Reports suggest that Apple is exploring the development of an enterprise AI server, which would be powered by two or four of its upcoming M8 Ultra chips. The company is considering using Nvidia's NVLink Fusion technology to connect these chips, as current Thunderbolt-based solutions for Mac clustering lack the necessary bandwidth for data center-level performance. The project, which could launch as early as 2029, is still in the planning stages, with no final decisions made on the server design or any partnerships with Nvidia. If realized, this would mark Apple's first purpose-built server since the Xserve, a product it discontinued in 2011 due to low demand. At the time, Apple suggested that users run server software on Mac minis and Mac Pros instead of using traditional rack-mounted servers. However, recent interest in Apple hardware from AI companies may be changing the landscape. Reports indicate that organizations like OpenAI have purchased large numbers of Mac minis and Mac Studios to train AI models, and Anthropic has rented Mac minis through Amazon Web Services. Apple, however, was initially unprepared for this surge in demand. The company lacked an enterprise AI strategy, dedicated engineering teams, and specific staff to manage large orders. As a result, Apple reportedly turned away some business clients seeking access to its Private Cloud Compute infrastructure. Despite this, Apple’s current CEO, John Ternus, who previously led hardware engineering, is said to have been a strong advocate for the project. The challenge lies in connecting multiple chips efficiently. Apple’s current M5 Ultra chip, found in the Mac Studio, uses a technology called UltraFusion to link four dies in a single package, offering high performance for on-device AI tasks. However, scaling this to a server-level product would require significantly more power and faster chip interconnects. Thunderbolt 5, Apple’s current standard, offers up to 10GB/s of bidirectional bandwidth, which pales in comparison to Nvidia’s NVLink Fusion, which can connect up to 72 accelerators at 3.6TB/s each. This highlights why Apple may be seriously considering Nvidia’s technology to meet the high-performance demands of enterprise AI workloads.