App Center 2026 migration window: hosted CI vs per-day cloud Mac runner for short-cycle mobile builds

Visual Studio App Center retires in stages: core build, test, and distribution capabilities ended after March 31, 2025, while Analytics and Diagnostics gained an extension into 2026 per Microsoft Learn — hence the “2026 migration window.” Teams often replace CI months before crash telemetry moves. Here we focus on short-cycle mobile pipelines where queue time hits revenue, and where private dependencies plus signing injection determine whether hosted CI or a rented Apple Silicon runner wins.

What the retirement timeline forces you to reorder

App Center used to bundle builds, device testing hooks, and distribution metadata. After sunset those pieces land in separate systems — pipelines in Azure DevOps or GitHub Actions, binaries in store consoles, telemetry in Azure Monitor or an ISV APM. The sequencing trap is migrating analytics first while nightly iOS jobs still assumed App Center signing helpers; teams then learn provisioning profiles, API keys, and private CocoaPods feeds never made it into vault-backed CI variables.

Hosted CI vs per-day cloud Mac self-hosted runner

Hosted CI wins when orchestration matters more than Apple Silicon wall time — Gradle/npm caches on Linux, Android matrix jobs, branch protections. Per-day cloud Mac runners win when macOS queue tax repeats, an Xcode minor stays pinned for weeks, or flaky archives need SSH. Both paths still demand templates, images, and on-call ownership.

Cost modeling is not only cents per minute: stuck archives that pull three engineers off feature work erase hosted-CI savings. Idle always-on Macs waste budget — align rental length with sprint cadence. Android-heavy teams often keep Linux elastic and cordon iOS onto one or two Apple hosts; monorepos sharing codegen may colocate both stacks on a larger Mac when artifact symmetry beats marginal VM density.

Private dependencies and signing injection: decision matrix

Private Maven, npm, NuGet, or CocoaPods endpoints usually fail for one of three reasons: DNS split horizons, IP allow lists that exclude hosted runners, or credential rotation that outpaces pipeline updates. Fix the policy layer before debating hardware — otherwise every runner flavor stalls on the same 403. For Apple code signing, prefer injecting an App Store Connect API key via ephemeral environment variables, importing certificates into a throwaway keychain per job, and pairing that with Fastlane Match or an internal PKCS#12 vault. Avoid copying developer IDs onto shared laptops; treat signing materials like database superuser creds.

React Native teams weighing hosted queues versus local builds should read Expo EAS queues, credential injection, and cache keys on per-day cloud Mac runners. Native iOS shops should align signing with Fastlane Match, encrypted Git, and HTTPS read-only layouts before cutting nightly archives.

FAQ — crisp answers for steering committees

Do we have to pick Azure Pipelines because Microsoft suggested it?

No — GitHub Actions, GitLab CI, CircleCI, Buildkite, and Jenkins all remain viable if macOS labels and artifact storage meet compliance. The migration cost is reproducing secrets, caches, and notarization steps, not the badge on the orchestrator.

Is a per-day cloud Mac runner “self-hosted” if we rent it?

Operationally yes: you accept responsibility for image hygiene, monitoring, and patching cadence even when the hardware is billed hourly. The vendor supplies metal; your team still owns launch agents or CI worker tokens.

Where should private dependency mirrors live?

Co-locate feeds with SAS or OIDC-scoped reads; if SaaS runners cannot reach your registry, add a pull-through cache on a trusted network instead of widening firewall exceptions.

What signing telemetry belongs in logs?

Profile UUID, cert thumbprint, xcodebuild exit codes, and notarization ticket IDs per artifact hash cover most audit questions App Center used to answer implicitly.