The "AWS vs GCP" question gets asked at almost every kickoff. The honest answer is that for a startup with fewer than thirty engineers, the differences that matter are not the ones the marketing pages talk about. Both clouds will run your workload. The question is which one your team will be productive on.
Compute and orchestration
ECS Fargate and Cloud Run sit at the same point on the simplicity curve — you ship a container, the platform runs it, you pay per second. Cloud Run is slightly faster to learn and has friendlier defaults; ECS gives you more control once you outgrow them. Either is a fine choice. If you know you will end up on Kubernetes, EKS is more mature than GKE Autopilot is in some regions, but Autopilot is genuinely the lowest-effort managed Kubernetes on the market.
Data services
This is where the choice matters. RDS for Postgres on AWS is the safe pick for almost any OLTP workload, and Aurora gives you headroom when you outgrow vanilla Postgres without forcing a migration. On GCP, Cloud SQL is solid but Spanner is a more interesting long-term bet if you expect global writes — at a meaningful cost premium and a steeper learning curve.
For analytics, BigQuery is in a class of its own. If your product has a real analytics surface and a small data team, that single service can be the reason you pick GCP.
Identity and access
IAM on AWS is more granular, which is both its strength and its tax. GCP's IAM model is simpler and easier to reason about, but you will eventually want to layer on Organization Policies and Service Account boundaries to match what AWS gives you out of the box.
The thing that actually decides it
The cloud your team already knows. The hiring market in your city. The discount your investor has negotiated. These three signals will out-predict any technical comparison most of the time.
If none of them apply, pick AWS for the broader hiring pool and the deeper ecosystem of third-party tooling, or pick GCP if BigQuery or Cloud Run are load-bearing for your product. Don't agonise.