Apple Reportedly Eyes an AI Server Return With M8 Ultra Chips and Nvidia NVLink

Apple is reportedly considering a return to selling enterprise server hardware, more than a decade after discontinuing Xserve in 2011. The reported product is not a conventional general-purpose server: it would be a rack-mounted system designed for AI workloads, especially inference. Crucially, Apple has not announced the project, and reporting indicates it would not reach the market before 2029 and could still be cancelled. 357

What Apple is reportedly building

According to reports citing people familiar with the work, Apple is developing an enterprise AI server for developers, businesses, and government customers that want to run models on their own infrastructure. Two configurations are under consideration: one with two M8 Ultra chips and another with four M8 Ultra chips. 1713

The M8 Ultra is a future, unannounced Apple-silicon processor, so there are no confirmed specifications, benchmarks, memory capacities, prices, or power figures. The meaningful point is the proposed multi-chip design: Apple appears to be exploring a purpose-built system rather than simply packaging individual Mac desktops for data-center use. 13

Reports describe an emphasis on AI inference—using trained models to generate responses—rather than positioning the hardware as a broad replacement for standard enterprise servers. 311

Why Nvidia’s NVLink Fusion matters

Apple has reportedly held discussions with Nvidia about using NVLink Fusion, a technology intended to provide high-speed connections among chips in AI systems. 111

For a two- or four-chip machine, the interconnect is central. Fast chip-to-chip communication can help the system handle workloads that must exchange large amounts of data across processors, rather than treating each chip as an isolated computer. That could be particularly relevant for running larger AI models across multiple processors.

But the Nvidia element remains tentative. Reports say Apple has considered NVLink Fusion; they do not establish that Apple and Nvidia have signed an agreement, finalized a server design, or committed to shipping a joint product. The server could be cancelled or potentially launched without Nvidia technology. 57

A potential return to the server market, not an Xserve revival

If Apple sells the system externally, it would mark its first return to the enterprise-server business since Xserve was discontinued in 2011. 38 Still, calling it an “Xserve revival” risks obscuring the difference: the reported machine is an AI-focused, Apple-silicon platform, not a confirmed successor to Apple’s former general-purpose server line.

Apple already has experience building server hardware for its own operations. Reporting on its Houston facility says Apple makes AI servers there, but the rumored M8 Ultra project would represent a different commercial step: selling dedicated enterprise AI infrastructure to outside organizations. 926

Why Apple may see demand beyond Mac mini and Mac Studio

Apple’s desktop hardware has attracted AI developers. An Apple silicon executive said the Mac mini and Mac Studio had become popular choices for running AI agents. 28 Those systems can offer a familiar macOS development environment and Apple silicon in relatively compact machines.

However, desktops are not the same thing as enterprise infrastructure. A dedicated server platform could be designed for rack deployment, dense multi-chip configurations, repeatable fleet management, and the support expectations of corporate or government buyers. The reported server effort suggests Apple may be testing whether demand for individual Apple-silicon machines can become demand for a supported data-center product.

That conclusion is an inference from the reported target market and design, not a confirmed Apple product strategy.

What the plan could mean for Apple and Nvidia

For Apple, a successful product could create a route from workstation-class Apple silicon into customer-operated AI deployments. It could also give teams that develop on Macs a potential production platform based on the same chip ecosystem.

For Nvidia, NVLink Fusion could broaden the company’s role in AI infrastructure beyond systems powered solely by Nvidia compute chips. If adopted, Nvidia would supply a key connectivity layer in a system built around Apple silicon.

Neither outcome is assured. They depend on a product that has not been publicly confirmed, an interconnect arrangement that has not been announced, and a launch window reported to be at least several years away.

The bottom line

The strongest conclusion is limited but significant: Apple is reportedly exploring an enterprise AI server with two or four future M8 Ultra chips, and it has discussed Nvidia’s NVLink Fusion as a possible way to connect them. 1311

Everything beyond that—including the final architecture, Nvidia’s role, pricing, availability, and whether customers will ever be able to buy it—remains provisional. Until Apple confirms the project, the 2029 date should be treated as a reported earliest target, not a launch promise. 57