Apple weighs building its M8 Ultra servers around Nvidia's NVLink Fusion interconnect

Nvidia's licence lets a partner slot its own compute blades into existing Nvidia racks built on the MGX designs ^[7]. The rest of the rack stays Nvidia's to sell. Even when another company supplies the accelerator, Nvidia can sell the CPUs, the NVSwitch appliances and the networking, keeping business inside racks that run non-Nvidia compute chips ^[8]. Nvidia has not disclosed what the licence costs. "There will be some licensing elements in place," said Dion Harris, Nvidia's senior director of high-performance computing and AI infrastructure ^[9].

The M8 Ultra is at a very early stage of development, and a product tied to a chip that far from release can change or be cancelled ^[10]. The licensee list rests on dated announcements from other companies. MediaTek adopted NVLink Fusion on August 31, 2026, alongside Nvidia's $3.5 billion investment in MediaTek convertible bonds ^[11]. D-Matrix followed on September 10, 2026, and expects to offer systems with up to 144 Raptor accelerators connected by a single NVLink fabric by the end of 2027 ^[12]. AWS plans to use it with Trainium chips ^[13]. Seven companies are named in the program ^[18].

Apple has already shipped work onto Nvidia hardware. The relationship has been frosty for years, from a Steve Jobs-era dispute over defective parts and trade secret allegations through more than a decade of poor support for Nvidia hardware ^[14]. Then in June 2026 Apple extended Private Cloud Compute to Google Cloud using Nvidia Blackwell GPUs with Nvidia Confidential Computing, and its most capable server model, AFM 3 Cloud Pro, was optimized for Nvidia GPUs ^[15].

Apple also builds servers for itself. It has been shipping Private Cloud Compute hardware from a 250,000-square-foot Houston factory since October 2025, ahead of its 2026 target, and it refuses requests for outside access to that hardware ^[6]. The reported line would sell comparable machines to governments and AI developers ^[4].

The one dollar figure tying Apple to Nvidia systems is an estimate. Loop Capital analyst Ananda Baruah estimated in March 2025 that Apple was ordering about $1 billion of Nvidia GB300 NVL72 systems, which he put at roughly 250 servers priced at $3.7 million to $4 million each ^[16]. At those unit prices, 250 servers come to between $925 million and $1 billion ^[17]. Apple never confirmed the purchase, and the claim was questioned at the time ^[16].

Between the report and the earliest planned release sit at least three years ^[19]. John Ternus, Apple's new chief executive, has backed the server project since it began about a year ago, and Apple would have to secure expensive components in large quantities well before any launch ^[20].