Apple Intelligence Servers Expected to Get All-New, Turbocharged Chip

[MacRumors]

Apple Intelligence servers are currently powered by the M2 Ultra chip, and they are expected to start using M4 series chips next year. In an eventual move away from Mac chips for server use, The Information today reported that Apple is developing a new server chip that will offer even faster performance for AI features.

The report said the new chip contains "many duplicates" of Apple's Neural Engine, so it sounds like it will offer turbocharged performance for AI processing.

At least a trio of companies are believed to be involved with the chip. Apple is said to be handling the overall design of the chip, while Broadcom is said to be providing some networking technology for it. TSMC is expected to begin mass production of the chip in 2026, using its third-generation 3nm process, known as N3P. That is the same process expected to be used for the A19 chips in the iPhone 17 models next year.

The chip will likely power some Apple Intelligence features that rely on server-based generation, such as Image Playground, according to the report:

Apple will likely use the new AI chip it is developing for inference, where the chip processes new data—such as a user describing a possible image—and applies them to the models to generate an output, like producing the image itself.

More on Broadcom's involvement, from the report:

Like Google, Apple is relying on Broadcom for technology to network or link the chips together so they can work in unison to compute data more quickly. That technology has been one of the key drivers of AI development, making it possible to compute the massive amounts of data required to train and run LLMs. Networking technology is one of Broadcom's key strengths.

The paywalled report offers many more details about the chip.

Apple has announced that iOS 18.2, iPadOS 18.2, and macOS Sequoia 15.2 will be released today with additional Apple Intelligence features, including Genmoji, Image Playground, Image Wand, and ChatGPT integration for Siri. There is also a new Visual Intelligence feature on all iPhone 16 models that allows you to quickly identify things in the real world using the Camera Control button on those devices. As part of Apple Intelligence, Siri will gain features like on-screen awareness and deeper per-app controls, likely starting with iOS 18.4. And with iOS 19.4 in 2026, Siri is expected to become more conversational like ChatGPT.

Article Link: Apple Intelligence Servers Expected to Get All-New, Turbocharged Chip

 

[vegetassj4]

Hello Turbo! Apple introduces the new Turbo Apple Intelligence servers.​

CUPERTINO, CALIFORNIA -- Apple today announced a beautiful new server joins the lineup, with incredible battery life, a powerful dual-SOC system for pro-level serving and intelligencing, and groundbreaking selfie capabilities. People love their Apple Intelligence and rely on it every day for all that they do, and now there’s an exciting addition to the lineup with a new Turbo Server, lightyears ahead of the competition, it is the best server ever.

 

[Apple Knowledge Navigator]

M4 Ultra its own custom chip? Hopefully…

 

[dgdosen]

Pretty interesting - Apple has one of the largest and sweetest plums of market segments that will leverage this stuff... Even as 18.2 is a bit underwhelming.

 

[CarAnalogy]

Awesome! More server side power that my iPhone 13 Pro is too weak to access! Somehow...

 

[Tagbert]

Hmmm... new, more powerful server chips. Sounds a bit like the Intel Xeons that were used in the Mac Pro. If they brought these chips to their Mac line, it might finally make the Mac Pro worth buying again.

 

[Havalo]

This may not resonate with everyone, but it is unfortunate to see Apple has become the subject of ridicule this year.

Timmy, don’t be like Mike Lazaridis, listen to your customers…

 

[WarmWinterHat]

What, exactly, are these M2 pro chips doing that is soooo demanding that they are ready to be upgraded? I know it takes a beefy processor to say "hmm", "working on it" and "still on it", but I think the current setup can handle it.

 

[dannys1]

Hmmm... new, more powerful server chips. Sounds a bit like the Intel Xeons that were used in the Mac Pro. If they brought these chips to their Mac line, it might finally make the Mac Pro worth buying again.

I doubt it, it sounds more like specific Neural Engine based processors, possibly dropping the performance cores all together just to handle ML and generative AI. They're more focused not more powerful.

 

[dannys1]

What, exactly, are these M2 pro chips doing that is soooo demanding that they are ready to be upgraded? I know it takes a beefy processor to say "hmm", "working on it" and "still on it", but I think the current setup can handle it.

Processing generative AI for a possible 100m iPhone 15 and 16 users?

 

[WarmWinterHat]

Processing generative AI for a possible 100m iPhone 15 and 16 users?

Isn't a lot (majority) of that supposed to be done on-device? Genuine question since I don't use generative AI.

 

[dannys1]

Isn't a lot (majority) of that supposed to be done on-device? Genuine question since I don't use generative AI.

I've tested to see if the image generation is - but generally I think it's mostly the small LLM that is done on device which runs (or will run) Siri, understanding context, writing tools etc

 

[CWallace]

M4 Ultra its own custom chip? Hopefully…

I doubt we will see this chip used outside of Apple's data centers.

 

[everlast3434]

My iPhone 16 pro was apparently built for Apple Intelligence. Do I already have it? Am I still waiting for it? I havent noticed it, does that mean it's working? Thanks for making things clear Apple.

 

[neuropsychguy]

This may not resonate with everyone, but it is unfortunate to see Apple has become the subject of ridicule this year.

Timmy, don’t be like Mike Lazaridis, listen to your customers…

What? Who is ridiculing Apple? Who does Tim Cook need to listen to and why? Is there a problem with Apple developing powerful chips for a specific purpose?

 

[neuropsychguy]

My iPhone 16 pro was apparently built for Apple Intelligence. Do I already have it? Am I still waiting for it? I havent noticed it, does that mean it's working? Thanks for making things clear Apple.

Have you tried using any of it? Are you using the features Apple will add in the future?

I use the email and message summaries all the time. I sometimes use the writing tools to quickly clean up emails and messages and save myself time. I'm looking forward to the ChatGPT integration, image playground, and more. I'm also looking forward to Siri improvements.

Aside from all that, I've been enjoying my iPhone 16 Pro.

 

[everlast3434]

Have you tried using any of it? Are you using the features Apple will add in the future?

I use the email and message summaries all the time. I sometimes use the writing tools to quickly clean up emails and messages and save myself time. I'm looking forward to the ChatGPT integration, image playground, and more. I'm also looking forward to Siri improvements.

Aside from all that, I've been enjoying my iPhone 16 Pro.

Nope. I don’t need Apple the “clean up my writing”. Glad you find it useful though.

 

[CarAnalogy]

My iPhone 16 pro was apparently built for Apple Intelligence. Do I already have it? Am I still waiting for it? I havent noticed it, does that mean it's working? Thanks for making things clear Apple.

Are you able to fake your way through life like the commercials promise you can?

 

[eas]

These are probably going to be CPU & GPU light and ANE heavy compared to their existing laptop and desktop chips. They seem unlikely to find their way into a desktop Mac unless Apple decides to make something specifically for generative AI (which seems unlikely).

 

[jouster]

I doubt we will see this chip used outside of Apple's data centers.

Not disagreeing, as I have no direct experience of the semi-conductor industry. But isn't CPU design/build extremely expensive? So if there was a way to amortize the cost by putting it in a product like the Mac Pro, it would make sense to do so?

 

[CWallace]

Not disagreeing, as I have no direct experience of the semi-conductor industry. But isn't CPU design/build extremely expensive? So if there was a way to amortize the cost by putting it in a product like the Mac Pro, it would make sense to do so?

If it is designed primarily around accelerating Apple-specific Large Language Models, that could limit its suitability to general computing tasks under macOS.

And Apple is probably paying a small fortunate to third-party OEMs now for the LLM-focused servers in their data centers, so moving that in-house should provide significant cost-savings and eliminate the need to use it in consumer/prosumer/professional products for general sale.

 

[eas]

Not disagreeing, as I have no direct experience of the semi-conductor industry. But isn't CPU design/build extremely expensive? So if there was a way to amortize the cost by putting it in a product like the Mac Pro, it would make sense to do so?

Yeah, designing and manufacturing another chip variant it's really expensive, and engineering is all about making smart compromises.

If this chip doesn't have multiple uses then one wonders why they didn't just do deals with their existing cloud computing partners. I'm sure they would have been able to get good deals from Amazon, Microsoft, and Google for use of their NVIDIA-free inference offerings. It's conceivable that this ends up as a dual-purpose chip that ends up in a high-end desktop as well as Apple's AI infrastructure.

On the other hand "dead silicon" (un or underused functional units) is expensive. On top of that, chips like the Max are near the reticle limit, a limit on the size of chips that can be reasonably manufactured. So, the cost of the underused silicon kind of becomes infinite because there is no way to add more functionality to the die at any cost (ignoring that Cerebras has techniques for building devices that span an entire wafer). This is a per-unit cost rather than the fixed cost of design.

How many units of this AI chip are they going to need? There are a couple of billion active Apple devices in the world right now. Only a minor fraction can currently use Apple intelligence, but in a few years, that'll be approaching 1 billion. That could easily translate into a need for hundreds of thousands, maybe even a million of these chips. I wouldn't be surprised if that's approaching the # of M4 Max's they produce.

 

[junesix]

Yeah, designing and manufacturing another chip variant it's really expensive, and engineering is all about making smart compromises.

If this chip doesn't have multiple uses then one wonders why they didn't just do deals with their existing cloud computing partners. I'm sure they would have been able to get good deals from Amazon, Microsoft, and Google for use of their NVIDIA-free inference offerings. It's conceivable that this ends up as a dual-purpose chip that ends up in a high-end desktop as well as Apple's AI infrastructure.

On the other hand "dead silicon" (un or underused functional units) is expensive. On top of that, chips like the Max are near the reticle limit, a limit on the size of chips that can be reasonably manufactured. So, the cost of the underused silicon kind of becomes infinite because there is no way to add more functionality to the die at any cost (ignoring that Cerebras has techniques for building devices that span an entire wafer). This is a per-unit cost rather than the fixed cost of design.

How many units of this AI chip are they going to need? There are a couple of billion active Apple devices in the world right now. Only a minor fraction can currently use Apple intelligence, but in a few years, that'll be approaching 1 billion. That could easily translate into a need for hundreds of thousands, maybe even a million of these chips. I wouldn't be surprised if that's approaching the # of M4 Max's they produce.

Apple Intelligence will be passing off more intensive tasks that can't be handled on-device to ChatGPT. Instead, they could develop their own cloud service and roll it up into the Apple One subscription.

 

[huge_apple_fangirl]

Hopefully Apple sells a server or Mac Pro with this chip.

 

[svish]

Not surprised. With better chips being developed every year, expecting the best ones to power these servers.