Apple is building an enterprise AI server powered by its M8 Ultra chips, targeting a 2029 release with configurations of two or four processors per unit—marking the company's return to the server market after exiting with Xserve in 2011.
CEO John Ternus backed the project a year ago while leading hardware engineering, before taking the top job Sept. 1. The server addresses real demand: OpenAI has acquired tens of thousands of Mac minis and Mac Studios for reinforcement learning training, while Anthropic has rented Mac mini units from Amazon Web Services.
Apple is exploring integration of Nvidia's NVLink Fusion networking technology to connect M8 chips within the server, according to discussions between the companies. But the project remains conditional—Apple could abandon it or proceed without Nvidia's tech.
The move reflects Apple's shift from consumer-focused hardware to capturing AI infrastructure economics. Mac mini and Mac Studio sales have surged as developers discovered they offer cost-effective alternatives to traditional accelerators for certain workloads. An Apple-built server would let the company control the entire stack and capture margin on the hardware layer of AI operations—a business model that has worked for Nvidia in GPUs but remains unproven at Apple's scale in servers.
The effort faces headwinds: a memory chip shortage driven by rapid data center expansion has spiked costs across consumer and enterprise hardware, pressuring both Apple's margins and customer willingness to pay premium prices for alternative architectures.
