Skip to content
Back to course

18.2 Scalability

Scalability is an AI system's ability to handle growing traffic and data without breaking or becoming too expensive. A demo that serves 10 users may collapse at 10,000 if you do not plan for load, latency, and cost.

TrainPackageDeployServeMLOps lifecycle
Scaling a model service horizontally behind a load balancer.

For instance, a mobile money app in Ethiopia may be quiet at night but spike on payday. Auto-scaling adds servers during the spike and removes them later, so the app stays fast without paying for idle machines all month.

Scenario

Your LLM API costs spike during predictable daily peaks. What scales most cost-effectively?

Check your understanding

1/4 · 40 XP

What does scalability primarily address?