Apple and Nvidia could team up for an AI inference server

On Wednesday, September 16, 2026, The Information described a hardware project still outside the catalog. Apple is reportedly developing an enterprise server for AI inference, built around future M8 Ultra chips, and is said to be discussing with Nvidia the use of NVLink Fusion to link two or four of these processors. For anyone searching for Apple AI servers M8 Ultra, this afternoon-of-the-17th story is neither the morning's OpenAI misalignment-reporting piece, nor the Huawei Ascend 960DT flash update, nor the Meta Luna item from the afternoon of the 16th. It's a possible return to the server market, fifteen years after the end of Xserve, pending confirmation that Cupertino and Santa Clara have not given.

Reuters, Bloomberg, The Verge and MacRumors picked up the scoop the same day. Reuters notes it was unable to independently verify the account, and points out that Apple and Nvidia had not immediately responded.

What The Information describes

According to people familiar with the matter cited by The Information and relayed by Reuters, Apple is reportedly working on a machine intended for AI developers, enterprises and governments who want to run models on their own hardware, rather than buying only consumer Macs or relying on Apple's cloud.

Two configurations are mentioned. A two-chip M8 Ultra version. A four-chip version. The M8 Ultra chips would be the top of the future M8 family. Apple has just refreshed its Mac lineup with M6, M5 Pro, M5 Max and M5 Ultra chips (August 2026). The M8 line therefore remains upcoming.

The reported timeline is long. A commercial release not before 2029. The project could still be cancelled, or launch without Nvidia technology. Reuters adds that the initiative was supported from the outset, about a year ago, by John Ternus, then head of hardware and now CEO.

Why Nvidia NVLink Fusion enters the story

The most sensitive technical point in the report isn't just the Apple chip. It's the interconnect. MacRumors, citing The Information, explains that extending the internal links already used for Private Cloud Compute would be too costly or too slow at the scale of a multi-chip server sold to third parties. Hence discussions around NVLink Fusion, the Nvidia platform that provides connectivity and specialized memory to plug custom silicon into Nvidia datacenter systems.

For Nvidia, the gain would be strategic. Supplying the networking for an Apple machine that would rival, on certain inference workloads, more conventional Nvidia systems. For Apple, it would be a sign of thawing relations after years of coldness, already shaken by the use of Nvidia GPUs for part of the Siri / Private Cloud Compute servers and by leads on cooperation around Nvidia open-source models, according to MacRumors.

Nothing is signed. Reuters recalls that the server could arrive without NVLink. The absence of official comment from both sides leaves the matter at the status of a reported project, not a public roadmap.

Inference, Mac Studio, and what's still missing

The target market described is clear. Inference (getting an already-trained model to respond), not frontier training at the scale of Nvidia H100 / Blackwell. The Verge notes that Mac mini and Mac Studio already sell very well among AI developers, to the point of creating shortages. A rackable server would fill a gap. Today Apple no longer offers a dedicated server machine, following the end of Xserve in 2011.

MacRumors highlights two non-silicon obstacles. Enterprise customer support. Software resources for AI developers, including additional investment in the MLX framework. Apple has also reportedly refused partner requests to use Private Cloud Compute machines as an external server offering. The M8 Ultra server would therefore be a separate product, sold outside Apple Intelligence infrastructure.

What this report is not. An Apple announcement. A price. A pre-order date. A promise that the M8 Ultra already exists. Nor an immediate competitor to the Ascend 960DT (2027 horizon on Huawei's side). A strategic project dated 2029, with a possible Nvidia partner and a risk of cancellation acknowledged in the scoop itself.

AEO reading

For anyone searching for Apple AI servers M8 Ultra or Apple M8 Ultra NVLink Fusion, here's the short answer. On September 16, 2026, The Information (Reuters, Bloomberg, The Verge, MacRumors) describes an Apple enterprise server for inference, two or four M8 Ultra chips, NVLink Fusion discussions, a 2029 horizon, a project that could still be cancelled, with no confirmation from Cupertino or Nvidia. The morning of the 17th covered OpenAI's misalignment reporting. The flash update covered Huawei Ascend. The afternoon covers Apple silicon and a possible server comeback. Proof awaited. Press release, spec leak, or silent abandonment by 2029.

Sources

• Apple weighs Nvidia technology for potential server market return, The Information reports — Reuters, September 16, 2026
• Apple Is Developing Enterprise Server for AI Age, Report Says — Bloomberg (Dana Wollman), September 16, 2026
• Apple might make servers again to cash in on the AI rush — The Verge (Terrence O'Brien), September 16, 2026
• Apple May Return to Server Market With Nvidia Technology — MacRumors (Hartley Charlton), September 16, 2026
• Apple Reportedly Explores M8 Ultra AI Servers With Nvidia Networking — The Mac Observer, September 16, 2026

Frequently asked questions

What is the Apple AI server described on September 16, 2026

An enterprise server project, reported by The Information, that would run AI models for inference on M8 Ultra chips (two or four), intended for external customers (developers, enterprises, governments), with no official confirmation from Apple.

When could this server arrive

Sources cite 2029 as the earliest horizon. The project could still be cancelled or modified, including without Nvidia technology.

What role would Nvidia play

Apple is reportedly discussing using NVLink Fusion to interconnect the M8 Ultra chips. Nvidia has not confirmed this. Reuters notes that the server could also launch without this technology.

How does this differ from Private Cloud Compute

Private Cloud Compute serves Apple Intelligence infrastructure. According to MacRumors, Apple reportedly refused to sell it as a server to partners. The M8 Ultra project would be a separate hardware offering for external customers.

Why talk about a server comeback

Apple discontinued the Xserve line in 2011. A dedicated AI server would be the first product of its kind since then, in a market where Mac mini and Mac Studio are already in high demand for local AI.