This article may contain affiliate links. Learn more on our Partner Program page.
Apple left the server market in 2011 with the Xserve. Now, a new attempt is reportedly in the works, this time for companies that want to run AI models on their own hardware. Surprisingly, Nvidia could supply a key component.
Apple is developing plans for a server with its own chips, which it intends to sell to external customers. This is according to The Information, citing sources familiar with the project. The server will reportedly be based on M8-generation chips, and Apple has apparently been in talks with Nvidia regarding the interconnectivity of the processors.
The background to this is demand from the AI industry. Companies are buying Mac minis and Mac Studios in large quantities to run AI models locally – the AI demand for Mac minis and Mac Studios had surprised Apple, according to an earlier report in the same newspaper. A dedicated server would represent the step from desktop computers to data center equipment.
Key Facts at a Glance
- Apple is reportedly considering an AI server with M8-generation chips for external customers.
- Nvidia's NVLink Fusion connection technology is being discussed as a way to link multiple Apple chips.
- A possible market launch is mentioned for 2029, but the project could be discontinued before then, according to the report.
- Apple CEO John Ternus is said to have supported the project when work began about a year ago.
What is known about the planned server
The server is intended for companies that want to run AI models on their own hardware. According to the report, the focus is on inference, i.e., generating answers from already trained models.
Nothing is set in stone. Neither the server itself nor an agreement with Nvidia is finalized. A possible launch date is 2029, though an earlier shutdown is explicitly possible. That would leave roughly three years between now and the launch – and 18 years since the end of Xserve.
According to the report, the project was initiated about a year ago when John Ternus was still leading hardware development. He is said to have supported the work at that time. He has been Apple's CEO since September 1st.
What role could Nvidia play?
One of the options under consideration is NVLink Fusion. Nvidia offers hardware and software that allows data to be exchanged between chips – even between processors from other manufacturers. Apple could use this to connect multiple M8 chips in a single system.
This would be an unusual situation for Nvidia. The company would be supplying technology for an Apple product that could compete with its own AI systems. At the same time, it fits with the strategy behind NVLink Fusion, which is to also sell the connection technology to companies with their own processors.
This collaboration wouldn't be the first. Apple runs the most powerful model of the third Foundation generation on Nvidia GPUs in Google servers that are exclusively reserved for Apple. The overview of Apple Intelligence and Data Privacy describes how this setup works.
Why Apple doesn't simply use its own technology
Apple already has its own server hardware. It runs in Private Cloud Compute, the infrastructure for Apple intelligence requests that the device itself doesn't process. However, according to the report, scaling Apple's existing chip connections to server size would cause cost and speed issues – hence the look at Nvidia.
Apple reportedly rejected requests from partners to use its own private cloud compute servers. Its own server infrastructure was already under scrutiny in the spring, following reports of unused AI servers at Apple based on modified M2 Ultra chips.
What the Mac Studio can already do today
Apple already sells hardware for local AI inference. The Mac Studio with M5 Ultra can be ordered with up to 512 GB of shared memory – a configuration that Apple says will be available at the end of October. The return of 512 GB to the Mac Studio was one of the key points of its presentation in August.
Multiple devices can also be paired. According to Apple's announcement of the new Mac Studio, Thunderbolt 5 enables clusters of multiple Mac Studio units that can perform distributed AI inference up to three times faster than a single system. The connection between the chips is also the point that Apple, according to the report, does not intend to scale with its own technology for a server.
What Apple still needs for a server business
The report itself identifies hurdles beyond hardware. Apple needs stronger support for business customers and more software for AI developers, including further investment in its own MLX framework.
The sources suggest an intention, not a plan. The Information relies on sources close to the project, not on statements from Apple or Nvidia. The timeframe mentioned is three years in the future, and the report explicitly considers a discontinuation a possibility.
For companies that now need local AI hardware, nothing changes: Apple's offering for this remains the Mac Studio, individually or in combination via Thunderbolt 5.
Do you think Apple can compete with Nvidia in the data center, or will the Mac Studio remain Apple's more realistic AI machine? Share your prediction in the comments.
