Skip to main content

Apple's Secret AI Server: Baltra Chip and NVLink Fusion

Apple's Secret AI Server: Baltra Chip and NVLink Fusion

Apple is reportedly developing an enterprise-grade AI inference server, internally codenamed Baltra. According to The Information, the company has discussed using NVIDIA's NVLink Fusion networking technology to connect custom Apple silicon, potentially marking a return to the server market after more than a decade. The news sent Apple's stock on a brief dip before it recovered and edged higher in pre-market trading.

The planned server would target AI developers, enterprises, and government agencies. It may come in configurations with two or four of Apple's upcoming M8 Ultra chips—described as the company's most powerful processor yet, though still in development.

A Chip Built with Broadcom

At the heart of Baltra is a custom AI server chip developed in partnership with Broadcom. The chip is expected to use TSMC's 3nm N3E process and a chiplet architecture. Broadcom helps design individual chiplets, while Apple assembles them into a single package. This split approach lets Apple keep the overall design under wraps, even from its partners.

Apple has also considered using NVLink Fusion to link M8 chips inside the server. Introduced by NVIDIA at Computex 2025, NVLink Fusion opens up NVIDIA's proprietary interconnect to other manufacturers, allowing custom chips to communicate at high speeds and form large-scale AI computing systems alongside NVIDIA GPUs.

For context, NVIDIA's fifth-generation NVLink supports full intercommunication among 72 GPUs, with 1,800 GB/s of communication speed and up to 130 TB/s of aggregate bandwidth. MediaTek and Marvell are already early partners, while Fujitsu and Qualcomm plan to integrate their CPUs with NVIDIA GPUs using the technology.

Xserve: The Last Server Apple Sold

Apple's last dedicated server line was the Xserve, launched in 2002. The 1U rackmount system ran Mac OS X Server and started with PowerPC G4 processors before moving to Intel Xeon. But Apple never gained much traction in the enterprise market, and the Xserve was discontinued on January 31, 2011. After that, the company focused squarely on consumer electronics.

So why jump back in now? A big reason: the Mac business is booming. Apple's latest earnings show Mac revenue hit $10.3 billion in a single quarter, up nearly 29% year-over-year—faster growth than iPhone or iPad, making it the standout in the hardware division. Many analysts attribute that surge to bulk purchases from AI companies.

In other words, Apple's AI server ambitions may not be a sudden change of heart, but a natural extension of where its hardware is already winning.

Key Points:

  • Apple is developing an AI inference server codenamed Baltra, possibly with M8 Ultra chips.
  • The server may use NVIDIA's NVLink Fusion for high-speed chip interconnects.
  • Apple partners with Broadcom on a custom AI server chip using TSMC's 3nm process.
  • The move would mark Apple's first server push since the Xserve was discontinued in 2011.
  • Strong Mac sales, driven by AI companies, may be fueling the return.