Apple's AI server initiative unveiled: baltra chip and nvlink fusion integration

Apple is reportedly developing an enterprise-level AI server, codenamed 'Baltra,' which will use its custom M8 Ultra chips and potentially NVIDIA's NVLink Fusion technology. The server is aimed at AI developers, enterprises, and government agencies, with versions featuring two or four M8 Ultra processors. The M8 Ultra, still in development, is described as Apple's most powerful processor. The Baltra chip, developed in partnership with Broadcom, will use TSMC's 3nm N3E process and a Chiplet architecture. Apple is responsible for assembling the small chips into a single package, maintaining design confidentiality. The server may leverage NVLink Fusion, a high-speed interconnect technology introduced by NVIDIA in 2025. This technology allows for high-speed communication between custom chips and NVIDIA GPUs, with the fifth-generation NVLink supporting 72 GPUs at 1800 GB/s. Apple's last dedicated server offering was the Xserve, discontinued in 2011. The Mac business has seen strong growth, with $10.3 billion in revenue in the latest quarter, driven by bulk purchases from AI companies.