Reports indicate Apple is evaluating a return to the server market by developing custom AI accelerators tailored for its own inference workloads.

Apple is exploring an AI server that would utilize its planned M8 Ultra chips and connect via Nvidia’s NVLink Fusion networking, according to The Information. The concept may not reach customers until 2029.

Apple has not sold a purpose-built server since discontinuing the Xserve in 2011, making even preliminary research a notable shift from its traditional focus on Macs and consumer devices. The Information reports that the proposed machine would be built around two or four M8 Ultra processors. NVLink Fusion could link these chips through the same high-speed interconnects that power many Nvidia-centric AI data centers. That matters because, as the report notes, “the hard part” in enterprise hardware is often integration. NVLink Fusion is designed to provide external chipmakers with standardized connectors and memory technology, enabling their silicon to slot into existing server racks and networking infrastructure rather than operating within bespoke, one-off clusters. The big caveat is that the initiative is still early and far off: the project could be canceled, delayed, or ultimately ship without Nvidia’s technology.

For markets, NVLink Fusion could serve as the gatekeeper for Apple’s server ambitions. In AI infrastructure, performance isn’t just about the processor; it’s also about how thousands of chips talk to each other without bottlenecks. If Apple uses NVLink Fusion to plug its own chips into the architecture that big data centers already rely on, its server would function as a rack-level enterprise option rather than “a really fast Mac.” At the same time, the development highlights Nvidia’s strategic position: even when a customer wants non-Nvidia compute, Nvidia can still win by supplying the networking layer that makes those systems deployable at scale.