苹果 AI 服务器曝光：代号 Baltra、搭自研 M8 Ultra，或采用英伟达 NVLink Fusion

Apple AI Server Exposed: Code Name Baltra, Uses Self-Developed M8 Ultra or NVIDIA NVLink Fusion AIbase基地 Published inAI News · 5 min read · Sep 18, 2026 5 According to The Information, Apple has discussed using NVIDIA's NVLink Fusion networking technology in a planned AI inference server. The server will be equipped with Apple's custom chips, indicating that Apple may return to the server market. After the news was released, Apple's stock initially declined but later turned positive and saw a slight increase during pre-market trading. The report states that Apple is developing an enterprise-level AI server, targeting AI developers, enterprises, and government agencies. It may offer versions equipped with two or four of the planned M8 Ultra chips. The M8 Ultra is described as Apple's most powerful processor, currently still in development and not yet officially released. Collaborating with Broadcom to Develop the First AI Server Chip Apple has partnered with Broadcom to develop its first AI server chip, internally codenamed "Baltra," which is expected to use TSMC's 3nm N3E process and be designed based on a Chiplet architecture. Broadcom assists Apple in designing each small chip, while Apple is responsible for assembling them into a single package—this isolated development approach allows Apple to keep the overall design confidential from partners like Broadcom. Apple has considered using NVLink Fusion to connect the M8 chips in this planned AI server. NVLink Fusion is a high-speed interconnect technology introduced by NVIDIA at the 2025 Taipei International Computer Show. It opens up the previously proprietary NVLink technology for other manufacturers, enabling customers' custom chips to communicate at high speeds, forming large-scale AI computing systems, and working in conjunction with NVIDIA GPUs. According to NVIDIA's official introduction, the fifth-generation NVLink supports full intercommunication among 72 GPUs, with a communication speed of 1800 GB/s and a total aggregated bandwidth of up to 130 TB/s. MediaTek and Marvell have already become the first partners for this technology, while Fujitsu and Qualcomm also plan to integrate their own CPUs with NVIDIA GPUs using it. The Last Time Apple Sold Servers Was the Xserve Apple's last time selling dedicated server hardware was with the Xserve product line. Launched in 2002, the Xserve used a 1U rackmount design and came pre-installed with the UNIX-based Mac OS X Server operating system. It initially used PowerPC G4 processors and later upgraded to Intel Xeon. Due to a lack of focus on the enterprise market and weak customer support, the Xserve never gained traction and was eventually discontinued on January 31, 2011. After that, Apple shifted its business focus entirely to consumer electronics. One of the factors driving Apple into the enterprise-level AI server market is the strong growth of the Mac business: Apple's latest financial report shows that Mac revenue reached $10.3 billion in a single quarter, with a year-over-year increase of nearly 29%, outpacing iPhone and iPad, making it the fastest-growing segment in the hardware division. The market largely attributes this growth to bulk purchase demands from AI companies. AI Inference ServerApple M8 Ultra ChipNVLink Fusion TechnologyBaltra Chip This article is from AIbase Daily Scan to view Welcome to the [AI Daily] column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with hot topics in the AI field, focusing on developers, helping you understand technical trends, and learning about innovative AI product applications. —— Created by the AIbase Daily Team © Copyright AIbase Base 2024, Click to View Source -