Cloud Mac compute power,
delivered near you

AI · Inference & batch tasks

We run inference services and overnight batch processing on a Mac mini (M4) in a cloud data center. Model files and the runtime environment stay fixed on that cloud machine; my own laptop is only used for editing code, reading logs, and writing quick scripts. Before, when running long tasks locally, filling up memory would slow down local dev tools and destabilize video calls. Since moving the heavy work to the cloud machine, my laptop basically no longer fights inference for the same memory.

We still run regression tests with the same sample set before each release — no shortcuts there. Our rough cost split: tasks that run for a long time and peg the machine go to the cloud; phases that involve frequent code changes and parameter tuning stay local, to avoid shuttling multi-GB model files back and forth.

Gao Yuan

Algorithm Engineering · Inference Platform

Builds · CI & certificates

For iOS releases, we assign a dedicated cloud Mac to each Git repo, using Fastlane for signing and creating archives. Each cloud machine holds only that repo's certificate, provisioning profile, and config files — avoiding the situation on a shared build machine where "someone else's private key ends up in the system keychain and nobody knows who signed what".

Multiple release pipelines can run in parallel without competing for the same build machine. Certificates and private keys still live in the team's central vault and are provisioned to the relevant cloud machine at build time. The cloud Mac solves the "who uses which environment" isolation problem; auditing, rotation, and access controls are still your own team's responsibility.

Ding Mo

Mobile Infrastructure · CI Lead

Cross-border · Collaboration & releases

With design in Singapore and dev/ops in Hangzhou, for UI reviews and walkthroughs we have everyone connect on a large screen to the same cloud Mac in the data center closest to the design team, viewing the same native macOS display via VNC. This way everyone takes a shorter network path, which is usually smoother than remotely controlling a colleague's personal machine in Singapore from Hangzhou — though latency still depends on the connection and time of day, and can still lag during peak hours.

Release timing no longer requires the whole team to be on call in the same timezone. Network quality, data compliance, and vendor ticket commitments are still things you need to verify before ordering. We've documented a "city/team ↔ preferred data center node" reference chart internally so new hires don't have to keep asking which region to connect to.

Su Qing

Global Product · Releases & Operations