Depot vs Self-Hosted vs Cloud CI
Three options for CI compute in 2026: managed acceleration (Depot.dev), self-hosted runners, or cloud hosted runners (GitHub Actions, CircleCI). Honest per-minute pricing, real break-even, and when each option actually wins.
Head-to-head dimensions
| Dimension | Depot.dev | Self-hosted runners | Cloud hosted |
|---|---|---|---|
| Per-minute rate (Linux baseline) | $0.004 (2-vCPU runner) | ~$0.001-0.002 (spot, high util) | $0.006 (no self-hosted fee) |
| Cache persistence | Yes (EBS volume) | Yes (manual setup, runner-local) | Ephemeral / GitHub-managed (limited) |
| Operational burden | None (managed) | High (runner controller, security, scaling) | None (managed) |
| Setup time | <1 hour (drop-in) | 1-2 weeks (real production setup) | Default (already there) |
| Multi-architecture builds | Native (ARM + AMD) | Manual (separate runner pools) | Native on hosted ARM runners |
| Pricing transparency | $0.004/min runners; $0.04/min container builds | Server cost + admin time (variable) | Published per-minute by OS |
Four scenarios, four answers
Startup, 5 engineers, 4K min/month, simple Linux pipeline
GitHub's free tier (2,000 min on Free, 3,000 on Team) covers most of it; the small overage is a few dollars a month. At this volume Depot's lower per-minute rate saves pennies and isn't worth the third-party dependency. Nothing to optimise.
Scaleup, 30 engineers, 25K min/month, Docker-heavy
GitHub hosted = 25K × $0.006 = $150/mo. Depot runners at $0.004/min are ~$100/mo at face value, and the persistent cache typically cuts Docker-heavy build time 30-60%, so real spend lands nearer $60-90/mo for faster feedback. For the image-build steps specifically, Depot's Container Builds ($0.04/min on a 16-vCPU machine) compress multi-minute docker build jobs further. Depot wins here on both cost and speed.
Enterprise, 100 engineers, 200K min/month, mature platform team
GitHub hosted = 200K × $0.006 = $1,200/mo. Depot runners = 200K × $0.004 = $800/mo with zero ops, less once speed gains cut minutes. Self-hosted on spot with high utilisation can reach ~$0.001-0.002/min (~$300-600/mo of compute) but adds ~0.5 FTE platform engineer ($75K/yr loaded). Self-hosted wins on raw compute only if you already run that platform team; otherwise Depot's managed $800/mo is the cheaper all-in. GitHub hosted is the clear loser at this scale.
macOS-heavy, 15 engineers, 8K macOS min/month
GitHub hosted macOS at $0.062/min × 8K = $496/mo. Depot now does offer macOS runners ($0.08/min for an 8-CPU M-series machine, roughly half of GitHub's large macOS class) but they cost more per minute than GitHub's standard macOS, so they win on speed, not price. Self-hosted Mac mini M2 cluster: ~$599 capex × 3 + ~$150/mo colo is roughly a 4-5 month payback on the hardware, then ~$350/mo saved. The real cost of self-hosting here is iOS provisioning and Apple-silicon admin, not the hardware.
The honest read on Depot
Depot.dev is a real product solving a real problem (cache persistence and faster builds), and their calculator tool is honest. The frame to get right is which product you are pricing. Depot's drop-in GitHub Actions runners are $0.004/min for 2-vCPU Linux, below GitHub's own $0.006/min, so on a like-for-like runner swap Depot is cheaper, not a premium. The $0.04/min number that floats around is Depot's separate Container Builds service (a 16-vCPU remote docker build machine), which is a different line item aimed at Docker-heavy pipelines.
The realistic decision: if you spend $300+/month on CI, run a 2-week proof-of-concept on Depot with your actual build pipeline. Because the runners are both cheaper per-minute and faster, the bill usually falls; the bigger your Docker layer surface area, the more the persistent cache (and the Container Builds add-on) pays back on top.
The one case where GitHub hosted still wins is the free tier: for small pipelines under ~2,000-3,000 minutes/month (pytest suites, ESLint, Go test, lightweight unit tests), GitHub hosted is free and a third-party runner isn't worth the dependency. Once you are paying for minutes, Depot's runners are hard to beat on cost.
Frequently asked questions
Is Depot.dev cheaper than GitHub Actions hosted runners?
Usually yes, on both price and speed. Depot's GitHub Actions runners cost $0.004/min for the 2-vCPU Linux class, which is below GitHub's own hosted Linux ($0.006/min after the January 2026 cuts), and Depot bills by the second with no round-up. Its runners are also faster thanks to a persistent layer cache and faster networking, so you spend fewer minutes too. The $0.04/min figure often quoted for Depot is its separate Container Builds product (a 16-vCPU remote Docker-build machine), not its drop-in runner. Where GitHub still wins is the free tier: below ~2,000-3,000 minutes/month it is free and Depot is not worth the switch. Above that, Depot runners typically cost less per month than GitHub hosted for the same workload.
When does self-hosted CI beat both Depot and cloud hosted?
Self-hosted wins on pure per-minute cost once volume amortises a flat server bill. A $30/mo VPS handling roughly 7,500+ minutes/month already beats Depot's $0.004/min and GitHub's $0.006/min; beyond 50,000 min/month on bigger or spot instances the gap widens further (no platform fee applies: the charge GitHub announced for March 2026 was postponed and never took effect). The catch is operational cost: a 0.5 FTE platform engineer at $150K loaded salary is real money that managed runners like Depot and GitHub hosted fold into the per-minute rate. Self-hosted is the cost-winner at scale, but only if you genuinely have the platform team to maintain it (security patches, capacity planning, runner controller updates, secret rotation).
What does Depot.dev actually do that hosted runners don't?
Depot provides three things GitHub's own hosted runners don't: (1) a persistent layer cache that survives across builds (GitHub's hosted caches are size-limited and slower to restore), (2) faster networking and NVMe-backed disks, so the same job finishes in fewer minutes, and (3) a separate Container Builds service ($0.04/min on a 16-vCPU machine) for remote docker build with native multi-architecture support. The drop-in runners themselves are $0.004/min for 2-vCPU Linux, below GitHub's $0.006/min, so Depot can be both cheaper and faster on the same pipeline.
What's the realistic break-even between Depot and cloud hosted for typical builds?
For runner-vs-runner there is barely a break-even: Depot's 2-vCPU Linux runner ($0.004/min) already undercuts GitHub hosted ($0.006/min) before you count its speed advantage, so Depot is usually cheaper per month from the first paid minute. The genuine break-even is against GitHub's free tier, below ~2,000-3,000 minutes/month, where GitHub hosted is free and Depot is not worth it. Depot's Container Builds product ($0.04/min on a 16-vCPU machine) is a different calculation, weighed against building Docker images on your own runners, and pays back when image builds are large and cache-heavy.
Is Depot a real competitor to self-hosted at scale?
It is the middle ground, and once you use Depot's real runner rate ($0.004/min, not the $0.04 container-build rate) it is more competitive than it first looks. Depot runners already beat GitHub hosted; against self-hosted the gap is narrower. At scale (50K+ min/month), self-hosted on spot instances with good cache discipline can reach ~$0.001-0.002/min and undercut Depot on raw compute, but only after you absorb the platform-team overhead. Below that scale, Depot's zero-ops model is usually cheaper all-in once you price the engineering time self-hosting demands. The procurement question is whether the per-minute savings of self-hosting actually exceed the loaded cost of the team that maintains it.