Built for a scale that never came

The second most expensive mistake I see is infrastructure built for a scale that never came

A startup spends three months building a multi-region, highly available, auto-scaling architecture. Kafka for event streaming. Separate read replicas for every database. Cross-region failover. The works

They have 200 users

Not 200,000. 200

The infrastructure can handle 50,000 concurrent users without breaking a sweat. 200 users use about 4 percent of its capacity. The other 96 percent runs continuously, full price, waiting for traffic that may or may not ever arrive

This is a real thing I've seen. Not once

The motivation is always good. Engineers want to build things right. They've read about the companies that got caught unprepared when they went viral. They don't want to be the team that had to rewrite everything under load. They build for the future to avoid the pain of not being ready

The problem is that the future is uncertain and the infrastructure bill is not

Every month you run infrastructure at 4 percent utilization is a month of runway spent on capacity you don't have the users to justify yet

The right question is not "can this architecture handle our growth?" but "what's the cheapest architecture that handles our current load and can be upgraded when we actually need to upgrade it?"

The answer at 200 users is almost always simpler than what teams build

One region. One availability zone, maybe two. A single appropriately sized database. An application server that can be vertical scaled or replicated when needed. Basic monitoring. That's it

You build the complex architecture when the growth makes it necessary, not when the ambition makes it appealing

The companies that write those viral "how we scaled to X users" blog posts almost never built that architecture at 200 users

They built it when they needed it. Because they still had runway to get there