Apple walked away from the server business more than a decade ago. Artificial intelligence may be giving it a reason to come back.
Apple is reportedly exploring a new generation of AI servers built around its own future M8 Ultra processors, with NVIDIA technology potentially playing an important role in connecting those chips together.
According to a report from The Information, subsequently reported by Reuters, the project has been under development for roughly a year and is tentatively aimed at 2029.
Nothing has been announced by Apple, and a project this far from release could change dramatically — or disappear altogether.
But the idea itself is significant.
Apple spent years turning Apple Silicon into one of its biggest competitive advantages. First came the iPhone and iPad chips. Then Apple pushed its own processors into the Mac.
The next destination could be the data centre.
Apple Silicon may be heading somewhere new
Apple’s current silicon strategy is overwhelmingly associated with devices people actually buy.
A-series chips power the iPhone. The M-series transformed the Mac and expanded into the iPad.
An M8 Ultra server would be something different.
Rather than putting the processor on someone’s desk, Apple would be deploying its silicon behind the scenes to provide the enormous amount of computing required by modern AI services.
The reported project is particularly interesting because it is said to focus primarily on AI inference rather than training giant foundation models.
Training is the enormously computationally intensive process that creates an AI model.
Inference is what happens afterwards — when millions of people actually use that model and expect an answer.
For Apple, that distinction makes sense.
The company doesn’t necessarily need to compete with OpenAI, Google or Meta to build the world’s biggest AI training cluster.
It does, however, need enormous computing capacity if Siri AI and increasingly sophisticated Apple Intelligence features are going to be used across hundreds of millions of devices.
And then there’s NVIDIA
Perhaps the most surprising part of the report is NVIDIA.
Apple is reportedly considering using NVIDIA’s NVLink Fusion technology alongside its own processors.
NVLink is designed to allow processors and accelerators to communicate at extremely high speeds — something that becomes crucial when large numbers of chips need to behave like a much bigger computing system.
That could allow Apple to continue designing the processors while using NVIDIA technology to solve part of the challenge of connecting them together.
It would also represent an interesting chapter in the relationship between the two companies.
NVIDIA graphics hardware once appeared inside Macs, but Apple and NVIDIA have had little meaningful hardware relationship for years.
Now AI could bring their technologies back into the same system.
Not inside a MacBook.
Inside Apple’s data centres.
Why would Apple build its own AI servers?
Control.
That has been one of the recurring themes of modern Apple.
The company designs the operating system, processor and much of the hardware inside an iPhone. The same strategy increasingly applies to Macs.
AI creates another layer Apple doesn’t completely control: the infrastructure required to run increasingly powerful models.
Building specialised Apple Silicon servers could change that.
Instead of designing only the device receiving an AI response, Apple could eventually design much of the hardware producing that response too.
Imagine the chain.
An iPhone running an Apple-designed processor sends a complex Siri request to an Apple data centre.
That request could eventually be processed by another Apple-designed chip sitting inside an Apple-designed AI server.
The answer then returns to the iPhone.
That would bring Apple’s vertical-integration philosophy into the AI era.
Apple already has an enormous AI infrastructure challenge
The timing matters because Siri has just undergone one of the biggest transformations in its history.
Apple released Siri AI this week alongside the latest generation of Apple Intelligence, bringing greater conversational ability, personal-context understanding, onscreen awareness and broader actions across apps.
Those capabilities make Apple’s products more useful.
They also increase the amount of AI infrastructure Apple needs behind them.
Some Apple Intelligence processing can happen directly on a device. More complex requests can require cloud computing.
As those capabilities become more ambitious, the amount of server capacity required to support them could grow considerably.
That makes Apple’s reported interest in purpose-built AI servers much easier to understand.
The M8 Ultra name tells us something too
There is another fascinating detail hidden inside the report.
M8 Ultra.
Apple’s Ultra processors traditionally represent the highest-performance end of its Mac silicon lineup.
Using an Ultra-class chip inside a server suggests Apple could increasingly blur the distinction between its personal-computing silicon and data-centre processors.
That doesn’t mean the reported M8 Ultra server chip would simply be a Mac processor placed inside a rack.
Apple has years to modify the architecture specifically for inference, networking, memory and data-centre efficiency.
But Apple’s existing silicon expertise gives it an unusually strong starting point.
NVIDIA doesn’t necessarily lose if Apple builds its own chips
At first glance, Apple designing AI processors might sound like another threat to NVIDIA.
The reality could be more complicated.
The AI industry is gradually moving beyond a world where every large system consists exclusively of NVIDIA GPUs.
Google has TPUs. Amazon has Trainium and Inferentia. Microsoft has developed its own AI accelerators. Meta is working on custom silicon.
Apple potentially joining that list would reinforce the trend.
But NVIDIA increasingly wants to provide technology surrounding those custom chips as well.
NVLink Fusion is part of that strategy.
If Apple ultimately uses NVIDIA’s interconnect technology to build an enormous cluster of Apple processors, NVIDIA could still benefit even if its GPUs aren’t doing the actual inference.
That’s a very different relationship from simply selling Apple graphics cards.
Apple’s AI strategy is becoming much bigger than the iPhone
For consumers, none of this means an Apple server is about to appear in an Apple Store.
The reported timeline points toward 2029, and Apple hasn’t confirmed the project.
But it provides a glimpse of how dramatically AI could reshape the company behind the products we see.
Apple Silicon started as something users rarely thought about.
Then it became one of the defining features of the Mac.
Its next evolution may happen somewhere consumers never see at all.
Inside enormous racks of computers quietly processing millions of Siri and Apple Intelligence requests.
If Apple ultimately builds those machines using its own M8 Ultra processors — connected with technology from NVIDIA — the company’s silicon revolution will have travelled from our pockets, to our desks, and finally into the data centre.
