GitHub names its own data centers as a cause of the August 6 Actions outage

GitHub's July 2026 availability report concedes that the GitHub Actions service at the centre of the August 6 outage is still running entirely in GitHub's own data centers, and that this was a contributing factor to the capacity shortfall during the incident ^[4]. For anyone running CI on a single provider, that turns a bad Thursday into a documented structural dependency: the vendor has now named the component, the constraint, and the fix it had deprioritised.

The specific admission is worth reading slowly. GitHub says the majority of Actions already runs on Azure, but that it had not prioritised migrating the launch service, the component that bridges the monolith to Actions, because the work is generally asynchronous and can queue in response to problems ^[5]. That reasoning is defensible right up to the point where it is not: GitHub says cascading failures produced an unacceptable delay in recovery ^[6]. Queueing is a buffer, not a capacity plan, and a buffer in a rack you cannot grow is a single point of failure with extra steps. GitHub is now accelerating the move of Actions to Azure, where it says it will have more headroom and more ability to absorb spikes ^[7], and accelerating its broader Actions architectural roadmap around isolation, resiliency and scale ^[3].

The company describes the incident as unacceptable in both impact and duration ^[1], and the root cause analysis is not finished; the full detail is promised in the August report, due in September ^[2]. So the accounting is partial by GitHub's own admission.

What is measurable is the migration underneath. Monolith read traffic served from Azure Central US peaked at 52.75% on July 28, which GitHub calls the first time it stayed consistently above half ^[8] - meaning 47.25% of that read traffic was still served elsewhere at the high-water mark ^[1]. Git traffic in Azure reached 47%, up from 43% in June ^[9], a four-point monthly gain ^[3]. Twenty-nine per cent of repositories now have a second replica in Central US to soften regional failover ^[10], which leaves 71% without one ^[2]. GitHub also reports the first authentication tables moving off its oldest shared database to dedicated infrastructure ^[12], a dedicated user service offloading more than a million queries per second at peak with 80% of a major authorisation lookup on the isolated path ^[13], repository content traffic running fully from Central US on dedicated hardware and a dedicated pull request service at 99.87% parity with the monolith for authenticated reads ^[14].

The smaller efficiency wins are the honest kind: one production change halved total query time on the artifacts table, and caching in the Git authorisation path cut authorisation service load by 18.2% while request volume grew ^[15]. All production search workloads now serve from Central US with extra headroom ^[16], and the stability validation process introduced after May's incident is applied to every major traffic expansion ^[11], following the deliberate pause GitHub described a month earlier ^[17].

Read together, this is a company mid-migration telling you which parts are not migrated yet. Actions was one of them.