Scaling in the Modern Era
The debate isn't just A vs B anymore. It's about 'Persistent Compute' (Containers/Kubernetes) vs 'Ephemeral Compute' (Lambda/Cloud Functions). Both have their place, but choosing the wrong one can bankrupt your startup.
Serverless: The Cost of Cold Starts
Serverless is amazing for 'bursty' traffic. If you run a ticket sales site that is dead 90% of the time but explodes for 10 minutes when tickets go on sale, Serverless is perfect. You pay $0 when idle. However, the 'Cold Start' (the 500ms-2s delay when a function wakes up) is a UX killer for real-time apps.
Microservices (Containers): Predictable Performance
If you have consistent traffic, Containers on Kubernetes are often cheaper than Serverless at scale. You pay for the reserved CPU, but you get sub-millisecond response times because the application is already loaded in memory.
The Hybrid Approach
The winning strategy in 2026 is often hybrid. Use a containerized Modular Monolith for your core API (User auth, billing, products). Use Serverless functions for async background tasks (Image resizing, PDF generation, sending emails). This gives you the speed of persistent compute with the infinite scaling of serverless for heavy lifting.
Conclusion
Don't pick an architecture based on Hacker News hype. Analyze your traffic patterns. If it's spiky, go Serverless. If it's flat/heavy, go Containers.