Apple is developing an enterprise server built on its own M8 Ultra chips and has discussed using Nvidia's NVLink Fusion to connect them, The Information reported. The machine would be sold to AI developers, companies and governments, with nothing reaching the market before 2029. For business, this is the first sign that Apple intends to compete for AI infrastructure budgets it has so far left to Nvidia, Dell and Supermicro.
What the reported server contains
Two configurations are described. The smaller carries two of Apple's forthcoming M8 Ultra chips, the larger carries four. Nvidia's NVLink Fusion would link them, the same interconnect Nvidia already licenses to Qualcomm, Arm, Fujitsu, Marvell and others. The report does not name a price, a launch customer or a manufacturing partner. It also notes the effort could be cancelled, or proceed without Nvidia's technology at all. Shares in both Apple and Nvidia barely moved on the report, which suggests investors treat the plan as early and unconfirmed.
The hardware would not be Apple's first server. The company already builds machines for its own use: Private Cloud Compute ran only on Apple silicon until June, when Apple extended it to Google Cloud running on Nvidia GPUs and Intel processors. Selling servers to outside customers is a different business. Apple has not done it since 2011, when the Xserve was discontinued after nearly nine years on the market. That gap matters: the company would have to rebuild sales, support and channel relationships it abandoned more than a decade ago.
The timing is tied to European procurement rules. In June the European Commission proposed a Cloud and AI Development Act that would grade public cloud contracts by tiers of sovereignty assurance. The highest tier, intended for defence and national security, requires a European cybersecurity certificate and effective control over every software component; no third country may control a provider's design, development or maintenance. A locally owned and operated machine is an easier argument to make than a service run by somebody else.
What this means for business buyers
For companies that buy AI infrastructure, the practical effect is a third procurement option alongside Nvidia-based systems and cloud contracts. A server a government or a bank owns and runs in its own building fits sovereignty requirements more cleanly than a managed service, which is why the reported machine is aimed at governments as much as at AI developers. Small firms will see little change: the cost of owning and operating such hardware, plus the staff to run it, keeps cloud the default. Large buyers with compliance obligations are the natural audience.
Several conditions remain unverified. The Act covers cloud services rather than hardware, so a server sale would sit outside it altogether, though the procurement instinct behind the rules would not. Europe has already shown how flexibly it applies them: in April a six-year, EUR 180M sovereign cloud framework went to four groups, one a Proximus-led consortium including a Thales and Google venture, alongside Mistral AI. American infrastructure qualified, and the Commission said non-European technologies operated within a strict framework can meet the minimum level of sovereignty. Bidders were judged against eight objectives, from legal exposure to supply chain transparency. Buyers should ask vendors which tier their offer targets and who controls firmware and updates.
The marker to watch is whether Apple confirms the programme and names an interconnect partner before 2029. A formal launch with NVLink Fusion would put Apple directly against Nvidia in enterprise AI hardware and give European public buyers a sovereign-operable option built on American silicon. Silence, or a shift to Apple's own networking, would leave the market as it is today.
