Infrastructure Capacity Planning Template

Quarterly for stable services, monthly for high-growth services. Trigger an off-cycle review after any capacity-related incident, before a major product launch, or when growth rate changes by more than 20%. The [Technical PM Handbook](/technical-pm-guide) covers how to integrate capacity reviews into your planning cadence.

Start scaling preparations at 70% utilization. Initiate scaling actions at 80%. Never let production workloads exceed 85% sustained utilization. These thresholds give you enough headroom to scale before users are affected. Some resources (databases, queues) need more headroom than stateless compute because they take longer to scale.

Use analogues from similar products, industry benchmarks, and conservative estimates. Build three scenarios: pessimistic (50% of target), expected, and optimistic (200% of target). Plan infrastructure for the expected case with the ability to scale to the optimistic case within your lead time window. Track actual versus projected weekly and adjust.

Auto-scale stateless compute (web servers, workers) where scaling is fast and low-risk. Manually scale stateful systems (databases, queues, caches) where incorrect scaling can cause data issues. Hybrid approaches work well: auto-scale within a pre-approved range, alert humans when auto-scaling hits the upper boundary.

Calculate the cost of downtime per minute (lost revenue, SLA penalties, customer churn risk) and compare it to the cost of the infrastructure investment. If 10 minutes of downtime costs $50,000 and the scaling investment is $3,000/month, the payback period is measured in days, not months. Use the [TAM calculator](/tools/tam-calculator) to size the revenue at risk. ---