Apple is developing a server using its own ARM-based chips for AI workloads, and may use Nvidia’s interconnect technology with the potential product, The Information reported, amid a boom in demand for AI processing power.

The server could be sold to AI developers, governments and other companies, the report said, citing multiple unnamed sources.

Two versions of the system are in development, a smaller one using two of Apple’s planned M8 Ultra chips clustered together, and a larger one that would use four M8 Ultra processors.

The servers, if Apple decides to go ahead, could come to market in 2029, the report said.

Apple announced its M6 range of chips in August.

Apple has reportedly considered connecting the chips using Nvidia’s NVLink Fusion, an on-chip communications technology based on Nvidia’s NVLink standard that Nvidia licences to companies including ARM, SiFive and Amazon Web Services for use in their data centre chips.

Apple has been working on the server project for around a year, according to the report.

John Ternus, Apple’s former head of hardware development who took over as chief executive this month, is reportedly in favour of the product.

Mac processing power

Apple previously made the Xserve range of servers, but discontinued it in 2011.

The report comes at a time of booming demand for AI processing power.

Apple’s powerful and energy-efficient ARM-based chips have proven popular with AI developers, who have clustered together multiple Mac Mini and Mac Studio computers running M-range chips to run AI workloads, leading to shortages of the devices.