I am waiting as well. Got my money saved up and everything... just waiting for the announcement.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple Says Mac Studio and Mac Mini Will Be in Short Supply for Months
- Thread starter MacRumors
- Start date
- Sort by reaction score
Rychiar
macrumors 601
That would be insane. They can’t just flat out not sell desktops til next year. Companies are looking to buy them now
So they are gobbled up by AI people. I guess it's not important that ordinary people can have computers anymore.
Btw, why I am greeted by Gemini AI for my gmail? Can't think of a single thing I need AI getting close to my emails? What did Google conclude in their meetings?
Btw, why I am greeted by Gemini AI for my gmail? Can't think of a single thing I need AI getting close to my emails? What did Google conclude in their meetings?
I'm not manipulating anyone. I have good experience with all OS. This includes an M1 MBP, a powerful gaming PC and a Steam Deck that's Linux base. Each to their own and it's fine if you consider Windows to be resource efficient. In my experience it's not, whether it's for productivity, gaming or LLM inference.
nathan_reilly
macrumors 6502
wow, i bet its particularly bad in those scary "urban" settings as well, huh?
@Apple, its time for you to get into the RAM Fab / SSD Fab game.
Take your $160 bn in cash and buy Micron (market cap ~$565 bn)
Take your $160 bn in cash and buy Micron (market cap ~$565 bn)
- This buys you the fabs, the expertise, the IP/patents and gets you producing your products at whatever speed you need to meet demand
- This is a huge opportunity that is getting away from you (Apple)
- Imagine if Google, Microsoft, Amazon, OpenAI, Anthopic, Meta or Musk (TeraFab, Tesla, SpaceX, and xAI) succeeded in buying Micron
- The need for these components will only go up exponentially
DrJR
macrumors 6502a
Talk about a glass half empty mindset…..ask Lenovo if they would like to sell out of current product……selling one’s inventory is a GOOD thing.
Never said anything about not selling desktops - clearly there's going to be shortages - but there's no point trying to overextend the supply chain just to cater for desktops when there's more profit to be made from laptops
Refurbished minis were a dime a dozen in various configurations last fall when I got my M4 Pro, now they literally stay on the refurb store less than a an hour or two. Refurbished studios are gone even faster.
Meanwhile some folks on social media are flaunting their refurb/second-hand Studios with 256+ GB of memory, purchased at absolutely insane prices ($15-20k?) due to high demand - that's the new normal, apparently.
Crazy times.
Meanwhile some folks on social media are flaunting their refurb/second-hand Studios with 256+ GB of memory, purchased at absolutely insane prices ($15-20k?) due to high demand - that's the new normal, apparently.
Crazy times.
They still come up on the refurb store.
There’s a tracking service that will send you an email alert when your chosen configuration is available
That’s how I got my studio
Plus it’s cheaper then new anyway
I would guess very few
The m5 studio was always going to be in the fall at the earliest. If there even is one.
Maybe it will be spring m4 ultra/m5 max
How many normal people do you actually think are running LLMs at home?
What would they even do with them?
There’s a tracking service that will send you an email alert when your chosen configuration is available
That’s how I got my studio
Plus it’s cheaper then new anyway
I don’t buy this narrative for a second.
I would guess very few
The m5 studio was always going to be in the fall at the earliest. If there even is one.
Maybe it will be spring m4 ultra/m5 max
How many normal people do you actually think are running LLMs at home?
What would they even do with them?
BSDnostalgia
macrumors 6502a
There seem to be a lot of "AI people" on this thread. For those of us who are not, what is a good source of benchmark results that shows processor, OS, model, memory size, but, let's say, less detailed than this site that I found via Google? https://llm.aidatatools.com/results-macos.php A site that would explain, for example, why Mac Mini M4 systems are so in demand? The results, I "gather", are correlated with the memory bandwidth available to the inference processing hardware if it exists? Apple Silicon systems surely can't be the best, but, I guess they must be the most per system $ perhaps maybe ? Or per KWH consumed? That seems more likely. Apple Silicon is very power efficient. I know nothing about how this AI stuff works.
You haven't heard of OpenClaw yet. They do "AI things" with them.
You don't have to believe the AI rush because you're not anywhere on the bleeding edge of tech. It's real, and it's here. Heck, I got my 128GB studio just in time. It's an M1 Ultra, and now they're like 3x the price if you can find them.
If you're any good and in software you better be doing some kind of AI at home. And it's literally the newest toy for enthusiasts. A mini was a lot less than a 16/32/64 GB graphics card, and now they're gone.
FatLouie
macrumors 6502a
I would be very surprised to see the M5 Studio being released. People are already clamoring for the M6, and, in my view, Apple would be better off keeping the M4 Studio in the spotlight and then releasing the M6 Studio late next year. Many users want a Studio or a Mini, but only a few truly need a Studio.
I would think that not too damn many, not professionally and at home anyway. There are plenty of hobbiests doing this, though.
So the real problem is not the lack of memory supply, but their whole production line is inadequate to the actual demand.
FatLouie
macrumors 6502a
Both.
Apple gulped up a wad of memory without bickering about the price. But that is a finite quantity, and the memory crunch for the AI buildout the world is doing is not finished yet.
As for the SoC production line, who really knows? Perhaps it is that the economy and supply chain are in tatters now and will be for a while. It takes oil for the Asian manufacturers to make almost anything. And oil is a large cost issue now.
What “ai things”
Can you give some examples of the sorts of things people would be doing?
Doing what exactly? I can never seem to get an answer as to what people are doing with these LLMs
Yes I know what open claw is.
You don’t know where I am or why I believe what I believe
…
I’ve got an m4 studio right here. What sort of “AI stuff” should I have it doing?
If you want to see how tight the market is, look for a Mini M1 with 16GB on eBay. I got one for my wife for business stuff last year for $285. Now they're nowhere to be found or $600+. It's crazy.
16GB is pretty much the minimum for local LLMs. You could get away with a decent Qwen model at that point.
AI things = running openclaw to handle emails/routine tasks. There are people who have OpenClaw call them, make appointments, take dictation, etc. It's sort of weird to me.
But for enthusiasts, learning how to drive AI is more important. Every model has its issues. How do you drive it to get what you want? What parameters are there? There are all these agent harnesses and tools, and figuring out how to use them effectively is actually quite expensive if you're using the commercial services. It's a full-time job keeping up with what's out there.
All the stuff that you heard about AI being just a text generator are now wrong. It's like saying "a brain is just a bunch of electrical impulses" - it's true, but irrelevant. And the good thing is that right now it's easy enough to poke local models and see how they behave/do and modify them to do "illegal" things. They can write apps and games, sort of. They can actually do a pretty good legal analysis.
For local LLMs, I (and other people) are getting them because there's stuff you don't want to send to the cloud, and that security is worth a lot of money. Sure you can use bedrock too, but for many local is better/easier.
Plus agentic coding is pretty neat.
16GB is pretty much the minimum for local LLMs. You could get away with a decent Qwen model at that point.
AI things = running openclaw to handle emails/routine tasks. There are people who have OpenClaw call them, make appointments, take dictation, etc. It's sort of weird to me.
But for enthusiasts, learning how to drive AI is more important. Every model has its issues. How do you drive it to get what you want? What parameters are there? There are all these agent harnesses and tools, and figuring out how to use them effectively is actually quite expensive if you're using the commercial services. It's a full-time job keeping up with what's out there.
All the stuff that you heard about AI being just a text generator are now wrong. It's like saying "a brain is just a bunch of electrical impulses" - it's true, but irrelevant. And the good thing is that right now it's easy enough to poke local models and see how they behave/do and modify them to do "illegal" things. They can write apps and games, sort of. They can actually do a pretty good legal analysis.
For local LLMs, I (and other people) are getting them because there's stuff you don't want to send to the cloud, and that security is worth a lot of money. Sure you can use bedrock too, but for many local is better/easier.
Plus agentic coding is pretty neat.
asdfjkl;
macrumors 6502
Waiting on next generation Mac mini whether M5 or M6, whatever they release. Hoping prices stay reasonable as I do require 24GB / 1TB configuration. Given they already charge exorbitant prices for upgrades to the base model I hope those upgrade prices don't change. Really hoping for upgraded base specs, too.
