Product2 publishers2 min readPublished
Apple weighs wiring an M8 Ultra AI server together with Nvidia's NVLink Fusion
The Information reports two configurations, one with two chips and one with four, aimed at developers, businesses and governments, with a possible 2029 launch. Apple already builds servers for Private Cloud Compute and has turned partners away from them.
The Product Desk · Product desk
What happened
- The Information reports Apple is developing an AI-focused server built on its planned M8 Ultra chips and intends to sell it to AI developers, other businesses and governments instead of only running it internally.
- Two configurations are under consideration, a smaller one clustering two M8 Ultra chips and a larger one with four.
- Apple has discussed linking those chips using Nvidia's NVLink Fusion, a set of switches, chiplets and software for chip-to-chip communication inside data centers.
- The plans are not final, the report puts the earliest sale in 2029, and it says the project could be canceled before then.
- Work began roughly a year ago with the backing of John Ternus, who was then Apple's hardware engineering chief and is now its chief executive.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- decision Teams already buying Mac minis and Mac Studios by the pallet for inference have nothing new to order, so their live choice is how much further to build on Apple's toolchain for a machine that may never ship.
- constraint Support commitments will gate procurement here, and a regulated buyer cannot put an unbuilt business-support organisation into a 2029 plan.
- exposure Nvidia would collect networking revenue from a box sold against its own AI systems. Its fabric business and its systems business end up on opposite sides of the same rack.
- precedent If Apple takes NVLink Fusion, Nvidia's interconnect becomes the default plumbing even in racks that deliberately avoid Nvidia processors, and the next alternative-silicon vendor has a well-worn path to follow.
The buyers for this machine already have a workaround on the shelf. AI companies have been buying Mac minis and Mac Studios in large quantities, according to The Information [15], and 9to5Mac says people are already using clusters of Mac Studios to run AI models locally [16].
The dates matter for anyone tempted to put this in a roadmap. 9to5Mac published the report on 16 September 2026 [19], the earliest sale is 2029 [5], and the work started about a year ago [6]. That is roughly three years of waiting on top of a year already spent, about four years from start to shipment [18].
Apple already builds servers. It makes the hardware behind Private Cloud Compute, and it has declined requests from partners to use those machines [10]. It sold servers until it discontinued Xserve in 2011 [8]. The change described in the report is a box outside customers can buy and operate themselves, aimed at companies that want to run models on their own equipment with a focus on inference [9].
Nvidia is in the story because of an internal limit. Scaling Apple's existing connections between chips would present cost and speed problems, according to the report, and that is why NVLink Fusion is under discussion [11]. Apple has already announced plans to extend Private Cloud Compute to Google Cloud using Nvidia GPUs [14].
The report's list of remaining work is mostly not hardware. It says there is still a need for stronger support for business customers and more software resources for AI developers, including further investment in Apple's MLX framework [17].
Start with whether the models you serve run well enough on Apple silicon and MLX today, since that is the only part of this a team can measure in 2026. Then ask whether the cost of building on that toolchain is recoverable if the server never arrives, because the report says the project could be cancelled [5]. The only Apple hardware an inference team can order in 2026 is the desktop line those AI companies are already buying [15].
What to watch
- Whether the NVLink Fusion discussion becomes an announced Apple-Nvidia arrangement, which the report says is not finalized.
- Whether Apple funds the enterprise support organisation and the MLX developer resources the report says are still needed.
- Whether Apple reverses its refusal to let partners run on its Private Cloud Compute servers.