Apple is reportedly building AI servers again, for the first time since 2011 -- and talking to Nvidia about the chip-to-chip link that would run them

Apple is developing rack-mounted AI inference servers built around future M8 Ultra chips, according to a report from The Information, marking a potential return to server hardware for the first time since Apple retired the Xserve in 2011. Two configurations are reportedly in the works: a smaller unit clustering two M8 Ultra processors, and a larger one with four. The machines would target AI inference -- running already-trained models to generate responses -- rather than the more compute-intensive work of training frontier models from scratch, and would be sold to AI developers, businesses and governments that want to run models on hardware they own instead of the public cloud.

The more striking detail is who Apple is reportedly talking to about connecting those chips: Nvidia, via its NVLink Fusion interconnect. The two companies have kept each other at arm's length for nearly two decades, after GPU reliability problems in early-2010s MacBooks soured the relationship and pushed Apple toward building its own silicon in-house ever since -- Nvidia chips have not shipped in a Mac since. A confirmed hardware partnership here would be a real thaw, not just a chip roadmap update. The project reportedly began about a year ago, with organizational backing from John Ternus, then Apple's hardware engineering lead and now, since Sept. 1, its chief executive. The Information reports that demand from AI teams already buying Mac mini and Mac Studio machines in volume helped sharpen Apple's interest in a rack-scale product -- but cautions the plan could still be cancelled, or ship without Nvidia's technology at all. Neither company has commented, and the report has not been independently verified.