Apple Reportedly Plans AI Server With M8 Ultra Chips
Two configurations with two or four M8 Ultra chips, and no launch before 2029: Apple is weighing a return to selling server hardware.
In short
Apple is reportedly developing a data center server built around two or four of its own M8 Ultra chips to run trained AI models, with no launch expected before 2029.
At a glance
- Two planned configurations: one with two M8 Ultra chips, one with four.
- The target workload is inference – running trained models, not training them.
- A launch is not expected before 2029, and the project could still be dropped.
- Apple is evaluating Nvidia's NVLink Fusion to link the chips inside the rack.
- Mac revenue rose nearly 29 percent last quarter to $10.4 billion.
Apple is building a data center server around its own M8 Ultra silicon, according to a report by The Information. Two configurations are said to be planned, one with two chips and one with four. The intended buyers are AI developers, businesses and government agencies. A launch is not expected before 2029.
Inference, not training
The box is meant to run models that have already been trained rather than to train new ones. That is a deliberately narrower job than the one the largest training clusters are bought for. It is also the workload where Apple silicon's memory bandwidth and power profile have the most to offer.
A Nvidia link in an Apple machine
For fast chip-to-chip communication inside a rack, Apple is weighing Nvidia's NVLink Fusion. The interconnect started out as Nvidia-only technology and was later opened to third-party hardware. Whether Apple actually adopts it is unsettled; the report explicitly leaves room for the project continuing without it.
Who is pushing it
Chief executive John Ternus is said to have backed the effort roughly a year ago, while he still ran hardware. The commercial case has grown since then. AI labs including OpenAI and Anthropic are reportedly buying Mac Mini and Mac Studio units in bulk for AI workloads, and Mac revenue climbed nearly 29 percent last quarter to $10.4 billion.
What is not confirmed
None of this comes from Apple. The report itself allows that the program may be cancelled outright. For this article only one of the two available write-ups could be retrieved, and it in turn relies on The Information, so no second independent newsroom was available for cross-checking. Memory capacity, pricing and sales channel are absent from the reporting that could be read.
FAQ
Which chip would Apple's AI server use?
The M8 Ultra. The report describes two variants, one carrying two of the chips and one carrying four.
When would Apple's AI server launch?
Not before 2029, according to the report, and the project may be cancelled before it ships.
Would the server be used to train AI models?
No. It targets inference, meaning it runs models that have already been trained for business and government customers.
