About Serverless vs Containers vs VMs

Where each wins; when serverless bites. Part of the DevTools Surf developer suite. Browse more tools in the Info / Guides collection.

Use Cases

• Architects choosing compute platforms for new microservice deployments
• Startup CTOs evaluating serverless to minimize infrastructure management
• DevOps teams deciding between ECS/Fargate and Lambda for event-driven workloads
• Engineering managers estimating infrastructure costs at different traffic scales

Tips

• Read the 'where each wins' matrix to match workload to platform
• Check the cold start analysis for latency-sensitive applications
• Review the cost comparison to understand when serverless becomes expensive

Fun Facts

• AWS Lambda, launched in November 2014, was the first major serverless compute platform. It originally supported only Node.js with a 60-second timeout.
• A typical AWS Lambda cold start adds 100-500ms of latency; provisioned concurrency (launched 2019) eliminates this but costs 3-4x more.
• VMs provide the strongest isolation (hardware-level), containers share the kernel (OS-level), and serverless functions share the runtime — each trades isolation for efficiency.

FAQ

When does serverless win?

Bursty traffic (pay nothing at idle), event-driven workloads (queue processing), and teams with no ops capacity. Cold starts hurt latency-sensitive endpoints.

When do containers win?

Steady traffic, long-running processes (WebSockets, background workers), precise performance tuning, and portability across clouds.

VMs — still relevant?

Legacy workloads, specialized hardware (GPUs for ML), or when you need full OS control. For new apps, containers or serverless are almost always better.

Can I mix them?

Yes — most mature stacks do. API behind containers (predictable latency), async workers on serverless (bursty). Pick per workload characteristics.