engineering
The economics of running CI infrastructure
Where the money actually goes when you operate a runner fleet: idle capacity, egress, and the cost of a three-second start.
Jess Okonkwo · Co-founder · 2026-01-29 · 9 min read
CI has an awkward load shape. Demand is bursty, correlated with working hours, and extremely latency-sensitive at exactly the moments it spikes.
Warm pools are the whole business
A three-second start requires machines already booted and waiting. Someone pays for that idle time. Getting pool sizing right — predicting demand accurately enough to hold few enough idle machines — is most of the engineering value a managed provider adds.
Egress is larger than people expect
Image pulls, package downloads, cache writes and artifact uploads all cross a network boundary. On BYOC deployments NAT gateway charges are frequently the single largest line item, and they are almost never in the initial business case.
Why self-hosting looks cheaper than it is
The compute bill is real and visible. Scaling policy, image pipelines, cache infrastructure and the on-call rotation are real and invisible. We think the honest comparison includes engineer-days, and most comparisons do not.
Your next build could be twice as fast, at half the price
Start free. Migrating away is the same one line, and we publish that diff too.