Apple Reportedly Building Its Own AI Supercomputer for 2029

Apple is reportedly building its own AI server — a rack of powerful computers that runs AI for many users at once — with a target date of 2029. According to The Information, the machine would use two or four of Apple's future M8 Ultra chips, a processor family that won't exist for years. Apple has not confirmed the project, set a price, or said how it would sell it.

Why now? AI companies love Macs. OpenAI has bought tens of thousands of Mac minis and Mac Studios to build AI agents (software that takes actions on its own), and Anthropic reportedly rents Mac minis through Amazon. That demand has caused shortages of some models. The server project reportedly started about a year ago with backing from John Ternus, who became Apple's CEO on September 1. It would also mark a return to selling servers, a business Apple quit in 2011 when it killed the Xserve.

The technical challenge is memory. Apple's current M5 Ultra chip can address 512GB of memory and move data at 1.2TB/s — plenty for running large AI models locally, but far below the 4.8TB/s Nvidia lists for its H200 chip. Apple is reportedly considering Nvidia's NVLink Fusion technology to connect chips together. Apple already leans on Nvidia: its AFM 3 Cloud Pro model runs on Nvidia GPUs inside Google Cloud.

For you, the practical takeaway is modest but real. A 2029 product is far off, could be canceled, and changes nothing about your iPhone today. But if Apple builds its own AI hardware, Siri and other features could get faster and cheaper to run — and Apple could keep more of your data inside its own private cloud instead of renting someone else's.