Apple is developing an enterprise AI server built around its planned M8 Ultra chips and has discussed using Nvidia’s NVLink Fusion networking technology to connect them, according to a report by The Information. The product would mark Apple’s return to the dedicated server hardware market for the first time since it discontinued the Xserve line in 2011, and it targets the AI inference market rather than general-purpose computing. The server is not expected before 2029, and the project could still be canceled or launched without Nvidia’s technology. Apple shares reversed an early decline and rose modestly in premarket trading on the report, while Nvidia gained as much as 1.3 percent.
Two configurations on the table
According to people familiar with the matter cited by The Information, Apple is considering two versions of the server: a smaller configuration with two M8 Ultra chips and a larger one with four. The M8 Ultra is Apple’s next flagship processor, and the company has not announced a launch date for the chip itself. The servers would be sold to AI developers, other businesses and governments, a customer list that closely mirrors the original Xserve’s buyer base.
The connectivity question is where Nvidia enters. Apple’s internally developed chip-to-chip technology is considered too slow and expensive for large-scale deployments, sources told The Information. Nvidia’s NVLink Fusion is a suite of switches, chiplets and software designed to let processors from different vendors communicate at high speed inside data centers. Nvidia opened the technology to external licensees, and Amazon’s cloud division said last month it would expand its use of NVLink Fusion for next-generation AI servers. Scaling Apple’s existing connections between its own chips would present cost and speed problems, which is why the Nvidia option is under consideration at all.
Why Nvidia benefits either way
For Nvidia, an Apple deal would add a major customer to its networking business even as Apple builds chips that could compete with Nvidia’s AI processors. The company earns roughly a fifth of its data-center revenue from networking gear, and NVLink Fusion lets it monetize interconnect technology independent of GPU sales. An agreement would also give Nvidia a role in an Apple product that competes with its own AI systems, an unusual arrangement that reflects how central the interconnect has become to data-center design.
The talks also signal a thaw between two companies that have been at odds for nearly two decades. Their relationship broke down after widespread MacBook failures linked to Nvidia graphics chips, and Apple severed ties. More recently, the two have edged closer: Apple has announced plans to extend its Private Cloud Compute infrastructure to Google Cloud using Nvidia GPUs, and Nvidia has welcomed third-party chipmakers into its NVLink ecosystem. The Information’s report suggests those contacts have matured into concrete product discussions rather than general partnership talks.
Apple’s server history and current footprint
Apple last sold dedicated server hardware under the Xserve brand, a rack-mounted machine it discontinued in 2011 after shifting its enterprise focus to standard Mac hardware and consumer devices. The company has not been absent from the data center since then, but its presence has been internal. Apple builds servers for Private Cloud Compute, the system that handles AI workloads too demanding to run on iPhones and Macs, and it manufactures those machines at its own facility in Houston.
The difference with the reported project is distribution. Private Cloud Compute serves Apple’s own services, and the company has declined partners’ requests to use it. The new server would be sold to outside customers, putting Apple in direct competition with the vendors it currently buys from, including Dell, Supermicro and HPE, and indirectly with Nvidia’s own DGX line. Analysts have noted that Apple’s silicon strength lies in performance per watt, which matters for inference workloads where operating cost dominates the buying decision. A rack that draws less power for the same output is a real selling point when data centers are queuing for grid connections.
Market reaction and open questions
Neither company has publicly confirmed the talks, and both did not respond to Reuters requests for comment. Reuters could not independently verify the report. The 2029 target date sits two chip generations away, and Apple has canceled internal hardware projects before.
The open questions are significant. It remains unclear whether the server would run macOS-derived software or a different stack entirely, and who inside Apple owns the project. Apple has also just refreshed its Mac lineup with M6, M5 Pro, M5 Max and M5 Ultra chips, saying the 2-nanometer M6 delivers up to four times faster AI performance, which suggests the M8 generation is still early in development. A product this far out could look very different by the time it ships, or not ship at all.
What is clear is the strategic direction. Apple has watched AI inference become a market measured in tens of billions of dollars, dominated by rivals who sell both the chips and the systems around them. Nvidia’s data-center business alone now generates revenue that exceeds the GDP of mid-sized countries, and hyperscalers are designing their own chips to escape that dependency. Apple, which already designs the most profitable consumer chips in the industry, has both the silicon competence and the balance sheet to enter. A return to servers with its own silicon, and possibly Nvidia’s networking, would put the company back in a business it abandoned fifteen years ago, this time with a product category that barely existed when it left.
For now, the report is a rumor with named sources, and Apple projects at this stage die quietly all the time. But the direction of travel across the industry points the same way: every large consumer platform is trying to own more of the AI stack, from silicon to networking to the data center itself. Apple joining that race, fifteen years late, is the story worth watching.