Over the past six months, three AI startups migrated from Vercel and Cloudflare to AWS Lambda. They hit predictable walls: background jobs, retries, queues, and cron tasks. While Vercel excels at Next.js deployment and Cloudflare at edge latency, both struggle with backend complexity. Lambda's integrated ecosystem—SQS, DynamoDB, and EventBridge—handles these needs under one network, often near $0 for small workloads.

The serverless hype cycle has faded, but the technology matured. Early tools like SAM were brittle, but modern frameworks like SST have simplified deployment. Cold starts remain a constraint, requiring a different architectural mindset. The key is recognizing serverless's sweet spot: request-based apps with variable traffic, especially AI-integrated APIs. It's not for every workload, but for the right use case, it's unbeatable.

Startups outgrow simpler platforms when they need robust backend services. Vercel's queueing is still in beta, and Cloudflare Workers have memory and runtime limits. For predictable high traffic, long-running processes, or GPU workloads, containers or EC2 are better. But for bursty, event-driven applications, Lambda's pay-per-use model and integrated tooling make it the default choice for teams building on AWS.