Apple is building a new server designed for artificial intelligence workloads, powered by multiple units of its most powerful in-house processors. The system could reach the market in 2029, marking the company's return to the server business after discontinuing the Xserve in early 2011.
Ars Technica reports the development based on coverage from The Information. Apple is exploring two configurations: one with two future M8 Ultra chips and another with four. The project began roughly a year ago and received backing from John Ternus, who led hardware development at the time and became Apple's CEO this month.
Mac hardware gains traction in AI data centers
The server plans come as Apple hardware sees growing adoption for AI workloads. Mac mini and Mac Studio systems are particularly popular. AI companies including OpenAI have purchased tens of thousands of these computers, according to The Information. They use the machines to train AI agents through reinforcement learning, a process where models learn which actions produce the best results through repeated attempts.
Anthropic also uses Apple hardware, renting Mac mini systems through Amazon Web Services. This developer interest gives Apple a foothold in a market it exited directly nearly two decades ago. For IT professionals building or managing AI for IT & Development workloads, the shift signals that Apple silicon is becoming a practical option beyond consumer devices.
Nvidia link technology under consideration
Apple may use Nvidia technology to connect the chips inside the server. The two companies are in talks about NVLink Fusion, which allows processors and accelerators within high-performance AI systems to communicate with each other. Apple is investigating whether the technology can link multiple M8 chips in a single chassis.
No decision has been made. The server project remains in development and could still be canceled. Apple might also release the system without Nvidia's interconnect technology.
Memory supply threatens timeline
The ongoing shortage of memory chips complicates the project. Rapid expansion of AI data centers has driven up demand sharply, affecting servers, PCs, smartphones, and other consumer electronics. Higher memory costs have already contributed to price increases across several Apple products.
For a server equipped with multiple Ultra chips and large memory pools, availability of memory components could become a deciding factor in whether the 2029 launch proceeds as planned.
Why this matters for IT and development professionals
Apple re-entering the server market would introduce a new hardware target for AI workloads that already run on Mac silicon in data centers. Developers optimizing reinforcement learning pipelines or deploying inference at scale may need to evaluate whether Apple's unified memory architecture and chip interconnects offer performance or cost advantages over existing x86 and GPU clusters. The 2029 timeline means teams have years to watch for developer tooling, OS support, and integration with orchestration platforms - but the architecture decisions being discussed now will shape those options. For software developers looking to build skills that map to this hardware direction, an AI Learning Path for Software Developers provides relevant grounding in the frameworks and deployment patterns these systems will likely support.
Your membership also unlocks:
