The Information reports that Apple is considering a return to the server business with machines built for artificial intelligence. The proposed hardware would come in two versions, one clustering two M8 Ultra chips and another with four. Apple has also reportedly discussed using Nvidia’s NVLink Fusion technology to connect the chips.

The target customers would be AI developers, businesses, and governments. The servers aren’t expected before 2029, and the project could still be canceled.

I hope it isn’t. Apple should absolutely do this.

The potential customer pitch is also easy to see. Plenty of organizations want capable AI models without sending every request to somebody else’s cloud. Developers want hardware they can control. Companies have sensitive data. Governments have even more of it.

If an M8 Ultra server carries forward the unified-memory architecture and performance-per-watt advantages of Apple’s current chips, it could be an interesting machine for private AI inference. Its job would be narrower than replacing Nvidia across the industry. Instead, it gives buyers another credible option for private workloads that demand local control without wasting power.

I see so many lawyers, doctors, financial experts, and others dealing with sensitive data going through the Robot Assistant Field Guide who are begging for hardware like this. If done right, I expect they’d sell these like hotcakes.