AWS Lambda and Fly.io represent two fundamentally different deployment paradigms: serverless (function-as-a-service) vs. containerized microservices (always-on compute). The choice is architectural, not just a preference for one tool over another.

AWS Lambda Model

Lambda functions scale to zero. You pay per million invocations (~$0.20 per million). Cold starts can be 100ms-10s. Execution time is limited to 15 minutes max. You write functions, AWS manages everything else.

Fly.io Model

Fly.io VMs are always-on (or scale manually). You pay per VM-hour. No cold starts. No execution time limits. You have a real server, just a lightweight one managed by Fly.io.

When Lambda Makes Sense

Traffic is bursty (flash sales, viral events) — scale to zero means you pay nothing during quiet periods
Event-driven workloads (S3 triggers, DynamoDB streams, scheduled events)
Infrequent invocations where pay-per-request is genuinely cheaper

When Fly.io Makes Sense

Consistent baseline traffic (chatbots, APIs with regular users)
Long-running processes (background jobs, data processing)
Stateful connections (WebSockets, SSE, persistent bot connections)

For OpenClaw (AI Chatbots)

Fly.io is the right model. AI chatbots have persistent connections to Telegram/Discord, maintain state across requests, and can't tolerate cold starts. Lambda's execution model fundamentally doesn't fit the chatbot pattern.

Pricing Reality

Lambda's per-request pricing sounds cheap but adds up with high-traffic chatbots. A chatbot receiving 10K messages/day runs $5-20/month on Lambda. Fly.io equivalent: $5-15/month. The price difference is smaller than it appears.

AWS Lambda vs Fly.io: Serverless vs Persistent Containers