A part of me is wondering if this is also a sign of major shake ups at the US plant where Minis, Pros, and Studios are assembled. I suspect that Apple has been repurposing that facility to build their Siri servers, and low sales/pending refresh of the desktop computers prompted them to dedicate that full space to make as many servers as possible to have up and running by WWDC 🤔
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.
Base Mac Mini Sold Out From Apple Online Store
- Thread starter MacRumors
- Start date
- Sort by reaction score
I still don't get the point of running a local LLM. And, with the security holes supposedly in OpenClaw, why would you?
TheMacDaddy1
macrumors 6502a
Refresh no doubt. Joe Consumer is not running local LLM's.
TheMacDaddy1
macrumors 6502a
The number of People buying Mac Mini's to run local LLM's is probably below 1%.
TheMacDaddy1
macrumors 6502a
No one is running local LLM's on 16gig Mac Mini's. Yes AI customers were buying lots of the 512 gig Mac Studios. 64gig is the starting point to run a local LLM IMHO but 128gig is much better.
CATiNTHESTORE
Suspended
Rychiar
macrumors 601
You can run competitive llms on 48GB+ Mac Mini🥰. Obviously 128GB is ideal if you can get your hands on one and have the money. You don't have to use OpenClaw. The reason why more people will do are the following:
OpenAI and Anthropic will IPO this year. The financials need to look healthy. Up until January 2026, inference had been highly subsidised and discounted. Claude Code for $200/month provided $3000-6000 worth of API usage.
This leads to cash burn that's to taper and the way it's done is by reducing subsidy/discounts. Before $200/plan would last.....a month. Now you hit daily and weekly quotas in 3-4 hours. The solution is to either spend $3000-6000 per month or for that money start buying Mac Minis.
I will be running Qwencoder 32b parameter model. I will basically run it 24/7 and send it instructions from my phone, iPad and laptop. Also it will work proactively for me like an EA on lots of things. In fact, the Mac mini won't be connected to a screen.
Basically the future for heavy users is API pricing. Openai and anthropic will offer compute to the highest bidders. Especially as they themselves struggle to secure more compute as many countries/states are increasingly against more data centres being built. Buying more GPUs from nvidia is easy whilst building more power plants is NOT!
Hence why Elon Musk is looking at building data centres in space!
Agent007
macrumors 6502a
TheMacDaddy1
macrumors 6502a
Less than 1% of 1% of 1% of "computer users" worldwide are using local LLM's. You do, that is great, but it is causing you major bias on this subject. NO ONE is buying a base Mac Mini to run a local LLM.
I recently read that 85% of AI usage is on free accounts, and that is all web based. The subject of the article is how the major AI companies will or will not reach ROI. No average consumer of AI, even knows what a LLM, local or otherwise is. They buy base Mac Mini's as a cheap computer that is not running Windows, and probably live in a web browser on that Mac Mini to do their "computing" needs....email, banking, looking up stuff etc.
My company uses Claud, Copilot, and Gemini, all paid accounts. We spend many thousands a month of AI, which has no become just part of the cost to run a business. We have taken down posts for developers because AI is making our developers more efficient, not replacing them. None of them ask to or want to run a local LLM. We have discussed setting up one in our local data center, to go against on-prem data sets but that has only been talk.
Don't consumers use their phone or a laptop as a computer?
Apple would not offer 128GB or 256GB or unified RAM and neural cores on the Mac mini for consumers to watch YouTube videos in 1080p. It's not just a computer but a highly space and power efficient server.
Label it whatever you want buy a Mac mini with high RAM is a bargain for me as a business owner. I replaced 5 software engineerings with cloud based LLMs and look forward to building out a farm of 3-5 Mac minis that will operate Task - code - test- iterate loop for several work streams.
I also love the fact that you can chain Mac minis. I will use Exo software for distributed inference across several Mac mini for LLMs that are too big to fit into the RAM of a single one.
Will get a nice rack like this one:
Then in 12-18 months something like this depending on stock availability:
I will continue to use Anthropic sparingly. Mainly for urgent bugs or just for planning with code execution being done by my Mac mini. I don't rly or use Claude Clode or Cowork as I don't want to become reliant on them, and I don't find Anthropic to be transparent about their products.
I feel stupid because I got one new for $399 in December. Wanted to run it as a local cluster with Exo with my other m4 mini I used as a Time Machine server at the time. It dissapointed me compared to my m4 max I had at the time so I returned it. Recently was looking for another $399 sale and saw I’d be lucky to even get one at MSRP lol oh well. Now I have a Neo and my other M4 mini as my main desktop alongside my gaming pc and 3090 Ti server.
CWallace
macrumors G5
And it looks like Apple is becoming that highest bidder based on reports.
Apple Is Reportedly Buying Up "All Available" Mobile DRAM At Very High Prices To Starve Out Its Competitors
Apple is not above playing dirty as it tries to corner the global mobile DRAM market to starve out competition.
wccftech.com
Apple Reportedly Agrees to 100% Price Hike on Samsung Memory Chips
Apple has agreed to pay Samsung twice as much for the LPDDR5X memory chips that it needs for ongoing iPhone 17 production, claims a report from Korean outlet Dealsite. According to the report's sources, Apple recently held emergency meetings with Samsung's semiconductor division to negotiate...
www.macrumors.com
This thread has been a JOY to read, especially whomever said this is due to Tim stepping down a couple of days ago. Incredible.
You can’t buy a Mac Mini or Mac Studio ANYWHERE - check Best Buy, BH Photo, Microcenter - they are sold out everywhere no matter what configuration.
boswald
macrumors 68030
And it looks like Apple is becoming that highest bidder based on reports.
Apple Is Reportedly Buying Up "All Available" Mobile DRAM At Very High Prices To Starve Out Its Competitors
Apple is not above playing dirty as it tries to corner the global mobile DRAM market to starve out competition.
Apple Reportedly Agrees to 100% Price Hike on Samsung Memory Chips
Apple has agreed to pay Samsung twice as much for the LPDDR5X memory chips that it needs for ongoing iPhone 17 production, claims a report from Korean outlet Dealsite. According to the report's sources, Apple recently held emergency meetings with Samsung's semiconductor division to negotiate...
those m5s everyone is waiting for are going to cost a fortune when they release this fall!
And how do you propose they sell you a physical product if they can’t get the parts to build it? I get what you are saying, but this pouncing method requires that they actually have the parts to build the product in the first place.
I'm really glad I got my m4 studio refurb when I did
I waffled on it a bit but is seeming more and more like the right choice every day
Yeah that’s true. I sold my 14” M4 max macboook pro about a month ago on marketplace for a couple grand and have the Neo to hold me over till I decide to (or heck even can) get something else. I used my max for just browsing Reddit and watching YouTube and tbh the Neo and my existing mini have kept me perfectly happy. I just play with llms as a hobby anyway and gaming stays in my gaming pc. I have an AI server I built with a 3090 Ti I had laying around from a. Couple years ago plus a 2080 so while only 32 gb of vram it’s still fun. I even used a couple raspberry pi’s I had in a drawer as docker servers where I can run my containers. All this stuff is just personal fun/learning anyway for me.
I did also get an iPad A16 and dare I say, I have used it more lately than either Mac just to browse on MR and Reddit and watch YouTube lol. Work provides me a beefy windows laptop for my development work, so as far as personal computers go, I guess I don’t really “need” a max again but will probably find myself gravitating to one again. I only got the max for my 3 display and the desire to replace my gaming pc. While crossover was great and I could play most single player games I threw at it, I still couldn’t play Valorant/arc raiders on the go at lan parties, so it stayed back anyway. I use GFN sometimes at lan parties now for arc raiders.
Full circle not sure if I really need anything else, just want it. I thought of getting an M4 max studio but realized I probably wouldn’t use it much anyway as it still has the same gaming limitations and it doesn’t have the display that the pro had for consumption.