Self-hosted CI runners: when they cost less than hosted

Quick answer

Self-hosted CI runners run on your own instances instead of the CI provider's, trading per-minute hosted pricing for instance cost plus the operational effort of running them. They cost less when build volume is high enough that the instance cost beats the equivalent hosted minutes, especially with autoscaling (scale runners to demand, to zero when idle) and Spot instances for fault-tolerant builds. For low or bursty volume, hosted runners' simplicity and zero idle cost usually win.

CI providers offer hosted runners (billed per minute) and the option to self-host runners on your own infrastructure. Self-hosting trades the per-minute charge for instance cost plus operational responsibility, so whether it saves money depends on build volume and how well you manage the runner infrastructure.

The tradeoff

	Hosted runners	Self-hosted runners
Cost	Per minute (zero when idle)	Instance cost + ops
Idle cost	None	Instances if left running
Best for	Low/bursty volume, simplicity	High volume, with autoscaling + Spot

Hosted runners cost per minute and nothing when idle, so they suit low or bursty volume, you pay only for builds. Self-hosted runners cost the instances they run on plus operational effort, cheaper per unit of compute at high volume but only if you avoid paying for idle runners. So the key to self-hosted savings is not running idle instances.

Autoscaling and Spot make self-hosted cheap

Self-hosted runners pay off most when they autoscale to build demand, spinning up on queued jobs and scaling to zero when idle so you do not pay for idle capacity, and when they run on Spot instances (CI builds are fault-tolerant, retry on interruption). Autoscaling Spot-backed runners can serve high build volume at a fraction of hosted per-minute cost. Static always-on self-hosted runners, by contrast, often waste money on idle time.

Choosing for cost

Use hosted runners for low, bursty, or simplicity-first CI, where zero idle cost and no operations matter more than per-minute savings. Move to self-hosted for high, steady build volume where autoscaling Spot-backed runners beat hosted per-minute pricing, and you have the capacity to operate them. The savings are real at scale but only materialize with autoscaling and Spot, otherwise idle self-hosted runners can cost more than hosted, part of overall CI cost strategy.

FAQ

When are self-hosted CI runners cheaper than hosted?

When build volume is high enough that the instance cost beats the equivalent hosted per-minute charges, especially with autoscaling (scale runners to demand and to zero when idle) and Spot instances for fault-tolerant builds. For low or bursty volume, hosted runners' simplicity and zero idle cost usually win, since self-hosted instances left idle can cost more.

How do self-hosted runners save money?

By replacing per-minute hosted pricing with instance cost, which at high build volume is cheaper per unit of compute, especially when runners autoscale to build demand (scaling to zero when idle) and run on Spot instances at a discount. The savings only materialize with autoscaling and Spot; static always-on self-hosted runners often waste money on idle time.

What is the downside of self-hosted runners?

Operational responsibility: you provision, secure, patch, and scale the runner infrastructure yourself, unlike hosted runners the provider manages. And if runners run idle (not autoscaled to zero), you pay for that idle capacity, potentially more than hosted. So self-hosting trades per-minute cost for instance cost plus operational effort, worth it mainly at high volume with good automation.

Can CI runners use Spot instances?

Yes, and they should for cost. CI builds are typically fault-tolerant, they can retry on interruption, so running self-hosted runners on Spot instances captures a large discount with minimal downside (an interrupted build reruns). Autoscaling Spot-backed self-hosted runners is the cheapest way to serve high build volume, well below hosted per-minute pricing.

Should low-volume projects self-host runners?

Usually not. For low or bursty build volume, hosted runners cost per minute and nothing when idle, so they are simple and cheap for occasional builds. Self-hosted runners' instance cost and operational effort only pay off at high, steady volume with autoscaling. For low volume, the simplicity and zero idle cost of hosted runners wins.

Does C3X estimate self-hosted runner cost?

C3X prices the runner instance infrastructure from your Terraform, so the compute cost of self-hosted runners appears before deploy, making it easy to compare against hosted per-minute pricing. Actual cost depends on build volume, autoscaling, and Spot usage, which you model as usage assumptions.

What to do next

Price your self-hosted runner fleet before you build it. C3X reads your Terraform and prices your resources against a live catalog. Start with the quickstart.