Apple is exploring a return to dedicated server hardware for the first time since it killed Xserve in 2011, developing enterprise AI inference systems around forthcoming M8 Ultra chips and discussing Nvidia’s NVLink Fusion to link those processors, according to a new report from The Information.
The planned machines would be sold to AI developers, businesses, and governments that want to run models on their own equipment rather than in the public cloud. Two configurations are under consideration: a two-chip system and a four-chip version built around Apple’s highest-performance M-series silicon. A launch is not expected before 2029, and the project could still be canceled or ship without Nvidia technology.
The effort began about a year ago with support from John Ternus, then Apple’s hardware chief and now CEO. Demand from AI teams already buying Mac mini and Mac Studio systems in volume appears to have sharpened Apple’s interest in a rack-scale product. Apple already builds servers for its own Private Cloud Compute stack, but has declined requests from partners to sell that hardware externally. Internal interconnect options were viewed as too slow and expensive for larger deployments, which is why NVLink Fusion is on the table.
NVLink Fusion is Nvidia’s suite of switches, chiplets, and software for high-bandwidth, low-latency links between chips, including processors that are not Nvidia GPUs. Opening that layer to other silicon makers lets Nvidia sell networking even when customers choose alternative compute. An Apple deal would be notable given a long, chilly relationship that dates to GPU reliability problems in MacBooks nearly two decades ago. More recently, Apple has used Nvidia hardware in cloud infrastructure supporting a revamped Siri.
Neither company commented, and the report has not been independently verified. Shares of Apple, which had been lower earlier, reversed modestly in premarket trading after the story circulated.
If it reaches market, the product would put Apple silicon into enterprise AI infrastructure rather than only consumer devices and Macs. It would also test whether/prove that efficient Arm-based M chips can compete in inference at rack scale, and whether Nvidia can remain the connective tissue even in systems built around someone else’s processors.
MacDailyNews Take: Even if nothing comes from this, it is already so nice to have “a product person, per se” back in charge at our favorite company! 😃
To paraphrase Steve Jobs, when the product people get driven out of the decision-making process, companies forget what it means to make great products. The product sensibility and the product genius that brought them wild success gets rotted out by people running these companies that have no conception of a good product versus a bad product.
Please help support MacDailyNews — and enjoy subscriber-only articles, comments, chat, and more — by subscribing to our Substack: macdailynews.substack.com. Thank you!
Support MacDailyNews at no extra cost to you by using this link to shop at Amazon.
I wonder what OS the customers want to run on these machines?
I recommend they consider a BSD Unix variant.
macOS is based on BSD Unix always has been
servers have to be redefined like the mac pro. for example is there any reason unified memory must be so slow. any reason why apple can’t move on to pci 8 and a graphic processing unit with 128 megabytes per card. pci 8 is looking slow but increasing it to 128 bit. servers no longer need to be big or have redundant power supplies.
With a product guy at the helm (and reportedly Ternus playing a significant part in birthing the “secret project” in ’25), this consideration seems quite timely. With the mini being an agentic AI choice for many…adding a server to the mix could mean the “Xserve” could play more broadly and actually endure as more than a bespoke option.
A serious/hopeful development, imo.