Apple Reportedly Building M2 Ultra and M4-Powered AI Servers

[MacRumors]

Apple has started building its own AI servers that use the M2 Ultra chip, Haitong analyst Jeff Pu reports.

Pu provided the analysis based on supply chain checks in a new note to investors seen by MacRumors. Foxconn is said to currently be assembling Apple AI servers that contain the M2 Ultra, with plans to assemble AI servers powered by the M4 chip in late 2025. Last month, a reputable source on Weibo said that Apple was working on processors for its own AI servers made with TSMC's 3nm process, targeting mass production by the second half of 2025, which may line up with this report about M4-powered AI servers.

Apple's decision to build its own AI server is reflective of the company's ongoing strategy to vertically integrate its supply chain. Apple will likely use its own chips to enhance the performance of its data centers and future AI tools that rely on the cloud.

While Apple is rumored to be prioritizing on-device processing for many of its upcoming AI tools, it is inevitable that some operations will have to occur in the cloud. By the time M4-based servers could be operational in data centers, Apple's new AI strategy should be well underway.

In line with broader industry expectations, Pu also expects Apple to announce on-device AI features at WWDC in June, such as audio recording summarization, photo and video editing, and live translation. While the iPhone 16 lineup will tout a range of AI features, 2025's iPhone 17 models will apparently be more focused AI devices.

Article Link: Apple Reportedly Building M2 Ultra and M4-Powered AI Servers

 

[CWallace]

Makes sense for them to leverage their own hardware, especially if their AI implementation is designed around their neural engine and GPU designs as opposed to "commodity / off-the-shelf" hardware.

And I do not expect Apple to offer these servers at retail, so this is not the "return of the Xserve".

 

[tonywalker23]

"Siri, directions to 123 example street" Here's directions to 123 example street [and instead of an address in the town I live, I get directions to somewhere halfway across the country.

 

[TallGuyGT]

If they're using M2 Ultras, I hope that means they're saving the M4 Ultras for a Mac Studio / Pro release at WWDC 24 (and not WWDC 2025).

 

[Think|Different]

Cannot wait to see what they do in this space! WWDC is gonna be a good one.

 

[m.x]

Makes sense for them to leverage their own hardware, especially if their AI implementation is designed around their neural engine and GPU designs as opposed to "commodity / off-the-shelf" hardware.

And I do not expect Apple to offer these servers at retail, so this is not the "return of the Xserve".

I guess a nice side effect is lower cost + higher yield of their Ultra chips due to the economics of scale. I don't think they have been that high in demand, well, now they are.

 

[bradman83]

If they're using M2 Ultras, I hope that means they're saving the M4 Ultras for a Mac Studio / Pro release at WWDC 24 (and not WWDC 2025).

The second-gen 5nm process the M2 series is built on is substantially cheaper than the first-gen 3nm process the M3 series is built on, while the Neural Engine in the M3 series is only 15% faster than the M2 (versus 60% faster than the M1). So Apple can build data centers with M2 Ultra chips for substantially less than an M3 chip without sacrificing much in the way of performance.

And if Apple is preparing to release an M3 Ultra at WWDC they'd likely want the chip supply dedicated to those for-sale units.

 

[gank41]

Would be amazing if Apple started offering a newer version of their Xserve lineup again.

 

[CrysisDeu]

Also means they will need to build and release software stacks for ML training on Mac, which is good

 

[anshuvorty]

Would be amazing if Apple started offering a newer version of their Xserve lineup again.

Ummm...why? there is a reason why Xserve failed....Apple's proprietary server tech just wasn't popular or performant enough for the market, hence why it failed and why Apple decided to end it.

I would imagine that AS-equipped servers, if Apple decided to sell them again, would end up with the same fate. Especially with regards to AI generation and Nvidia taking a huge lead with that too...

 

[anakin44011]

If this is true, I'm guessing they are just using up the remaining M2 Ultra stock that was manufactured. I can't imagine they are selling very many of these after the initial round of Mac Studios and Mac Pros...especially if the benchmarks on the M3 Max are true.

Anyway, I'm still rooting for an M4 Ultra sooner rather than later...and an option for multiple M4 Ultra daughtercards/modules on the Mac Pro for a personal/private render farm. I don't really need one for myself...but I'm still of the mind that Apple needs a more impressive flagship desktop computer.

I'm probably so wrong on this...but I'll keep hoping nevertheless.😜

 

[gank41]

Ummm...why? there is a reason why Xserve failed....Apple's proprietary server tech just wasn't popular or performant enough for the market, hence why it failed and why Apple decided to end it.

I would imagine that AS-equipped servers, if Apple decided to sell them again, would end up with the same fate. Especially with regards to AI generation and Nvidia taking a huge lead with that too...

Why? Because maybe people want something in a smaller OU than the Mac Pro for their rack?

 

[257Loner]

There is only one reason Apple is making their own AI servers: to avoid paying Nvidia billions of dollars!

 

[usersince86]

I asked Siri, "Should Apple build its own AI servers?"

Screen shot of Siri's "response..." 😁

 

[tkaravou]

It makes sense for Apple since they always do their own thing. They're not the type of company to buy stuff from a high margin competitor (NVidia) when they can keep the high margins to themselves. They're also not invested in CUDA afaik so they don't have the software dependency either.

 

[sunny5]

It sounds like Apple is making their own server with RTX 3060 graphic cards.

 

[Antony Newman]

Using up all the remaining & unused M2 Ultra sounds like a smart move - even as a testing ground before Apples data centre chips make an appearance.

If Apple needs a disaggregated supercomputer from their own silicon - It must have been a crazy cost (and lead time) if they had opted to use Nvidia.

Could the Hidra actually be the results of their Data Centre chip research? An architecture that can scale properly (just like the 'ARM' chips used in Fugaku)?

In the next two years - Apple will also have at their fingertips:
- wafer scale fabrication (courtesy of Cerebras R&D)
- silicon photonics (low latency more distant memory)
- TSMC tile technologies (to move to follow AMD's effective use of chiplets)
- Gate all around transistors - that can allow for '3D' design of layouts

.. I wonder if Apple are partly responsible for the dearth of memory chips in 2025?

 

[coolfactor]

They've proven the advantage of controlling the hardware and software for consumer devices. Bring that to servers where iCloud exists, and server-side Siri and AI processing all happen and we might see them deliver an experience unlike anything available from other cloud providers. Time will tell.

 

[macfacts]

What's the point of AI servers if apple users have on device AI?

 

[DanVM]

They've proven the advantage of controlling the hardware and software for consumer devices. Bring that to servers where iCloud exists, and server-side Siri and AI processing all happen and we might see them deliver an experience unlike anything available from other cloud providers. Time will tell.

Major cloud providers like MS, Google and Amazon already have their own AI processors running on custom-built software. Also, MS and Google already have AI integrated into their services, like MS 365 Copilot and Google Gemini in Workspace. Apple is the one behind, so we'll have to wait and see how they respond.

 

[tkaravou]

What's the point of AI servers if apple users have on device AI?

For GPT, they need something called a LLM (Large Language Model). Think of this as a massive snapshot of the internet at a given point in time. Once that snapshot is generated, they can copy it locally to the user. So when you ask it something, it checks the local LLM (on device) and gets you the answer.

So what's the point of the AI server?

1. Generating those models takes a LOT of GPU power for a long period of time. This is what all these companies are paying the big bucks for.

2. Those snapshots get stale quickly, so they need to run a new model every so often. It can cost billions to generate a new model so you don't want to do it too often.

3. As they improved their code, they need to regenerate parts or all of those models.

So does that mean if the snapshot is from Jan 1 I won't get any recent data?

Yes.

To fix that, a lot of the GPTs will leverage the internet to get more up-to-date results. That's slower and the results aren't as good. This part will not be handled on-device obviously.

This is why Apple is trying to strike a deal with Gemini and OpenAI. It's also possible that Apple is looking to handle the fresh data on their own, which would require more servers.

 

[itti]

Huang: 😭

 

[MichaelDroste]

Very interesting 🤔 how much ram? I’ve seen rumors of half a terabyte? When is the m4 ultra going to hit?

 

[kemal]

I can't imagine Apple's laptop chip comparing favorably to what Nvidia is offering. Too bad Apple hates Nvidia.

 

[JosephAW]

Hopefully this will benefit macOS