Sources: Apple is developing AI servers that could use M8 Ultra chips connected via Nvidia's NVLink Fusion, with plans to sell them to developers and others
Apple has been working on plans for an enterprise server using its own chips that could incorporate networking equipment from Nvidia …
The Information
Context & Ripple Effects
Apple’s reported server effort began as an in-house deployment: in 2024, its AI features were said to rely on M2 Ultra server chips for more demanding tasks, alongside on-device processing. A later report tied its first dedicated server chip to Broadcom, while July coverage said Apple was exploring chip acquisitions amid reported upgrade and performance difficulties.
The reported plan to offer systems to outside developers would therefore extend Apple’s silicon work from supporting its own services into enterprise infrastructure. The possible use of Nvidia networking is notable because it would pair Apple compute with an external rack-scale interconnect rather than a wholly proprietary system.
First-order effects
If the reported plan proceeds, Apple would become a prospective supplier of AI servers to developers and other customers, rather than limiting its server chips to internal workloads.
Nvidia could gain a new customer for NVLink Fusion if Apple adopts the networking technology for systems built around its own M-series processors.
Second-order effects
Apple’s server design choices would make interoperability between its silicon and Nvidia’s interconnect a purchasing consideration for developers evaluating the reported platform.
If Apple commercializes the reported design, AI infrastructure could increasingly be assembled from specialized compute and networking layers rather than sourced as a single-vendor stack.
The potential combination of Apple processors and Nvidia fabric points toward heterogeneous AI systems in which chip vendors compete at the component level while partnering at the rack level.
The trend: AI infrastructure is moving toward heterogeneous rack-scale systems that combine proprietary compute with specialized networking.
I believe this could be the M7 Ultra (unless Apple is doing M8 Ultra simultaneously) - I wrote a few months ago Apple is working on a new AI server to compete with Nvidia built around M7 for 2029. There's also a new Apple AI server going live early 2027. https://www.bloomberg.com…
It was obvious Apple would get back into server, still think with dedicated custom ASIC. Many predicted this, including us years ago. The NVLink is interesting. I'd be super skeptical of that, but if it did happen it just emphasizes points we have made in our recent reports that …
Big Apple scoop. You watch these businesses long enough and they go in and out of businesses with the times. @aatilley @QianerLiu @_pheebini https://www.theinformation.com/ ...
🍎new: Apple, in an uncharacteristic move, has been working on a plan to SELL AI SERVERS to developers why? Apple stumbled into AI success wits its M-series chips for Macs, causing a run on Mac computers Apple talked to Nvidia about using NVLink networking for these servers
$AAPL Apple is considering an AI server built around its own future M8 Ultra chips and may use Nvidia's NVLink Fusion to connect the system at rack scale. The interesting part is what Apple would be outsourcing. Apple can design the compute silicon itself while borrowing an inter…
I joined Apple twenty-two years ago to support the Xserve and Xserve RAID - rack-mount servers and storage that Apple sold for a few years before abandoning the product line. What made servers not work for Apple then is still true today: - Apple refused to support older operating…
@twid As someone who deployed hundreds of xServes during my time at Apple, I do think they're better positioned now than they once were. I could see Apple winning a lot of govt bids on local AI projects in the datacenter. Your point about OS version support though is valid.
It's a shame Steve Jobs is dead or I could hear him say that Apple is humble to go into the server market for the first time *again* — (that was the launch of Xserv RAID, the second generation of their second go at servers) [embedded post]