I've seen this take a few times now and it's making me second-guess our setup. We're a small SaaS team and our build volume is all over the place—some weeks we push tons of tiny fixes, other weeks it's quiet.
Everyone recommends the big cloud CI platforms, but the pricing models seem to scale with every parallel build and minute used. For those of you who moved off cloud CI/CD because of cost spikes, what did you switch to? Are self-hosted runners on something like GitHub Actions a realistic fix, or is there a better middle ground? I need real numbers if you have them—our last bill scared me.
Your experience is exactly why we see so many teams hit a wall with that pricing model. Self-hosted runners on GitHub Actions can absolutely be a realistic fix, but they introduce operational overhead you have to be ready for.
We documented a similar transition last year where a team's costs dropped by about 70% after moving critical workflows to a few dedicated machines. The catch is they spent non-trivial engineering time on maintenance, security patches, and scaling logic. For a small team, that trade-off might be worth it, but you need to factor in that hidden cost. Have you looked at your build patterns to see if a hybrid approach, using cloud for peak times and self-hosted for baseline, could smooth out the spikes?
—HR
You're right to be spooked by that pricing model. We made the switch to self-hosted runners on GitHub Actions after our Jenkins cloud bill tripled in a month due to a dependency update that triggered mass rebuilds.
The real numbers: our monthly CI costs went from an average of $1,2k on a major cloud platform to about $200 for the compute and a rough estimate of 15 engineer-hours per quarter for maintenance. The initial setup was about 40 hours. The hybrid approach user1013 mentioned is smart, but in practice we found it added complexity we didn't want. We just run everything on a small, auto-scaling group of spot instances now.
The key is having your infrastructure defined as code. If you treat your runner setup like cattle, not pets, the operational overhead is manageable. Did you analyze which specific job types or workflows are driving your cost spikes? That data should guide whether you move everything or just the heavy jobs.
connected