Apple may take “several months” to catch up to Mac mini and Studio demand
- Thread starter JournalBot
- Start date
Despite its all-in focus on Copilot, Microsoft continues to lose the software developer market to Apple (just look at the number of MacBooks that you see floating around Silicon Valley offices, or even within Microsoft itself). The Mac mini/studio are particularly nice for modern vibe-coders because they both have the hardware to plausibly support running local LLMs and because Apple's supply chains have somewhat insulated them from the global RAM shock.
Upvote
58
(59
/
-1)
I forsee a future where smugglers ship:
- Nvidia GPUs from the USA to China, and
- Consumer RAM from China to the USA
Upvote
24
(24
/
0)
When even Apple has component supply issues, you know shit has truly hit the fan, and it's likely several magnitudes even worse for everyone else.
Upvote
97
(97
/
0)
Despite, or because of? Microsoft is hell bent on bolting Copilot onto everything, even Notepad, where no one asked for it in search of a reason to claim usage statistical growth. Excel at this point in a brand new blank spreadsheet has 10+ intrusive ways of invoking Copilot even accidentally. Click on a blank cell? There a Copilot button popup!
And don't get me started on the shoddy QC for update of late.
Upvote
51
(52
/
-1)
ConnertheCat
Ars Centurion
As someone who wants to pick up a M5 Studio this year… god, I'm so fucking sick of AI.
Upvote
72
(74
/
-2)
Recently they've started pulling back on some of the dumber copilot integrations (like notepad) and other Windows apps. Still in early release builds but it's at least something.
https://www.cnet.com/tech/services-...-branding-from-windows-11-but-the-ai-remains/
Upvote
15
(15
/
0)
It is still mind-bending to me that with the demand TSMC experiences, no one is able to seriously challenge them. I realize that the technical and financial challenges are close to insurmountable, but still. The reward is enormous.
Upvote
55
(56
/
-1)
I've had two computers for the last 5 years. An M1 MBA for travel, and a high end desktop PC for my CAD work and heavy lifting.
I'm saying goodby to Windows and solving both problems with an M5 MBA. I know I'm not a big enough customer for MS to care or even notice, but Windows has become so enshitified I can no longer deal with it.
Upvote
59
(59
/
0)
MilanKraft
Ars Tribunus Angusticlavius
Yep. To the latter point, on the same earnings call Cook noted that starting in Q2 their memory costs went way up and will likely continue increasing. Will not be surprised if in 4-6 months, all Macs are $300-600 more expensive at any given BTO memory point, than they were just a few weeks ago.
Just another "benefit to humanity" from our glorious AI revolution, to go along with the "assisted" suicides, additional climate warming, impending water shortages, mental health issues, agent-deleted data, etc. So glorious I can hardly take it. It will even tell you how to cook pasta — this was never possible before now!!
(And to anyone about to say "but, but the ability to detect things in medical scans that humans cant," and similar things — that's called GOFML (Good Old Fashioned Machine Learning ("fake rebrands" not allowed)) and has been around for several years. Also does not require anywhere near the same resources the dipshit chatbots do.)
Last edited:
Upvote
48
(49
/
-1)
Agentic AI is bash/zsh native, and that's the kicker. Mac Mini's take the cake here because they're really the only consumer product available in a stationary form factor for Open Claw deployment. The truth is that you could run the same thing on a raspberry pi in many cases.
The fact that so many people are unaware of this reinforces my assumption that a lot of these minis and studios swept up for these projects will end up on the second hand market soon enough.
The fact that so many people are unaware of this reinforces my assumption that a lot of these minis and studios swept up for these projects will end up on the second hand market soon enough.
Last edited:
Upvote
15
(16
/
-1)
Upvote
4
(4
/
0)
I dumped my Windows system for only-gaming duties. First with an M2Pro Mac Mini, but I found 16GB of RAM not enough...and when I went to replace it with an m4, the higher RAM configs were all out of stock on the refurb store--and so have a Framework Desktop running Bazzite instead. Couldn't be happier--well except for HDMI being jagoffs and refusing to allow AMD Linux drivers to run the 2.1 spec because they are jagoffs.
I wish them luck, seriously.
They have a 10GB+ OS full of legacy garbage and half-finished half-baked features...and they kept bolting onto it, and patching on top of patching, for years. Someone on Ars in a prior MS story was of the opinion that MS basically needs an OSX moment, where they break legacy compatibility and start of fresh with a sane and maintainable and secure OS from ground up....because there's probably too much legacy garbage and patches to fix without it being basically a Jenga tower. And, I am prone to agree.
Upvote
24
(25
/
-1)
2026: the year it became impossible to buy a new computer so that the AI industry could corner the market on RAM, so that nvidia could make a bunch of GPUs, so that those GPUs could sit in a warehouse in Taiwan rapidly depreciating because the data centres they were supposed to be deployed to haven’t been built, and even if they had been built they couldn’t operate because the power stations also haven’t been built, so that the AI industry could lose money.
Upvote
58
(59
/
-1)
I just poked at the Apple Store website and apparently it's so bad they're not even letting you place an order and they'll ship it when they ship it. Disappointing since that was a legit differentiator from how how most other companies handle inventory shortages and was way more civilized than making you hover over the website waiting for the 5 nanoseconds it comes back in stock.
Upvote
4
(4
/
0)
Windows is even worse than that. Remember how you used to be able to have one screen to click on to select/deselect "run in background" for applications?
Yeah, that shit's gone.
And it seems every time windows updates, you have to do it all over again.
It's ridiculous. Whatever happened to an OS being an Operating system?
Upvote
6
(6
/
0)
Each fab line is like a 2KM long manufacturing line with bleeding edge everything and oh yeah, it's all in a clean room. New fab lines cost in the neighborhood of $50B in real capital. (And mind you that's $50B that it cost TSMC who for sure gets tax breaks and other incentives from Taiwan that nobody else is going to get
So unless you're Morris Chang's other cousin and you're opening your new fab in Taiwan, I am guessing you can stack another $5-10B on top. Who's going to invest $60B for a moving target?
Upvote
21
(21
/
0)
Just goes to show how hard it is to predict the future in tech, and how fast things can change. Could anyone have guessed that the Studio and Mini would suddenly become hot items?
The effects of the AI boom continue to ripple in the most curious ways.
The effects of the AI boom continue to ripple in the most curious ways.
Upvote
8
(8
/
0)
The limit is basically literally everything. ASML is only part of a huge picture.
In additional the hugely long manufacturing lines, they have lots of odd machines, some of them doing "that one thing" and oh yeah that machines is something from the 1990s and they basically bought 2000 of them that are stuck in a warehouse somewhere.
Upvote
3
(3
/
0)
And not only that, that is just the fab. You need packaging and so on. And all that stuff got built next door to each other in industrial centers in China and Taiwan. IIRC a couple of the new fab facilities in Airzona started producing silicon. Except they have to send product back to Taiwan/China for finishing because the packaging and so on is back there.
It is not dissimilar in concept to the Detroit Autos needing to send an engine they "build" back and forth to Canada a couple times. Because not all the steps in the production chain are domestic.
Upvote
10
(10
/
0)
The call seemed to hint that DRAM is the biggest issue given absolutely sky high prices, they probably just aren't getting shipments at all.
Upvote
2
(3
/
-1)
At one point there was a Symbolics Lisp Machine in many fab lines that were irreplicable and stockpiled.
Upvote
6
(6
/
0)
Yes, Tim is over-hyping the platform, as always.
"Amazing." Pfft.
Let's see how non-CEOs describe the platform wrt AI and agentic tools.
1 https://www.getopenclaw.ai/openclaw-mac
"Mac is the best platform for running OpenClaw. Apple Silicon runs it fast and silent 24/7. Mac Mini M4 is the go-to host."
2 https://dev.to/hex_agent/openclaw-on-macos-the-complete-setup-and-optimization-guide-igm
"macOS is the best platform for running OpenClaw."
3 https://openclawai.io/blog/openclaw-mac-mini-always-on-setup/
"The Mac Mini is quietly the best hardware for running OpenClaw. It's silent, sips power (under 10W idle on Apple Silicon), runs macOS natively (unlocking Shortcuts, iMessage, Apple Notes, Calendar), and fits on a shelf."
4 https://www.makeuseof.com/openclaw-introduction-and-review/
"OpenClaw is available on Windows, macOS, and Linux. But it works best on macOS."
5 https://techfundingnews.com/m4-mac-mini-clawdbot-ai-boom/
"Clawdbot takes full advantage of the Mac Mini's features: the M4 Neural Engine for local AI tasks, unified memory for smooth multitasking, quiet operation for nonstop use, and easy integration with Apple apps like iMessage and Siri."
6 https://moltbotwiki.org/install/mac-mini/
"The Mac Mini M4 is arguably the best hardware for running self-hosted AI agents like Moltbot. Its neural engine and unified memory architecture make it perfect for local inference and agentic workflows."
7 https://www.hashtechwave.com/moltbot-mac-mini-guide/
"While you can run Moltbot on a $50 Raspberry Pi or a $5/month VPS, the M4 Mac Mini has become the gold standard."
8 https://meincmagazine.com/gaming/2023...m-demo-shows-new-material-crafting-abilities/
"Fusing a long stick with a pitchfork can give you a longer attack range, for instance, and fusing various materials to arrows can create useful side-effects like freezing powers or a homing capability."
"Amazing." No one else would exaggerate to that degree.
Upvote
-13
(13
/
-26)
Exactly the kind of things governments are for, to fund those insurmountable things, where the return on investment isn't literal dollars and the timeline is generations. Unfortunately here in the US we don't currently have a functional government that wants to make smart investments.
Upvote
14
(15
/
-1)
Every OS will need to do that once people get serious about security. I'm betting that all OSes will need to be immutable and verifiable by hashes to access certain things within the next decade or so. We're headed into a zero trust future and there's really no other way. Apple kinda already does this with MacOS and Linux has some solid projects doing it (Fedora Silverblue, etc.).
Upvote
9
(9
/
0)
Man, I was really hoping for a 27" iMac to replace my 2019 machine. My other computer is a (rarely-used, but that's not its fault) 8GB M1 MBA. I know I'm going to have to replace them both over the next few years, but now I'm beginning to wonder if I shouldn't just go ahead and do it while I can.
There's some really cool stuff rumored to be coming down the line—like OLED MBPs, OLED iPad minis, and an iMac Pro; all of which I would consider throwing down the clams for—but now I'm wondering if they'll arrive before the total collapse of the world economy and the coming of the anti-Christ one-world government, lol(?)...
There's some really cool stuff rumored to be coming down the line—like OLED MBPs, OLED iPad minis, and an iMac Pro; all of which I would consider throwing down the clams for—but now I'm wondering if they'll arrive before the total collapse of the world economy and the coming of the anti-Christ one-world government, lol(?)...
Upvote
1
(1
/
0)
You are 107% correct. But let me add that things look even better for Apple in the local model and model fine tuning space than what your comment leads on.
Even with the loss of the 512GB RAM Studio model, the 256GB RAM model doubles what AMD (Ryzen AI Max+ 395) or nVIDIA GB10 can offer in a sigle machine. And in cluster, it gets even worse, as both AMD and Apple can go to 4x clustering, while the GB10 can only go to 2X.
Also, please bear in mind that Apple (and AMD) can instantiate 2 different local models (or two instances of the same model) without the models fighting for compute*, by having one model to use the GPU and another to use the NPU.
The GB10 can not do this, as it lacks an NPU, it only has its (admitedly massive) GPU available.
Since Microsoft can only use the AMD offerings (as it seems that their ARM exclusivity with Qualcomm is still ongoing), Microsoft's Local AI ambitions are hamstrung by the hardware partners.
* The two models will still fight for Electrical Power, Thermal Headroom, and memory bus Bandwidth, but will not fight for compute.
Upvote
8
(8
/
0)
Which isn't surprising. The TLDR of his statement: We have input-component stock and are burning through it as a buffer against increased prices. The same thing most companies did with panic-buying and warehousing lots of stock to delay the pain of the Trump tariffs.
Upvote
5
(5
/
0)
China does not produce enough RAM (or NAND Flash, for that matter) to feed the smuglers.
All the RAM (and NAND Flash) China produces can be legally absorbed by their internal markets + External markets friendly to them, where it is legal to sell it, at market prices, no shaddy deals, no illegal activity, no discount for the smuglers, no nothing.
I expect to see in the streets of Caracas (SO)DIMMs with Chinese CXMT memory chips, and M2 Drives with YMTC NAND Flash modules no problem... no smuggling.
Upvote
0
(0
/
0)
I also want to pick up an M5 studio. But I want it for running local models sooooooo….
Upvote
5
(5
/
0)
I just can't understand the popularity of OpenClaw. Mine doesn't do shit. I don't know what people are doing with it but there's just an insane amount of hype. Plus, when heavily subsidized plans go away (which seems like that's happening before the end of the summer) it's going to be ridiculously expensive to operate
Like before I knew you could plug your Github Copilot key into it for unlimited GPT-4o for $10/mo, I ate through $5 of Anthropic bucks in 15 minutes of basic setup and chatting, no continuously running workflows. That translates to, what, a hundred bucks in a month of normal usage?
So it's like wtf are people even doing with this thing? It can only connect to things that want you to connect, so very little ability to automate web tasks unless you're, again, paying for some API key.
I can't imagine shelling out $600 for a Mac Mini just for this, with API spend on top of that, but apparently people are doing it to such a degree that Apple is out of stock.
Like before I knew you could plug your Github Copilot key into it for unlimited GPT-4o for $10/mo, I ate through $5 of Anthropic bucks in 15 minutes of basic setup and chatting, no continuously running workflows. That translates to, what, a hundred bucks in a month of normal usage?
So it's like wtf are people even doing with this thing? It can only connect to things that want you to connect, so very little ability to automate web tasks unless you're, again, paying for some API key.
I can't imagine shelling out $600 for a Mac Mini just for this, with API spend on top of that, but apparently people are doing it to such a degree that Apple is out of stock.
Upvote
8
(9
/
-1)
Intel Foundry and Samsung Foundry are the only viable alternatives from the Lithography angle, and they are not too distant to TSMC really. The differences from a litho and performance point of view are small, and there is some movement in that direction with companies (particularly AI companies) using (or planning to use) their services in lieu of TSMC's.
But while TSMC lives or dies by being a good foundry, Samsung and Intel can always fall back to their internal usage, and therefore, are not as good foundries from the "customer care" angle (tooling for external customers, recipee variation for different types of logic chip on the same process node, willingness to really listen to and accomodate customer requests and sugestions, etc). That explains why TSMC services has such high demand.
As for new, or new-old entrants, is simple, each Fab cost ~U$D 10Milliard, and you need to add on top the R&D to develop your process. That's why there are not many entrants in the market, neither from companies that dropped out of the Race (Like Global foundries who abandoned a functioning 7nm and were stuck at 12nm) or completely new entrants.
The only new entrant in the market is Japan's Rapidus, aiming to bring 2nm in 2028, when the other three will already be in sub-1nm processes nodes, so, not bad, but not direct challenge either.
Upvote
6
(6
/
0)
macosandlinux
Ars Praefectus
For sure, Apple just underestimated demand and the AI bros buying up all TSMC can output gives them no options to alter orders. Cook, the supply-chain genius, kind of blamed himself "we just under-called the demand" and will leave his role as CEO on a miss here, which is kind of ironic.
Upvote
-4
(0
/
-4)
I'm waiting for a Fast and the Furious reboot where they steal RAM instead of DVD players.
My wife needed a Mac for grad school and her Surface just shit the bed. I was hoping it'd last a couple more months so I could buy the Mac over the summer but I had the choice of either a used M2 for around $600 or a brand new M5 for about $1000. We splurged and got the M5 to be safe, and it looks like that bet paid off and buying it early was a blessing in disguise. Things are going to get ugly very fast.
Last edited:
Upvote
7
(7
/
0)
My wife wants to sell hers M4 to replace it with a MacBook.
Didn’t realize that it would be an appreciating asset when we bought it.
Didn’t realize that it would be an appreciating asset when we bought it.
Upvote
5
(5
/
0)
Even when you've got governments with the capital and willingness to fund decades-long projects, there's no guarantee that they can manage the program successfully (i.e. California's high-speed rail boondoggle) or overcome the enormous technical challenges involved with leading-edge semiconductor manufacturing.
There's the question of whether ASML will even sell you the expensive machines you need and much of the know-how for the most advanced manufacturing nodes is either covered by thousands of patents or closely-guarded trade secrets. You need thousands of people with very specialized skills to operate the fab, too.
Intel had close to half a century of leading semiconductor fabrication and they basically blew up their technology lead with bad technological choices.
Upvote
2
(5
/
-3)
I mean, Microsoft kind of already had an OSX moment, and they had it around the same time that Apple did. Windows XP, the consumer OS, came from Windows NT and 2000, not 98 and Me, and that came with a drop in legacy support. Now, is the comparison exact? No, it isn't. But the other problem is that for a not-insignificant part of the user base, the legacy compatibility is the point.
Microsoft is hoist on the petard of its own success. Why would you stay with Windows if you lost backwards compatibility? Some people would, sure; Microsoft Aperture (or whatever, lol) would continue to find adoption in corporate environments...but even there, there would be many edge cases where users or whole departments would have to be left on legacy Windows.
Meanwhile, in the consumer marketplace, everyone who has been fantasizing (even if it was a sort of hate-fantasy) about switching to Linux or MacOS would suddenly have zero incentive not to switch. I did it half a decade ago. Do I regret it? Not in the slightest. I still use Windows, but not by choice, and not on hardware I paid for.
Upvote
12
(12
/
0)