Rumor mill: Apple could be exploring plans for an AI server powered by its future M8 Ultra chips and potentially Nvidia networking technology. The system would be aimed at companies that want to run AI models on their own hardware, particularly for inference work. A launch would not come before 2029, and the project could still be shelved.
According to a report in The Information, Apple is considering versions with either two or four M8 Ultra chips. It is also discussing whether to use Nvidia's NVLink Fusion technology to connect the chips. NVLink Fusion combines hardware and software designed to move data between processors. Apple's existing methods for linking chips could become too expensive or too slow at the scale required for a server.
However, Nvidia's involvement is not certain. Apple could build the server without NVLink Fusion.
The idea follows growing demand for Apple's higher-end Macs from AI companies. OpenAI has bought "tens of thousands" of Mac mini and Mac Studio systems, according to the report. The machines have been used for reinforcement-learning work in which AI agents improve through repeated trial-and-error runs. Anthropic has also rented Mac minis from Amazon Web Services.
Those purchases have shown that Apple's M-series chips can appeal to AI developers beyond their traditional Mac user base. Mac mini and Mac Studio systems offer Apple silicon in relatively compact form factors, while the proposed server would package several future Ultra chips into equipment intended for business customers and data-center deployments.
Apple already runs its own server hardware through Private Cloud Compute, the infrastructure used for Apple Intelligence requests that need more processing power than an iPhone, iPad, or Mac can provide locally. Apple has not offered those servers to outside customers. The company has also declined partner requests to use its Private Cloud Compute hardware, according to the report.
A commercial server would mark a larger move into enterprise infrastructure. Apple last sold a server through its Xserve line, which it discontinued in 2011. The proposed product would have a narrower role than Xserve: it would be built around Apple silicon and designed for AI inference, rather than general-purpose server workloads.
The effort also reflects a potentially closer relationship between Apple and Nvidia. Apple has announced plans to expand Private Cloud Compute to Google Cloud infrastructure using Nvidia GPUs. The companies have also reportedly talked about other AI work, including Nvidia's open-source models.
But the project would arrive in a difficult hardware market. Demand from AI data-center operators has tightened memory-chip supplies and increased costs for systems that need large amounts of memory. That could be a particular issue for Apple because its M-series architecture uses unified memory, which is important for handling larger AI models.
Apple would also need to build out the business side of the effort. Enterprise customers typically expect deployment help, developer tools, software support, and long-term service commitments. The report said Apple would need stronger support for business customers and more resources for AI developers, including further work on its MLX machine-learning framework.
Apple could return to the server market with M8 Ultra hardware and Nvidia networking
