Apple considers return to server market after 18 years, in talks with Nvidia

Apple is considering a return to the enterprise server market for the first time in 18 years and is in discussions with Nvidia about using its server interconnect technology, The Information reported Wednesday (local time), citing sources.

According to the report, Apple is exploring the development of enterprise AI servers powered by its in-house M8 Ultra chip, targeting AI developers, corporations and government agencies. The company is also in talks about incorporating Nvidia's open server interconnect technology, NVLink Fusion, into the design.

The servers under consideration would come in two configurations: a smaller model linking two M8 Ultra chips and a larger one connecting four. Both would combine multiple chips into a unified computing system to handle AI inference workloads. The target launch date is 2029, according to the report.

However, the project remains in its early stages, and plans could change or be scrapped entirely during development, The Information said.

A return to the server market would mark Apple's first such venture since it discontinued its Xserve server line in 2011 — roughly 18 years ago. People involved in that project said it was largely a failure, attributed to a lack of interest in the enterprise server market and insufficient customer support.

NVLink Fusion is an open interconnect technology Nvidia unveiled last May that allows non-Nvidia chips to connect across servers using Nvidia's networking infrastructure.

Amazon Web Services decided to adopt the technology last December and said last month it would expand its use.

Dion Harris, Nvidia's senior director of high-performance computing and AI infrastructure, said the company's networking equipment strategy is expanding from in-house production to open collaboration.

The discussions come amid observations that the long-strained relationship between Apple and Nvidia has recently begun to thaw.

The two companies fell out in 2008 over a dispute about sharing after-sales service costs stemming from a graphics card defect, a rift so deep that Nvidia GPUs disappeared from Apple's Mac lineup after 2019.

The animosity ran so deep that Steve Jobs was said to have treated Nvidia executives as if they were invisible during meetings.

A turning point came when Apple, in revamping its Siri voice assistant unveiled in June, pivoted toward using Nvidia's Blackwell chips on Google Cloud — a move observers said helped lay the groundwork for a rapprochement.

Apple's consideration of a server market re-entry also aligns with its broader push to expand its hardware business in response to surging demand for AI computing.

Apple's Mac sales reached $10.4 billion in the third quarter of its last fiscal year, up 29 percent from the same period a year earlier, with analysts attributing the growth in part to rising demand for Mac Studio and Mac Mini units used for AI inference.

The Information noted that if Apple wants to build a competitive enterprise server business, it will need to hire staff and offer software tools comparable to Nvidia's CUDA.

Awni Hannun, the researcher who led Apple's on-device AI framework MLX, has left the company to join Anthropic.

OpenAI is also reportedly developing its own AI chip, known as "Jalapeño," underscoring the intensifying race among major technology companies to build proprietary AI infrastructure.

Adding to the irony, Taiwan's Foxconn — Apple's longtime iPhone contract manufacturer — has grown its AI server business to the point where it could become a potential competitor to Apple.

Foxconn is already Apple's largest hardware partner and the world's top AI server manufacturer. Should Apple enter the AI server market with its own product, it could once again turn to Foxconn for assembly — putting Foxconn in the position of building competing AI servers for multiple big tech companies at once.