Enterprise servers are a market Apple walked away from in 2011. A report from The Information says the company is exploring a way back, with an AI machine built around two or four future M8 Ultra processors and possibly Nvidia’s NVLink Fusion for the chip-to-chip links.
The hardware would be sold to outside customers rather than kept in-house, and is not expected before 2029. John Ternus, who moved from hardware engineering into the CEO seat on September 1, has backed the effort since it took shape roughly a year ago.
AI teams have quietly become a Mac customer class. OpenAI’s reinforcement-learning work alone accounts for tens of thousands of Mac mini and Mac Studio purchases. Anthropic takes a different route, renting Apple’s small desktops from a cloud catalog.
Apple’s own interconnect may not scale cheaply enough for a rack, which is why Nvidia’s technology is on the table. The project could still be shelved, or ship without it. Nvidia already sells networking to builders using non-Nvidia processors.
A commercial server would be a strategic shift. Apple runs Private Cloud Compute hardware today but keeps it internal and has declined partner requests to use it. Separately, the company is designing a cloud chip for 2027 deployment with Broadcom’s help that carries the codename Baltra.
Memory economics cut both ways. Unified memory gives Apple large model capacity per node, yet it trails Nvidia on bandwidth, and AI demand has already driven memory prices sharply higher.
