Estimate compute cost of running a CI/CD pipeline based on runtime and agent pricing.
Total CI/CD pipeline cost is rarely just compute — it's the sum of agent compute hours (the dominant driver for most fleets), storage for artifacts and workspaces, and network egress for pulling images, dependencies, and transferring artifacts. Breaking the total into these three components makes clear where cost-optimization effort pays off most; compute usually dominates, so reducing build time or agent idle time typically yields more savings than storage or network tuning alone.
totalCost = (agentHours × costPerHour) + storageCost + networkCost
Agent-hours scale with build frequency and duration across potentially hundreds of daily builds, while storage and network costs tend to grow much more slowly and are easier to cap with retention policies, so compute is both the largest and most variable cost component.
Use spot/preemptible instances for non-critical builds, right-size agent instance types to actual CPU/memory needs, enable agent auto-scaling to avoid paying for idle agents, and reduce build duration through caching and parallelism.
Yes — for self-hosted (non-cloud-billed) agents, 'cost per agent hour' should be derived from amortized hardware/depreciation and power costs divided by available agent-hours, rather than an on-demand cloud rate, since idle self-hosted capacity is a sunk cost either way.