THE MACBOOK PRO HOLDS MORE MODEL WEIGHTS THAN THE MAC STUDIO #3841
NathanTheThien
started this conversation in
General
Replies: 1 comment 1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Apple's desktop lost to Apple's own laptop, and nobody at Apple will tell you why.
I went to buy a Mac Studio to run AI locally under MLX. I had $2,500 to $4,000 in cash, ready to spend. There was nothing worth buying.
The Mac Studio tops out at 96GB of memory. The 16-inch MacBook Pro with the M5 Max ships with 128GB. The laptop holds more than the desktop. Read that again.
I closed the tab. Apple didn't lose that sale to NVIDIA. It lost it to itself.
And look at how it happened. This spring Apple quietly dropped the 128, 256, and 512GB versions of the Mac Studio, blaming the price and supply of memory. Then in June the Studio got more expensive anyway — base went from $1,999 to $2,499, the Ultra from about $3,999 to $5,299. More money, less memory, same three months. Meanwhile the MacBook Pro still ships with 128GB today, in stock. A shortage doesn't decide that the laptop should beat the desktop. A person decides that. Somebody picked which product would eat the shortage and picked the one their own developers use to run models. That's not a supply problem, it's a decision they made and never told anyone about.
But this is bigger than one machine I couldn't buy. The home server is where everything is going, and Apple is asleep on it.
Every family makes more data than they can hold and more than they want to rent. Photos, video, backups, documents. And now AI, which wants the model sitting on a drive you own, not streaming out of somebody's datacenter. And it has to follow you. I don't want a box that only works at home — I want it at home, my phone at the airport, my laptop in a hotel, all hitting the same drive over the internet with nobody in the middle taking a cut. Tailscale already does this. WireGuard does this. Synology sells it in a box. Companies a tiny fraction of Apple's size solved this years ago. Apple has every piece to build it and sells none of it.
Picture one box in the house. It runs a private AI on your own data. It holds the family's photos and backups. It runs your email. It blocks ads and trackers and scam sites for every device on the network before they ever load. It reaches your phone anywhere in the world, no subscription in the middle of any of it. That one box wipes out a stack of monthly bills. The VPN — gone, the box is the tunnel. The private email service — gone. The cloud AI subscription — gone, it runs on your hardware and nothing leaves the house. The cloud storage rent — gone. The ad blocker and DNS filter — gone, the box does it for everything, even the devices that can't run apps.
Proton already sells most of this as one bundle and people pay for it happily. Proton proved people want it, and it also showed the limit: it's still their server, and it's still a monthly bill. Apple could end the bill. It's the only company that could put the whole thing in one quiet box a family plugs in once and forgets, because it already owns the chips, the sign-in, the device trust, and the encryption. All of it is already in the building.
And look at the alternative. Today your choices are a stupidly expensive NVIDIA rig, or a PC you build yourself that most families can't build or keep running — no battery, no clean shutdown when the power cuts, nobody to call. Apple's chips are the only thing that could run all of this quietly, in a closet, on a battery, for ten years. Tesla didn't compete with the dealership repair shop, it killed the reason the repair shop existed — fewer parts, fewer breakdowns, fewer visits. The maintenance business shrank and Tesla didn't care, because it was never in that business. Same move here. Don't compete with the subscription. Kill the reason for it.
And here's the part that actually pisses me off. A computer doesn't have to die to become useless. Five years in it's just slow. Old. It still turns on, but everyone's moved past it. So what do you do? You throw out the whole machine and buy another $4,000 box — the drives, the case, the power supply, the battery, all still perfectly fine, in the trash, because the chip got old. Who wants to spend $4,000 every five years? That's stupid. Nobody signs up for that on purpose. And it's the exact thing nobody will fix, because the throwaway is the business. Let me swap the one part that aged. Just the compute. Leave the drives, the case, the battery, the power supply alone. That's the whole ask — a machine you feed for a decade instead of replacing whole every five years.
People will call this charity. It's not. It's a better business, and Apple won't want to hear that. Drive the whole stack down to one number and hold it — call it ten bucks a month. All of it: private AI on your own hardware, storage in your house, email, ad and tracker blocking, secure access from anywhere. One bill, one box, no tiers, no upsell when you hit 5GB. Raise it one percent twice a year, small and predictable and announced ahead of time. Nobody churns over that, and nobody feels farmed.
What makes that ten bucks honest instead of a trap is that you can leave. The data's on your own drives, and it all exports in one click, in a normal format, onto a drive you carry to any machine on earth. Cancel the subscription and you lose the convenience — the syncing, the follow-you-anywhere, the updates. You don't lose a single photo. You're paying for the service, not for a key to your own stuff. A subscription you can walk away from with everything is a product. One you can't is a toll booth. That's the whole difference.
Do the math from the customer's side. Right now a family pays for iCloud, a VPN, a private email service, an AI subscription, and a DNS filter — easily $40 to $60 a month, split across five companies, each holding a piece of your data. Apple could fold all of it into ten bucks, run it on hardware you already paid for, and still make money, because the cost of running it on your silicon in your closet is basically zero. Apple isn't paying for the compute. You already did. That's the highest-margin thing Apple could ever sell, and the only one where the customer says thank you. Right now Apple's privacy pitch is a promise — this turns it into a product you can hold: the company that ended the rent on your own data, whose box you plug in once and it just works for ten years and nobody sees your photos but you. Instead Apple charges $32.99 a month to hold photos it didn't take, on hardware you don't own, and calls it Services growth.
Which brings me to the thing nobody at Apple will say. iCloud storage is a joke as a product — 5GB free in 2026, on a phone that shoots 4K video. That's not a technical limit, it's bait. The free tier is built to fail so the $2.99 tier catches you, then $9.99, then $32.99, every month for the rest of your life, for data you already own. You're not buying storage. You're renting access to your own photos. And it traps you on purpose: no local copy you actually control, no drive you can hold, no array you can rebuild, no one-click way out that leaves you whole. That's the tell. A real product lets you leave with everything. A toll booth makes leaving cost you a weekend and a piece of what's yours.
That's how Apple dies. Not next quarter, not loud. Quiet. A whole generation grew up paying rent on their own memories, and someday somebody sells them a box that ends the rent, and it won't be Apple's box. Apple will have taught them the lesson and let someone else cash it. Trust is the one moat Apple has that money can't buy, and renting a family their own baby photos forever at a rising price is exactly how you burn it down without noticing.
So here's what I actually want. Not a better Mac Studio — one box that replaces four. To do what I need right now, I have to buy a Synology NAS ($600 plus drives, because Apple killed Time Capsule in 2018 and never replaced it), a UPS ($200, because no Apple desktop has a battery), a Mac Studio or MacBook Pro for the compute ($2,500 to $4,000), and a separate box for DNS. Four devices, four things to break, four things my family can't set up, won't set up, and shouldn't have to. Every one of them is a slice of one Apple chip with a battery and some drive bays.
The chip already exists and it's the best design in the industry for this. Put a battery inside so it shuts down clean when the power dies — every Apple laptop already does this, they make the part. Let me plug in any drives and have it build the backup array itself, the way Synology does for $600. Make it reachable from anywhere — phone at the airport, laptop in a hotel, same drive, no middleman — which Apple could do today because it already owns the sign-in and the device trust and the encryption. Ship the chip and its memory as one sealed cartridge I can swap in year four for a faster one while the case and drives and battery and power supply all stay. Sell it as a kit — I buy the battery, the power supply, the board, and my own drives, and upgrade each when I want. Build the case to last ten years and two or three cartridges. And let me take my data out any time, one click, normal format, onto a drive I can carry anywhere — the exit isn't a favor, it's the thing that makes it a product instead of a cage.
Somebody's going to say the memory has to be welded to the chip or the speed dies. Right, and that's not what I'm asking. I'm not asking to unsolder the memory. I'm asking for the chip and its memory to ship together as one sealed board I can swap. The distance between them doesn't change, the speed doesn't move, the only thing that comes out is the board. And don't tell me it can't be done — Apple already built swappable compute into the Mac Pro, designed the connector, shipped it, the know-how is in the building. Framework, a company a fraction of Apple's size with none of its advantages, sells "swap the board, keep the shell" as its whole business and turns a profit doing it. Somebody smaller than you is already doing it. You taught them how.
The reason nobody at Apple will say any of this out loud is simple: a box that holds every photo and backup forever, at home, with no monthly bill, eats iCloud. iCloud is Services, Services is Apple's highest-margin business, and Wall Street pays more for it than for hardware. Every family that buys this box stops paying $9.99 to $32.99 a month, forever. Apple already ran this play — Time Capsule, same product, built it and sold it and killed it in 2018 and walked away from the category. That wasn't an accident, they thought about it and quit. So say the real thing: Apple would rather rent me my own data than sell me the box that holds it. That holds up right until somebody offers the honest version — a box you own and a subscription you can quit without losing anything. The day that exists, renting people their own memories stops being a strategy and becomes the reason they leave. It's a fine business decision. It's not an engineering one. Stop pretending it's the second.
So here's what I'm asking, and I'll give it two weeks. Is dropping the 128GB-and-up Mac Studio temporary or permanent — one sentence, because people are spending $2,500 to $5,300 blind right now. Will the next Mac Studio hold at least as much memory as the MacBook Pro, or is a desktop that holds less than a laptop just the plan now — give a date or give a reason. Is running big models locally something Apple actually supports, or is MLX a toy and the real plan is to push inference to Private Cloud Compute — say it straight, because if the answer is the cloud I'll stop waiting and so will everyone reading this. And is there any plan at all for a swappable compute module on the desktop, any form, any timeline, yes or no — the Mac Pro proved the connector and Framework proved the demand.
Bottom line, I've got $2,500 to $4,000 set aside. It goes to Apple if there's a high-memory box with an upgrade path. It goes to a PC with a swappable card if there isn't. Not because I want the PC — because a ten-year plan needs a ceiling that goes up, and Apple's is going down. I'll keep buying the laptops, Apple wins that one outright and I won't pretend otherwise. But the server is the anchor. It decides where a family's whole life lives for ten years — photos, backups, the kids' files, all of it — and once it lands somewhere it never moves. Apple isn't fighting for that. PC and Android are taking it by default, not on merit, just because Apple didn't show up.
And to the MLX team, if you're reading: none of this is at you. MLX works. It works well enough that the hardware is the only thing left in the way, which is what a good project looks like — it makes the next wall visible and the next wall isn't yours. You didn't cut the 128GB config, somebody above you did, and I'd bet you argued against it and lost. I'm posting here because you're the people who get the constraint, and I'd rather this come from someone who wants Apple to win than from a comment section that wants it to lose. If Feedback Assistant is the right place, tell me and I'll file it there too.
I did enterprise IT in a health system for years, so I've bought infrastructure that had to last and had to work. I want Apple to build this. I'm telling you why I can't hand over the money yet. I'm saying this because my psych personality type is ENTJ, I am pragmatic and practical. I don't mind spending money, but not to buy more worries.
Beta Was this translation helpful? Give feedback.
All reactions