Apple Eyes Nvidia NVLink Fusion for AI Server Return

Apple has discussed using Nvidia's NVLink Fusion networking technology in a planned AI inference server powered by its own chips, The Information reported on Wednesday, citing people familiar with the matter — a development that would mark Apple's first serious return to the enterprise server market since it killed the Xserve line fifteen years ago. Reuters carried the report but cautioned that it has not independently verified it, and both Apple and Nvidia declined to comment.

That caveat matters and we are putting it first: this is a single sourced report about a product that does not exist yet, may never exist, and is reportedly not expected before 2029. But the reported details are specific enough — and the strategic implications large enough — to be worth taking seriously.

File photo: an Apple Xserve cluster — Apple's last commercial server line, discontinued in 2011. (Photo: NASA via Wikimedia Commons, public domain)

What The Information reported

According to the report, Apple is evaluating Nvidia's NVLink Fusion technology for an AI inference server expected to use Apple's planned M8 Ultra processors. Secondary coverage of the paywalled report — notably a detailed AppleInsider write-up — says Apple has considered two configurations, with either two or four M8 Ultra chips, and that the system would target AI developers, other businesses, and governments rather than consumers.

Every one of those specifics should be treated as reported, not confirmed. The report also says the project is roughly a year old, that nothing is expected before 2029, and — critically — that the product could still be cancelled or redesigned without Nvidia technology. This is Apple thinking out loud about a category it abandoned, not Apple announcing a product.

Dion Harris, an Nvidia executive, declined to confirm whether Apple is an NVLink Fusion customer but said: "Apple has always been a great customer and partner of ours." That non-denial is the closest either company came to engaging with the story.

What is Nvidia NVLink Fusion?

NVLink Fusion is Nvidia's interconnect technology that lets third-party chips — not just Nvidia GPUs — plug into Nvidia's NVLink fabric, allowing custom CPUs and accelerators to share memory and bandwidth with Nvidia hardware at high speed.

For Apple, which designs its own silicon, the appeal is architectural: a reported AI inference server could pair an M8-class chip with Nvidia's networking ecosystem without Apple surrendering its chip-design independence. (Apple has not confirmed any of this — it remains a single unconfirmed report.)

Why NVLink Fusion matters

To understand why this would be a big deal, start with what NVLink Fusion actually is. Per Nvidia's official newsroom, it is a rack-scale platform that lets customers build semi-custom AI infrastructure inside the NVLink ecosystem — meaning a partner can integrate its own custom chips or accelerators into Nvidia's networking fabric rather than being locked into Nvidia's GPUs end to end. Verified partners already include Marvell (announced March 31, 2026), MediaTek (August 31, 2026), and d-Matrix (September 10, 2026).

For Apple, that architecture solves the central tension of its AI infrastructure strategy. Apple designs world-class silicon and has spent years insisting that on-device processing is its priority — but the largest AI models still need cloud-scale compute, and Apple's internal server infrastructure, Private Cloud Compute, was built for a different era. The Information reported separately in June 2026 that Apple is tapping Google's Nvidia Blackwell B200 fleet to power a Gemini-driven Siri, with Nvidia's confidential-compute feature encrypting data during processing. Apple's own in-house AI server chip effort, reportedly codenamed Baltra, was said in July 2026 to have hit delays.

NVLink Fusion would let Apple do what it has always wanted to do: keep its own silicon at the center while borrowing the industry's best interconnect to scale it. It is also a notable pivot given that Apple has sat on the board of UALink — the open interconnect standard positioning itself as a rival to NVLink — since January 2025. Nvidia CEO Jensen Huang has framed NVLink Fusion as precisely the kind of open-arms move that lets custom-silicon companies stay in the Nvidia orbit rather than building competing ecosystems.

Huawei is also gunning for Nvidia's AI chip crown.

Why inference servers are the prize

There is a reason the reported product is an inference server, not a training system. The AI industry's center of gravity is shifting from training models to running them: every agent, assistant, and copilot that businesses now deploy runs on inference, and serving that demand at scale has become one of the industry's key battlegrounds. Reuters' coverage of the AI-chip market this month noted that companies are racing to build specialized processors to challenge Nvidia's dominance precisely in inference — where the economics come down to memory capacity, memory bandwidth, and power, and where serving models "dramatically cheaper and more energy efficient" is the competitive edge.

That is a market Apple is structurally suited to enter. Inference serving rewards efficiency over raw throughput, and efficiency is Apple's oldest engineering habit. It also explains the Nvidia side of the reported equation: NVLink Fusion exists so that companies with their own silicon can scale inside Nvidia's networking fabric instead of building competing ecosystems. An Apple inference server running Apple's chips on Nvidia interconnect would be the ecosystem play working exactly as designed — for both sides.

Beyond compute, memory is the industry's other choke point.

Ghosts of Xserve

Apple has been here before, and it walked away. The Xserve — a 1U rack-mount server line — was announced on November 5, 2010, and Apple made it unavailable after January 31, 2011, exiting the rack-server business entirely. The company continued designing ever more powerful processors for Macs and its own internal workloads, but it left commercial data-centre servers to Intel, AMD, and Nvidia.

That history is what makes the report interesting rather than merely plausible. Apple is the most valuable company in the world, and it has spent fifteen years watching the server market from the sidelines while building one of the most formidable silicon design teams on the planet. An AI inference server would not be a nostalgic return to Xserve — it would be a bid to sell the picks and shovels of the AI boom, powered by the one thing Apple has that almost nobody else does: its own chips, tuned for its own software stack.

And the timing is not accidental. The AI infrastructure build-out has turned data centres into the most contested real estate in technology — the same build-out we have been tracking from the memory-chip wars to the sovereign data-centre land grabs. Apple, which just weeks ago unveiled the iPhone Duo foldable Apple's accelerating AI hardware ambitions, is visibly accelerating its AI ambitions across hardware, software, and now — reportedly — infrastructure.

What Apple's server comeback would mean

Keep the skepticism close. The Information's report is the only source; Reuters could not independently verify it; the product is years away if it exists at all; and Apple cancels projects as a matter of routine. Nothing in the report says Apple has committed to Nvidia, or that Nvidia has committed to Apple.

But consider what the discussions themselves reveal. Apple is thinking seriously enough about enterprise AI infrastructure to evaluate Nvidia's interconnect for its own silicon — which means the company's AI ambitions extend well beyond putting a smarter Siri in millions of iPhones. For Nvidia, an Apple endorsement of NVLink Fusion would be a strategic coup: the ecosystem play, validated by the one customer every vendor wants. For the US enterprise market, an Apple-built AI server would land in a world already crowded with Nvidia, AMD, and a wave of custom silicon — and fifteen years after Xserve, Apple would be betting that its vertical integration is finally an advantage in the data centre, not just on the desk.

Apple NVLink Fusion server FAQ

Did Apple confirm the report? No. It is a single report from The Information via Reuters, and Reuters has explicitly not independently verified it.

What chip would Apple use? Reported (not confirmed): an M8 Ultra-class part in 2- or 4-chip configurations.

When would it ship? Reported (not confirmed): before 2029.

Why is Apple returning to servers? AI inference demand and its Private Cloud Compute stack; Apple exited the server market when it discontinued Xserve in 2011, fifteen years ago.

Sources: The Information via Reuters (Sept. 16, 2026 — Reuters explicitly notes it has not independently verified the report); AppleInsider and business-news-today.com coverage of the report (Sept. 16, 2026); Nvidia official newsroom (NVLink Fusion partner announcements, Mar.–Sept. 2026); Macworld/Engadget/AppleInsider (Xserve 2010–2011 history); The Information via MacRumors (June 2026 — Apple/Google Blackwell B200 reporting). All reported details on the Apple server project are explicitly framed as reported-not-confirmed. The Xserve cluster photo depicts Apple's discontinued server hardware and is captioned as a file photo.