Fifteen years after walking away from servers entirely, Apple might be circling back. The Information reports that Apple is weighing an AI server based on its upcoming M8 chips, and the company has held talks with Nvidia about using its networking hardware to tie those chips together.
Apple hasn’t sold server hardware to outside customers since it killed off Xserve back in 2011. This time, the target audience looks different. Reports point to companies wanting to run AI models on hardware they own and control themselves, with a heavy emphasis on inference work, the process of actually generating responses from models that have already been trained.
Nvidia’s NVLink Fusion is the specific piece Apple has reportedly looked at. It handles the hardware and software side of moving data between chips quickly, something Apple would need if it wants multiple M8 processors working together inside a single server. Nothing here is locked in yet. Neither the Nvidia partnership nor the server project itself has been finalized, and the earliest anyone expects a release is 2029, assuming Apple doesn’t scrap the whole thing first.
Interestingly, current CEO John Ternus reportedly backed this project roughly a year ago, back when he still ran Apple’s hardware engineering division rather than the whole company. That timing lines up with a broader trend Apple has noticed lately: AI companies buying up Mac mini and Mac Studio units in unusually large numbers, treating them as compact, efficient compute options rather than typical consumer machines.
Apple isn’t new to running server infrastructure either. It already builds the hardware behind Private Cloud Compute, the system that handles Apple Intelligence tasks too demanding for a single device to process locally. According to the report, scaling that existing chip-to-chip connection technology further would run into real cost and performance problems, which is reportedly the exact reason Apple started looking at Nvidia’s alternative in the first place. Apple has also turned down outside requests to use its Private Cloud Compute servers directly, keeping that infrastructure strictly internal so far.
A deal like this would put Nvidia in an unusual position, helping build a product that could eventually compete with Nvidia’s own AI hardware business. NVLink Fusion technically exists to let Nvidia sell its networking tech to companies building competing processors, so this wouldn’t be entirely out of character even if it seems counterintuitive on the surface.
There’s a bigger relationship shift happening here too. Apple and Nvidia have a long history of friction, yet the two companies already revealed their plans to bring Private Cloud Compute onto Google Cloud running on Nvidia GPUs earlier this year. Reports suggest the two sides have also discussed other collaboration opportunities, including potential use of Nvidia’s open-source AI models within Apple’s own systems.
Even if Apple moves forward, the hardware itself is only part of the challenge. The report notes Apple still needs stronger business support infrastructure and more resources for AI developers, including continued investment in MLX, Apple’s own machine learning framework built for Apple silicon.
