Apple could return to selling enterprise servers after more than a decade's absence, according to The Information. The Cupertino company is reportedly working on an artificial intelligence server that would use its upcoming M8 Ultra processors and could incorporate Nvidia's NVLink Fusion interconnect technology. The product, still in development, would be offered in two variants and wouldn't reach the market before 2029, although the plan could still be canceled or substantially changed.
The initiative comes at a time when demand for AI hardware has skyrocketed , and the Mac mini and Mac Studio have become a common choice for developers seeking local processing power. Apple has reportedly held talks with Nvidia about using its inter-chip interface, a noteworthy move given the strained history between the two companies. According to sources, Apple engineers believe Nvidia's solution is the best available for this project.
A long-awaited return after Xserve
Apple abandoned the dedicated server segment in 2011 with the discontinuation of Xserve. Since then, the Mac Pro, Mac Studio, and Mac mini have assumed server functions, but without a specific offering for data centers. The new project would revive that category with a focus on AI inference —that is, running pre-trained models to generate responses. The configurations being considered include a version with two M8 Ultra chips and a more powerful one with four, according to the information gathered.
This decision is no coincidence. Apple's current CEO, John Ternus, supported the project when he led hardware engineering, and his appointment to head the company has given the initiative a boost. Ternus is very familiar with Apple's internal infrastructure, such as Apple Private Cloud Compute , which processes Apple Intelligence tasks on its own servers. However, these servers are not sold to third parties, and in fact, Apple has rejected requests from partners to use them.
NVLink Fusion: Nvidia's connection
The potential collaboration with Nvidia is one of the highlights. NVLink Fusion is an interconnect platform that allows chips from different manufacturers to work together in the same rack, sharing memory and data at high speed. Apple is reportedly evaluating this technology to link the M8 Ultra processors , as scaling its own chip-to-chip connections would be expensive and slow, according to the report. Data transfer between processors is critical when an AI model is distributed across multiple chips, and efficient interconnection reduces latency.
Nvidia has already licensed NVLink Fusion to companies like Qualcomm, Arm, Fujitsu, and Marvell, and Apple could be next. This would give Nvidia a role in a product that could potentially compete with its own AI systems, but the company seems keen to expand its ecosystem. The talks don't guarantee a final agreement , and the server could be released without this technology or even canceled altogether. Neither Apple nor Nvidia has officially confirmed these plans.
The role of John Ternus and Mac's demand for AI
Demand for Macs for artificial intelligence has been a key driver. In the last fiscal quarter, the Mac division grew nearly 29% year-over-year, fueled by purchases from developers who use Mac mini and Mac Studio to train and deploy models. This pressure has led Apple to rethink its offerings, developing a more professional solution that addresses needs such as remote administration, energy efficiency, and enterprise support—details of which are still forthcoming.
John Ternus, who took over as CEO in September 2026, has prioritized this project. His hardware expertise and knowledge of the internal infrastructure would be key to bringing the product to market. Furthermore, Apple has already integrated Nvidia GPUs into its private cloud for demanding workloads, demonstrating a more pragmatic relationship in recent years.
Possible configurations and schedule
The server, if confirmed, would be offered in two versions: one with two M8 Ultra chips and another with four. These chips have not yet been officially announced, and Apple may be developing the M7 and M8 Ultra in parallel, according to some sources. The launch is not expected before 2029 , and it remains to be seen whether the project will survive internal changes and negotiations with Nvidia. Shares of both companies barely moved after the news broke, suggesting the market is still hesitant.
In any case, the news confirms Apple's interest in AI hardware and its willingness to collaborate with a long-time competitor. The company already manufactures servers for its own use, but opening up to the external market would be a significant step. Developers and companies that need local computing power will be watching closely for the next moves, even though the wait may be long.
Table of Contents