the choice today.

CDK in one sentence

AWS CDK lets you describe infrastructure in TypeScript (or Python, Go, etc.) and compiles it to CloudFormation. You get real types, real abstractions, real unit testing — and you're locked into AWS and CloudFormation's quirks.

Terraform in one sentence

Terraform uses HCL, its own domain-specific language, and talks directly to provider APIs. It's multi-cloud, has a massive provider ecosystem, and the state model is explicit and portable.

What actually decides it for SaaS

• If you're a one-cloud shop (typically AWS): CDK wins. The ergonomics of writing infrastructure in the same language as your app, with real type checking and testable constructs, compounds over time.
• If you're multi-cloud or plan to be: Terraform. HCL isn't fun, but the provider coverage is unmatched and the state model travels.
• If your team already knows one deeply: stick with it. The tool is almost never the bottleneck — the discipline is.

Where CDK bites you

CloudFormation state is opaque compared to Terraform state. Stack drift, ROLLBACK_COMPLETE, and REVIEW_IN_PROGRESS states require pre-deploy cleanup scripts in CI to stay sane. We learned the hard way and now treat those cleanup scripts as part of the delivery.

Where Terraform bites you

State files become a shared resource you have to guard, and the module ecosystem rewards discipline punishingly: one sloppy module and you're rewriting half your infrastructure six months in.

Our default for new SaaS

AWS CDK v2 with TypeScript. It lines up with the app language, constructs are composable, and with a disciplined pipeline it's fast. If a client requires Terraform or multi-cloud, we flex — but the default tends to be CDK.

We bake this opinion into our SaaS Platform so new projects inherit the known-good pipeline on day one.