Apple subsidiary tests Harbor to enable multi-mac AI compute clusters

Apple's subsidiary, Claris, is developing a capability that allows Macs to function as local AI servers through its application, Harbor. This software enables Apple chip-equipped Macs to run large language models directly on the device, bypassing the need to transmit sensitive data to external cloud services. Harbor supports running smaller models on a single Mac, but it can also link multiple machines into a cluster using Thunderbolt 5 connections. This clustering functionality leverages Thunderbolt 5 RDMA support, which was recently integrated into macOS Tahoe 26.2, allowing for faster data transfer between grouped devices. The application offers one-click local inference by automatically detecting Thunderbolt 5 hardware and providing access to open-weight models. While Harbor is currently in an early-access phase and its App Store listing has been removed, this development signals Apple's strategic move to build a modular, on-device compute pool. This approach suggests Apple is positioning its own hardware as a viable alternative to third-party cloud AI platforms, offering users an energy-efficient, private computing solution for their idle Macs.