Apple is discussing an AI server built around future M8-series Apple silicon that it would sell to third-party enterprise customers, potentially ending its 18-year absence from the server market since it discontinued Xserve in 2011. According to reporting by The Information surfaced on September 16, 2026, Apple engineers have held internal discussions with Nvidia about using NVLink Fusion to interconnect M8 processors inside the new box.
Chasing The Local-AI Enterprise
The proposed server would be aimed at companies that want to run large AI models locally rather than lean on cloud services, with an emphasis on inference workloads. It is an obvious extension of Apple silicon's momentum: enterprise buyers have already been bulk-ordering Mac mini and Mac Studio systems as low-power AI inference nodes, and Apple's own Private Cloud Compute stack, which handles sensitive Apple Intelligence tasks, has quietly proved that its chips can run at data-center scale.
Why It Needs Nvidia NVLink Fusion
Chaining a large number of M8 chips together into an inference cluster is where Apple's current in-house fabric hits cost and latency limits, according to the report. Nvidia's newly opened NVLink Fusion program lets third-party silicon plug into the same high-bandwidth interconnect that binds together Nvidia's own GPUs, offering Apple a way to scale M8 rack designs without inventing a competing fabric from scratch. It would also be a striking role reversal after years in which Apple minimised its reliance on Nvidia hardware inside macOS.
Timeline, Risks And Competitive Read
Sources cited in the report peg a potential launch at 2029, with the project explicitly unfinalised and cancellable. Even so, the direction of travel matters: Apple is signalling that it wants a piece of the enterprise inference market currently dominated by Nvidia DGX systems and by hyperscaler-provided racks. That reads as one more pressure point on the space we covered in our reports on NVIDIA's Hugging Face acquisition and the Cisco-Splunk AI Pod launch, both of which are chasing the same on-prem-AI budget.
What To Watch
Two signals would confirm the plan is real. First, an M8-class silicon variant tuned for rack density and thermals rather than the MacBook envelope. Second, a public NVLink Fusion partner slot for Apple alongside the customers Nvidia has already listed. Without those, the current disclosure is closer to a directional bet than a product commitment — but it is the loudest Apple has spoken about enterprise hardware since discontinuing Xserve.
Reporting based on coverage from The Information, MacRumors, Digitimes and The Verge.