Apple Weighs Return to Servers With M8 Ultra AI Machines, May Use Nvidia NVLink Fusion

Apple is working on an AI server built on its own chips. If it ships, it would be the company’s first server product since it retired Xserve in January 2011. The Information first reported the project, and The Verge, Ars Technica, Bloomberg and Reuters followed. According to the report, Apple has also talked with Nvidia about using Nvidia’s networking technology in the machines.

Two or four M8 Ultra chips

The enterprise server would come in two configurations, with either two or four of Apple’s future M8 Ultra chips, Ars Technica reported, citing The Information. It isn’t expected before 2029. Apple only announced the M6 family in August, so M8 is still a ways off, The Verge noted. The project reportedly had backing from John Ternus when it started about a year ago, while he still ran hardware engineering. Ternus took over as Apple CEO from Tim Cook on Sept. 1.

To link the chips together, Apple is considering Nvidia’s NVLink Fusion, an interconnect that lets multiple chips work as one. The two companies have had a frosty relationship for years, but Apple recently turned to Nvidia chips for its revamped Siri servers. The Information cautioned that the project could still be canceled or go ahead without Nvidia’s technology.

Built on demand Apple already sees

The idea follows strong sales of the Mac mini and Mac Studio to AI developers, which have caused shortages. The Information reported that AI companies like OpenAI have bought “tens of thousands” of Mac minis and Mac Studios to train AI agents with reinforcement learning, and that Anthropic has rented Mac minis from Amazon Web Services. A purpose-built server would let Apple sell into that demand directly, on the strength of the power efficiency of its Arm-based M-series chips.

Long road ahead

A 2029 launch gives Nvidia, AMD and the cloud providers’ custom chips years to move further ahead. Apple would also have to build enterprise sales and support it hasn’t had in more than a decade. And it faces the same memory-chip shortage hitting the rest of the industry as companies race to build AI data centers, Ars noted.