Serverless (Lambda, Cloudflare Workers, Vercel Functions) vs. dedicated VMs (Fly.io, DigitalOcean, EC2) is one of the first architectural decisions for a new project. The tradeoff is real and the right choice depends on workload characteristics.

The Core Tradeoff

Serverless: You trade control for convenience. The platform handles scaling, availability, and resource allocation. You pay per request/use.

Dedicated VMs: You have full control and predictable pricing. You handle scaling (or use a platform that helps). Pay per time.

Serverless Advantages

Scale to zero: Pay nothing when no traffic
No ops burden: Platform handles availability and scaling
Global edge: Most serverless platforms have better geographic distribution
Cost at low traffic: Can be dramatically cheaper for infrequent workloads

Dedicated VM Advantages

Predictable pricing: Pay per month, not per million invocations
No cold starts: Real servers, always ready
Full compatibility: Any runtime, any library, any binary
Long-running processes: No execution time limits
Stateful connections: WebSockets, SSE, persistent TCP work naturally

When to Choose Serverless

Highly variable traffic (including zero)
Event-driven workloads (webhooks, scheduled jobs, IoT)
Simple, short-running functions
Teams without ops capacity who want maximum convenience

When to Choose Dedicated VMs

Consistent baseline traffic
Stateful or long-running processes
Specific runtime requirements (Python with C extensions, etc.)
Need for full system control
Predictable budgeting

For OpenClaw AI Backends

Dedicated VMs (Fly.io) are the right model. AI chatbots need persistent connections, can't tolerate cold starts, and often run at consistent traffic levels. The serverless model is a poor fit.

Serverless vs Dedicated VMs: The Architecture Decision