Apple Plans a Return to Servers: M8 Ultra AI Inference Machines by 2029

Apple is planning a return to the server business it exited two decades ago, and the target is the hottest part of the market. Per The Information and Reuters’ follow-up, Apple is developing enterprise AI inference servers powered by its own chips, in configurations pairing two or four M8 Ultra chips, has discussed using Nvidia’s NVLink Fusion networking technology to connect them, and is targeting a launch not before 2029. The plans are not finalized and could change.

Why Apple’s Silicon Logic Points at Servers

The M8 Ultra inference play follows the same logic as every Apple silicon generation: memory bandwidth and power efficiency. Inference is the workload where those attributes matter most, because serving models is dominated by memory movement and energy cost rather than raw training FLOPS. Apple’s Ultra chips already pair massive unified memory with best-in-class performance per watt, and an inference appliance built on that foundation could undercut GPU economics for the workload every enterprise is discovery-limiting on right now: token serving. The Nvidia conversation is the tell that Apple understands its gap: NVLink Fusion, Nvidia’s open networking standard announced for third-party silicon, would let M8 Ultra clusters plug into the accelerator ecosystem rather than fight it.

What to Watch: 2029 Is a Statement About Sequencing

The date is the strategic information. A 2029 target means Apple intends to enter after the current accelerator buildout matures, with its own next-generation silicon and, presumably, its own AI services (the company is already building Private Cloud Compute capacity for Apple Intelligence and just shipped the Gemini-trained Siri AI). Reading the sequence: Apple lets the training-infrastructure war (the one Dell’s backlog and the neocloud financing rounds are fighting) play out without it, then enters at the inference layer where its efficiency advantage compounds and its privacy positioning differentiates. The risk is equally clear: enterprise servers are a relationships-and-roadmap business, Apple has been out of it since the xserve era, and NVLink Fusion participation on Nvidia’s terms is a dependency, not independence. Watch whether the NVLink talks conclude, because that single answer determines whether these are Apple’s servers or Nvidia’s network with Apple-branded nodes.