Apple Reportedly Plans M8 Ultra AI Server for 2029

Apple is reportedly designing an M8 Ultra-powered AI server for release in 2029, as the company explores returning to dedicated server hardware. The Information states that the device would feature either two or four M8 Ultra chips and may utilize Nvidia networking. This project began approximately a year ago with support from former hardware engineering chief, John Ternus, who joined Apple as CEO on September 1.

If the server launches, it would mark Apple's return to enterprise hardware since it discontinued Xserve in 2011. Apple's current M5 Ultra chip supports up to 512GB of unified memory and 1.2TB/s memory bandwidth. However, it remains unclear whether multiple M8 Ultra chips could share a single memory pool or if memory would remain isolated to each chip.

Nvidia technology may be used to connect the chips, with NVLink Fusion being a potential option. The M8 Ultra server would move Apple away from using Macs for AI workloads and towards a hardware solution designed specifically for large-scale inference. Although Apple relies on Nvidia GPUs for some AI compute, it could use its own processors in this new server.

The current M5 Ultra chip demonstrates both the benefits and constraints of Apple's approach, as its 512GB memory capacity offers more local model storage than many conventional GPUs, but its memory bandwidth is significantly lower than that of Nvidia's H200. This information suggests that Apple may be aiming to expand its AI infrastructure stack by incorporating Apple silicon into a more significant role.