Apple Caught Off Guard by AI Demand for Mac Mini and Mac Studio

[deconstruct60]

Apple reality picked a weird time to kill the Mac PRO.

Not really.

1. They likely don't have enough complete M5 Ultras to fullfill Studio demand. (BTO pre-orders for larger Studios is current at the end of November and hasn't shipped yet). How many more Mac Pros going to sell if don't have any SoCs to put into them to basically drive the system?

[ If Apple can incrementally sell more Mac Studios than they were selling Mac Pros 2023 then the number of Ultras being sold goes up. The system building costs go down. Which likely means profits are up. ]

2. The prices of the Mac Studio in the larger configurations was running into the Mac Pro price zone. The memory and NAND apocolypse impact on pricing would impact the Mac Pro to an even larger proportional extent. $10K Mac Studios means $14-5K Mac Pros.

Some folks think there is enough mainia buying that Apple could slap a $50-60K on the Mac Pro and they would still sell. But a product that is solely depending upon mainia buying isn't a stable product.

3. Since Apple is building their own semi-custom AI/PCC (private cloud compute) boxes elements of Mac Pro team could be tasked with desiging/testing those. They don't have to hire anyone new (or at least not that many ). Long update cycles inside of Apple are less critical than the Mac Pro iterating on 3-6 year cycles.

Until the PCC servers start generating direct revenue, controlling the costs there matters more than scoring points.

An optimal PCC server is not necessary a great Mac Pro foundation. AI datacenter SoC don't need 'graphics' output. There is no single user with a monitor connection as the nominal configuration. There are more high speed networking pressures than graphics ones. Dumping Thunderbolt and display engines for far better data center optimized networking would be a far better trade-off for die space usage.


4. It appears the 'Extreme' didn't work out (given Apple's 'chiplet' approach that wasn't surprising.) . There was very high overlap between the Ultra as the core for both the Studio and MP. Pretty good chance Apple was reserving part of the 1,400W power supply for an even high power consuming M-series option. Once that didn't show up.

Yes the Mac Pro could do more , but how many folks are in the non-intersection gap?

 

[gugy]

i want this bubble to burst so bad

Yep. As a veteran designer, it saddens me to see everything looking the same, the lack of creativity and surprise. Sure, AI can be very helpful for annoying tasks and speed up the process; however, when it becomes the norm, it just destroys what humans do so well.

 

[deconstruct60]

There used to be a rumour about an “M extreme” as the top tier above M Ultra.

Pre-AI that was probably not worth the effort for Apple, the Ultra was already overkill for 99%(?) of customers.

But now that Extreme would likely fly off the shelves, regardless of price.

Stratospheric demand at stratospheric high prices? Probably not. There are a ton of software gap around what Apple has. Apple also have a very flakey track record. Mac Pro 2013 ... 6 years until upgrade. Mac Pro 2019 ... four years until upgrade. Massively high price on vendor that is going to pull a Rip-van-Winkle and disappear for more than a couple of years? They would loose out customers who prefer to spend money on a timeline that they can see and is committed to.

The RAM Apocalypse has even bigger negative impacts for the Extreme than it has for the Ultra and less so for the Max. When minimum RAM size goes up , so does the prices.

Apple reported cancelled the Extreme in part because it was too expensive. The AI hype train wasn't going to 'fix' that. Even if did managed to get some FOMO, 'spend money like drunken sailors' to spend on it for one round... it really isn't a viable product. At some point the drunken sailor spending goes away and what really had was a 'fad' , not a product.

Apple's 'datacenter' chip probably won't look like an Extreme. So there really wasn't a 'data center' will save the Mac Pro path either. The data center chip needs more data center level I/O and less Thunderbolt and graphics output. It won't make for a great single user GUI workstation SoC. Modular RAM likely isn't coming either.

Most of the 'AI taxis' stuff is to better weather locations. Waymo is just getting to Detroit on initial ramp basis. The fact that all of the 'run away' from bad weather, potholes , and more challenging areas to drive in, is demonstrative that there is substantive hype around 'we are practically almost done' .

 

[deconstruct60]

Today, many people ride in cars or taxis that drive themselves using AI. In 10 years, the cars will be better, and in time, human drivers will be banned from public streets.

That's not going to happen. First, there is no way to economically deperceate the current automobile fleet completely off the streets in that amount of time.
When California (and others) bans new gas cars from being sold there is no way they can ban all gas cars in 10 year span. It won't work.

Second, there is nothing that has reached level 5. Level 5 has been just around the corner according to the hype train for years now. Until get to level 5, folks can't reasonably ban anything. Nobody is going to put up with 'sorry no abulance because the weather is too much for the AI to handle so can't transport you to the hospital' situations.

And folks in Rural area aren't going to give up their work vehicles used to visit/inspect/work on their fields/work areas any time soon. may need to drive into a field from time to time. there is no 'road' there, but some commonsense guidliness for what to and what not do . Lacking on common sense is all over current AI.

 

[burgman]

It never will. AI will take your nice indoor air-conditioned job shuffling paper and flipping bytes though. Corporate AIs will send compliance reports to the government's AIs with stunning speed and efficiency as long as efficiency is not measured by electric power consumption.

If your job could be done in a cubicle or at home AI and CEO’s have in their sights on you now. Some interesting reports on backlash because the numbers aren’t as rosy as AI companies raise prices

 

[germanbeer007]

See the Gene Simmons/Tom Selleck vehicle “Runaway.”

not going to sit through 99 minutes.

 

[turbineseaplane]

Yep. As a veteran designer, it saddens me to see everything looking the same, the lack of creativity and surprise. Sure, AI can be very helpful for annoying tasks and speed up the process; however, when it becomes the norm, it just destroys what humans do so well.

Agree. It's sucking the soul and joy out of everything.

 

[eifelbube]

Yeah.

Eurocomment.txt ("We need a an EU AISA/AIMA Act with lots of government control and no more greed and halt development until we know what's going on" or variants thereof.)

The world's innovation doesn't wait for the EU's regulations to catch up.

Yeah, the great innovators of tbe world Google, Meta, and cohorts. What is so innovative about addictive algorithm designs prompting adolescents to commit suicide? What so progressive about Google’s manipulation of search results to prefer your their own businesses? I am relieved that such “innovation” aka. innundation did not spring from European cultures. And if it had, regulation would have stopped or at least significantly reduced the damage that comes from convenience tech. Regulation stifles innovation, a myth. The real problem is that innovation stifles regulation. See more here or there.

 

[Flytrap21]

Yup and we were caught off guard by the massive price increase. From $599-$899 in a few months.

When you say you were caught off guard by the massive price increase in a few months, do you mean in 48 months (because that's how long the price has been $599)?

Mac mini entry level prices

• 2019: $799
• 2020: $699
• 2021: $699
• 2022: $699
• 2023: $599
• 2024: $599
• 2025: $599
• 2026: $899

 

[Wahlstrm]

Not really.

1. They likely don't have enough complete M5 Ultras to fullfill Studio demand. (BTO pre-orders for larger Studios is current at the end of November and hasn't shipped yet). How many more Mac Pros going to sell if don't have any SoCs to put into them to basically drive the system?

[ If Apple can incrementally sell more Mac Studios than they were selling Mac Pros 2023 then the number of Ultras being sold goes up. The system building costs go down. Which likely means profits are up. ]

2. The prices of the Mac Studio in the larger configurations was running into the Mac Pro price zone. The memory and NAND apocolypse impact on pricing would impact the Mac Pro to an even larger proportional extent. $10K Mac Studios means $14-5K Mac Pros.

Some folks think there is enough mainia buying that Apple could slap a $50-60K on the Mac Pro and they would still sell. But a product that is solely depending upon mainia buying isn't a stable product.

3. Since Apple is building their own semi-custom AI/PCC (private cloud compute) boxes elements of Mac Pro team could be tasked with desiging/testing those. They don't have to hire anyone new (or at least not that many ). Long update cycles inside of Apple are less critical than the Mac Pro iterating on 3-6 year cycles.

Until the PCC servers start generating direct revenue, controlling the costs there matters more than scoring points.

An optimal PCC server is not necessary a great Mac Pro foundation. AI datacenter SoC don't need 'graphics' output. There is no single user with a monitor connection as the nominal configuration. There are more high speed networking pressures than graphics ones. Dumping Thunderbolt and display engines for far better data center optimized networking would be a far better trade-off for die space usage.


4. It appears the 'Extreme' didn't work out (given Apple's 'chiplet' approach that wasn't surprising.) . There was very high overlap between the Ultra as the core for both the Studio and MP. Pretty good chance Apple was reserving part of the 1,400W power supply for an even high power consuming M-series option. Once that didn't show up.

Yes the Mac Pro could do more , but how many folks are in the non-intersection gap?

The ”mania” with AI is more about almost over night deciding to build 100 000 data centers all over the planet and in space..
(And mostly on credit and “I Owe You’s”).

Not so much people buying new computers.

The future of “AI”(LLM’s) is most likely in big part a future where you run this on your own hardware. So that part is less of a fad.

Apple wouldn’t need to produce the literal M extreme, but some kind of desktop Soc.

Since the Ultras they’ve made so far has all been built using laptop chips, just creating a pure desktop version of the Soc without the power and cooling constraints of the MacBook-line would likely not be too complicated and yield even more performance for a MacPro.

 

[carestudio]

I really think all were planed. No surprise. People started clustering Macs last year for local AI and even few years ago, Mac mini was popular for AI already. Anyway, as long as AAPL is doing good, I am happy.

 

[vantelimus]

Yep. As a veteran designer, it saddens me to see everything looking the same, the lack of creativity and surprise. Sure, AI can be very helpful for annoying tasks and speed up the process; however, when it becomes the norm, it just destroys what humans do so well.

Get used to it. That's here to stay. It was here long before AI. People copy one another. AI, as does any technology, only amplifies what humans value. Your discomfort isn't with the value. It is with the change in relative weights. When anyone with an idea can execute at the level of a craftsman, and craftsmanship floods the market, it just doesn't seem so special.

The overuse of AI is not what is meant by "the AI bubble". The term concerns the valuation of AI-related companies and investments. In finance, a bubble occurs when hype pushes the price of an asset above its true value. When reality catches up, the bubble pops and prices collapse.

 

[MacBH928]

Many have already addressed the obvious privacy aspect but there's also been a significant closing of the gap between local vs cloud AI skill. A year or two ago, the gap may've been 2-3 years in favor of cloud. Now that lead has been cut in half and will likely cut another 1/3 or 1/4 of the gap by years end. There's also no rails (for better or worse) on local.

Because you don’t want to divulge the whole of your life to big tech.

1. That $20K is cheaper over time

2. Privacy

I am all for privacy but $20K setup to keep privacy is a bit ridiculous for the average joe. It is also handicapped and runs in slow motion. There are solutions for privacy orients AI usage.

There are privacy solutions like Proton Lumo and Venice AI.

There are people who run through thousands of dollars worth of tokens in a day when they’re doing real work with these things.
Meanwhile, sharing data with all the greedy Tech Bros.
The Mac mini would pay for itself in a week.

yea, at internet dial-up speed

 

[mrhick01]

Glad that there's at least some people in this thread that understands what is going on and why. So many other posters in here seem to lack the understanding about how Apple accidentally created extremely good AI machines by accident and the AI base is turning towards Apple. It's not a "Apple didn't understand AI" thing, it's an "Apple's hardware turned out to be absolutely perfect for AI as opposed to Windows aimed off-the-shelf systems".

Watching the developer community swing the Titanic around like a teenage boy drifts a Subaru WRX. This sudden turning to Apple was unexpected by everyone, not just Apple. Apple do need a rack mounted system as that would absolutely kill, but not the "Pro". If anything this M6 development and its optimising for AI has informed Apple that the Pro architecture offered nothing to this wave of AI development. There's a reason Apple thumped an additional that Neural Engine into the M6 as they saw how it gave the M series chips the advantage with AI, paired with the unified memory structure that has made it so efficient for this role.

I believe that is why Ternus/Srouji will gear the M7 for those purposes, and why the M6 will not get higher SKUs for this generation.

Cook, before he departs, has locked in price stability (although at significantly higher prices) for memory acquisition until 2030. I speculate that Apple is limiting the M6 because the PC parts market has been so volatile the last 18 months.

Preparing for a full array of M7 on an improved 2nm die will allow them to build up their supply chains, and have a new lineup of desktops and tablets that will do all the things that Apple excels with, plus run localized LLM models well. Whether it's long-term planning genius, stumble-bum accidents of pure luck, or something in between, is less relevant.

This isn't anything that I know, but I follow LLM news and have been observing the trends, and I think this is a probability.

 

[deconstruct60]

I really think all were planed. No surprise. People started clustering Macs last year for local AI

Apple has been doing cluster demos for AI/ML since 2022 . It is closer to 4 years ago (at least ) at this point.

" ...
Apple did a WWDC demo of a 4 system Mac Studio cluster several years ago. ( WWDC 2022 ).

" Horovod uses a ring all-reduce approach. In this algorithm, each of N nodes communicates with two of its peers multiple times. Using this communication, the worker processes synchronize gradients before each iteration. I'll show this in action using four Mac Studios connected to each other with Thunderbolt cables. For this example, I will train ResNet, a classifier for images. "

Accelerate machine learning with Metal - WWDC22 - Videos - Apple Developer​

Discover how you can use Metal to accelerate your PyTorch model training on macOS. We'll take you through updates to TensorFlow training...

developer.apple.com

The they have been pushing this concept almost as long as there has been a Mac Studio. And technically similar things were possible earlier ( just not "ML" label on it. )

 

[deconstruct60]

So, the real question is...is the new M5 Ultra going to be impossible to get?

Impossible? No Preorders are showing 10-11 weeks

Shop Mac Studio

Get a new Mac Studio desktop with M5 Max or M5 Ultra from only $208.25 per month. A powerful platform for AI. Select a model or customize your own.

www.apple.com

Probably December, but possible. Just not timely. If these dates keep sliding back the 512-model probably will slide back more.

 

[deepfoo]

Apple's unusually timed announcement of new Mac mini and Mac Studio models this week was driven by unexpectedly strong enterprise appetite for AI hardware, according to The Information.

Apple normally releases new Mac models in the autumn, closer to October or November, making this week's announcement unusually early, falling just before the anticipated arrival of new iPhone models. The Information says that the AI-driven boom in Mac Studio and Mac mini sales is behind the early launch.

Apple noticeably promoted the ability to link multiple Mac Studios together into a single, more capable system for running large frontier AI models, a feature aimed at business and developer customers rather than everyday consumers.

Apple highlighted the Mac mini and Mac Studio's shift toward business buyers in June, with a "Business at the Park" event involving executives from major companies Ford, Disney, and Anthropic. The Mac mini was said to be the "darling" of the event.

Even so, enterprise's rush toward powerful desktop Macs more broadly took Apple by surprise. The company reportedly did not possess an engineering team dedicated to business customers or staff focused on developer relations, and lacked an enterprise AI strategy.

Businesses that approached Apple asking to buy access to the company's Private Cloud Compute infrastructure were reportedly turned down. Apple is instead leaning on partners such as WebAI and Mount Thor, which provide AI tools and execution environments built on Apple hardware.

A surge in demand for Mac hardware to run AI models has coincided with the global memory shortage, leaving many Mac mini and Mac Studio configurations out of stock for months. Some enterprise customers are reportedly turning to other hardware such as Nvidia's DGX Spark, a compact AI desktop launched late last year in a form factor similar to the Mac mini's, as Apple's own high end configurations remain difficult to get hold of.

Article Link: Apple Caught Off Guard by AI Demand for Mac Mini and Mac Studio

CUDA. Without that, I have no idea what you hope to do as far as AI/LLM goes. OpenClaw? I guess. But why? DGX Spark can run circles around these rigs for LLM application. As to the lower end M6 - why - why did they even bother to release it? 32gb is not even going to be baseline for graphics/video apps. 64gb closer. Written on a wheezing Gen 1 2020 M1 MacBook Pro. I will tape this thing together if I have to before laying out this kind of money for a new one.

 

[deconstruct60]

The ”mania” with AI is more about almost over night deciding to build 100 000 data centers all over the planet and in space..
(And mostly on credit and “I Owe You’s”).

Not so much people buying new computers.

No, the 'mania' is down at the personal computer buying level also.

OpenClaw 2.0 pours glitter on slow-burning security dumpster fire

Making installation easier and putting a new wrapper on the interface while leaving most of the security to users is a recipe for more trouble with the popular agent harness

www.theregister.com

Unit numbers are closer together . The price/computer is just different. And not going to get those 100,00 data centers actually operational because not enough power to do that.

The future of “AI”(LLM’s) is most likely in big part a future where you run this on your own hardware. So that part is less of a fad.

There is a deeply funding race to build the most biggest AI possible. So as personal computer sized hardware gets better then 'even bigger hog' will just grow larger to be too big to fit. ( The individual models probably have an upper cap of usefulness, but 'mixutre of experts' where can just have a very large group of experts to incur and even larger footprint.

That is different from 'good enough' models shrinking down to PC level. A major part of the investment is moving at those 'good enough' , free to download from the web models. [ Even the free-to-download ones have a motivator to draw folks back in. If not at some point folks don't start 'repaying' those folks training those free download models the work will stop. Nobody is going to spend infinite money into a black hole forever. At some point sevicess and revenue will be in the loop. ]

Apple wouldn’t need to produce the literal M extreme, but some kind of desktop Soc.

Since the Ultras they’ve made so far has all been built using laptop chips,

That has more to do with practical economics than missing out on performance. Apple's objectives are Pref/Watt. It is high perfrormance they are deploying. But also need to make the die affordable.

The Mac Pro completely decoupled from the rest of the Mac line up isn't economically viable for chip development. In the Intel era, the Mac Pro's chip was primary paid for by Intel server products that Apple primarily didn't sell. Intel's whole Xeon-Workstation (upper end) line up was die repacking of their server chips designs. The Mac Pro (and other Xeon-W systems) made contributions there , but the bulk 'paid for' was through the servers.

Decoupling from Xeon-W meant decouplijng from those paying the bills.

If the Max (and Pro) were paying the bills that who is could have to couple to. The MP volumes ( some low mulitle of 100,000) are enitrely not sufficient to be viable for a large chip(s).

just creating a pure desktop version of the Soc without the power and cooling constraints

Over and inefficient consumption of power is one of the principle reasons why those 100,00 datacenters aren't going into operation.

The Mac Pro class sytem hasn't been the core of the Mac desktop line up since before the Intel transition. The Mini and Studio are doing just fine. ( being sold out and backordered is pretty good).

The "laptop version" is beating the 'desktop' cores that the competition is putting onto the market.

" ...
Multi-core performance
SKU. Score. Difference

20-core RTX Spark. 23,126 Baseline

18-core RTX Spark. 21,776 -5.84%

Snapdragon X2 Elite (Extreme X2E-94-100). 25,075 +8.43%

Intel Core Ultra 9 275HX 20,783 -10.13%

AMD AI Max+ 395 20,609 -10.88%

Apple M3 Max (14-core) 22,385. -3.20%

Apple M4 Pro (14-core). 24,901 +7.68%

..."

https://www.tomshardware.com/pc-components/cpus/two-variants-of-nvidias-rtx-spark-show-up-on-geekbench-revealing-a-cut-down-18-core-model-full-20-core-beats-most-x86-mobile-chips-across-multi-core-and-single-core-tests

NOTE: how this August 2026 chart only wants to compare the M4 Pro to the lastest Windows competition when the M5 Pro/Max was out. Because the gap would be even more large.

Individual the CPU cores that Apple is using are just as a good as any 'desktop' cores. The GPU cores in M6 aren't 'slacking' either in term of per core performance. ( besides mismatches on a subset of AI compression encodings. FP4 , FP8 , )

of the MacBook-line would likely not be too complicated and yield even more performance for a MacPro.

Excessively overclocking what Apple builds isn't going to buy much in terms of performance. As pointed out above the cores in play are already fast. A bigger collection is more about whether the bigger collection can be done affordably or not. It isn't that Apple needs a different core design team for 'desktop'.

 

[EsTriFee]

Yup and we were caught off guard by the massive price increase. From $599-$899 in a few months.

As someone's that's still using a 2017 macbook pro 15 inch... I am quite sad i didn;t update in the last year 🙁

 

[apple4everOG]

When will this lunacy end.

When people stop wasting their time on meaningless things using AI

 

[Unregistered 4U]

Which laptop would that be? M5 Max already exists in MacBook Pros and there wont be non-base variants of the M6. The MB Ultra will have an M5 Max at best.

Oh yeah, the most performant mini is an M5 Pro with 15-core CPU with 5 super cores and 10 performance cores. The M5 MacBook Pro released in March has an 18-core CPU with 6 super cores and 12 performance cores that ALREADY beats that. Anyone wanting peak performance at any price already has a system that outperforms a mini. So this isn’t a “WOAH, people like AI, better ship something performant!” It’s just a refresh putting appropriately downscaled solutions inside a product that will sell far fewer units.

 

[seek3r]

“The Information makes up why they didn’t have the information about the new systems.”
We’ll see if analysts report an increase in Mac unit sales based on these two lines OR if they remain a low single digit percentage of hardware sales. If they sell the same “less than 30 million”, then no, they weren’t taken by surprise, they’re doing what they ALWAYS do when it’s time to refresh hardware, clearing the deck of the less important products so full attention can be focused on their more desired lines. To be sure, a laptop more performant than the Mac mini is on the way.

Remember, making up news about Apple isn’t a bug, it’s a feature. 🙂

FWIW that depends on whether they’re supply constrained. If they *cant* sell more than that 30M then it doesnt matter what demand is, the number isnt going budge

 

[Unregistered 4U]

FWIW that depends on whether they’re supply constrained. If they *cant* sell more than that 30M then it doesnt matter what demand is, the number isnt going budge

They’ve very rarely sold over 30 million even when not supply constrained. There’s just not as much demand for Macs. And companies buying “thousands” won’t affect that by much. Companies buying millions might, but I doubt companies are buying millions.

 

[gugy]

Get used to it. That's here to stay. It was here long before AI. People copy one another. AI, as does any technology, only amplifies what humans value. Your discomfort isn't with the value. It is with the change in relative weights. When anyone with an idea can execute at the level of a craftsman, and craftsmanship floods the market, it just doesn't seem so special.

The overuse of AI is not what is meant by "the AI bubble". The term concerns the valuation of AI-related companies and investments. In finance, a bubble occurs when hype pushes the price of an asset above its true value. When reality catches up, the bubble pops and prices collapse.

It's not a matter of getting used to. That's not the point. You probably are not a creative person to understand the main point. Plus nobody said anything about AI bubble. SMH

 

[Smartuser]

There used to be a rumour about an “M extreme” as the top tier above M Ultra.

Pre-AI that was probably not worth the effort for Apple, the Ultra was already overkill for 99%(?) of customers.

But now that Extreme would likely fly off the shelves, regardless of price.

Apple could also tier the RAM, like they do on most other devices.

MBA max out at 32, need more then buy a MBP. Mini max out at 64, need more then buy a studio etc. The Studio currently tops out at 256 with 512 being promised to come later.

Apple could keep 512 and perhaps 1024GB and 1536GB as an option for the Mac PRO.

For customers who’s workflow actually need these specs then the better cooling and perhaps the space to add a few dozen TB of storage for huge data sets or backups etc.

For some it would perhaps be nice to have the option of rack mount their “AI server”.

Redundancy in power supply, networking etc are also much appreciated amongst enterprise customers.

There’s plenty Apple could do with it.

It's effectively fan fiction. There are rack mount enclosures for the Mac Studio, example:

https://www.bhphotovideo.com/c/product/1727467-REG/sonnet_rack_std_2x_rackmac_studio_pro_3u.html/?ap=y&ap=y&smp=y&smp=y&store=420&smpm=ba_f2_lar&lsft=BI%3A514&gad_source=1&gad_campaignid=21407301024&gbraid=0AAAAAD7yMh2swtzIBm9vltv_mo1Dj9Ugr&gclid=Cj0KCQjwkt_UBhDMARIsALpnOAycM3lfefuLgmTe-lvGYw-imns0NIX0LcxA0kUE2OIwlLUgExu06i8aAmlSEALw_wcB

More exist.

Lots of things could be done. I was talking about the actual Mac Pro as it was built.