Mac Studio 512GB RAM Option Disappears Amid Global DRAM Shortage

[johnvale]

We are not talking about regular RAM here. This unified memory on apples chips is more comparable to vRAM on GPU’s. This demand for high memory is mainly due to running local AI. Regular RAM is just isn’t fast enough. Flagship GPUs have even higher bandwidth but they don’t ship with that much of vRAM. Apple with its M chips is uniquely positioned in the middle - it’s fast enough and you can have loads of it. M ultra chips is a hot buy for local inference.

All leading AI labs are offering their models at a loss, so inevitably they will jack up the prices once everyone is hooked. Meanwhile local models are getting only better. So there could be a point of inflection where companies just run some jacked up local hardware instead of paying ever increasing subscriptions. This is not investment advice, but it’s the reason I have been loading up as much Apple stock lately. If they release M5 ultra with one 1TB RAM and manage to keep up with the demand, even if it’ll cost over 20K it will be flying of the shelves (as it will still be much cheaper than comparable stuff from Nvidia).

 

[jole]

This is Apple's halo car. Insanity to stop selling it. Ok to do a price correction — all buyers understand the reason.

Apple should instead release M5 Ultra with 1TB RAM. Does not matter how much it costs — $20k would be much better than non-available. It is a significant AI narrative Apple needs.

 

[vantelimus]

Could be the 512GB stock is all going in to Apple datacenters.

Or the 512GB stock has been sold and they don't want to manufacture more Studios with that configuration since a replacement with M5-Ultra is what they are currently producing for release in the next few months.

Honestly, I want at least a 1TB option in the next Studio. I hate it when MacOS pauses a bunch of tasks and tells me it is out of system memory. Yes, it happens even on 512GB machines when you open large AI models.

 

[Apple MessagePad]

Folks here who were never gonna buy this machine nor with 512GB of RAM, complaining about the lack of the option for something that was never going to impact them in any way.

 

[noraa]

If you think AI is a bubble you have no idea. AI has escaped from the labs and is happening so fast that nobody can keep track of everything. There are hundreds of thousands of really smart people building all kinds of stuff with nothing but thought. Is it production ready? Not yet...but it will be. Is it good enough? Probably, yes.

At this point the main thing holding back AI is QA test cases. I'm sure at some point someone will come up with a mechanism for AI to do a black box stare-and-compare kind of QA, if they haven't already.

The fact is, AI is probably better at any given thing than 80% of the humans out there. Could it, say, come up with a better, more cost effective and lower-impact way of making steel? Maybe. Could you? Probably not.

The problem with AI has nothing to do with its capabilities (or lack thereof), but the economics of it. Right now companies are pouring billions and billions into AI (mostly In the construction of data centers, but also R&D), but not getting any money out of it. Google, OpenAI, Anthropic, Microsoft, etc. are desperately trying to monetize runs, but thus far have failed to do so. Eventually all of these companies are going to realize that there isn’t enough money in AI and drastically scale things back - at which point the bubble will burst.

AI isn’t just going to go away at this point, but sooner than later the hype train around it will fade and fall off a cliff.

 

[SeaFox]

Nobody offers upgradable GPU memory

Nice red herring. We're talking about general system RAM. The fact it's used for graphics too does not make it "GPU memory". Using your definition, Apple used to offer it, as I own older beige Macs that used system RAM for graphics (and that RAM was on SIMMs I could replace/upgrade myself).

 

[vantelimus]

Not for AI work. Large Language Models can be really, really large. Even a terabyte of RAM isn’t enough to run the biggest open models at full quality.

Quantization is your only friend at this point. A full Deepseek-R1 requires 5-bit quantization to fit on a 512GB Mac. Lately, I've been favoring Qwen3.5-397B-A17B-MLX-8bit for image analysis. It requires 394.17GB. It can be slow, but it is thorough. If you want fast and interactive, I've been finding NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-8bit (32GB in memory) to be very zippy.

 

[ThailandToo]

I noticed this as I wanted to add another and I thought it meant perhaps Apple would update the Mac Studio now. I guess what I have must be worth more money now? Surely the pricing of the memory on the M5 Ultra will be much higher.

 

[vantelimus]

If you think AI is a bubble you have no idea

AI funding is in a bubble. AI itself is here to stay. If we don't kill ourselves or let AI do the job for us, this will be a very different world in five to ten years.

 

[Corefile]

Billy Gates said Macs only needed 640K maximum and so I think Billy is good with this.

 

[Orange Bat]

The problem with AI has nothing to do with its capabilities (or lack thereof), but the economics of it. Right now companies are pouring billions and billions into AI (mostly In the construction of data centers, but also R&D), but not getting any money out of it. Google, OpenAI, Anthropic, Microsoft, etc. are desperately trying to monetize runs, but thus far have failed to do so. Eventually all of these companies are going to realize that there isn’t enough money in AI and drastically scale things back - at which point the bubble will burst.

AI isn’t just going to go away at this point, but sooner than later the hype train around it will fade and fall off a cliff.

This is exactly it. While AI is here to stay, people are trying to figure out how to make money with it. Many who have implemented AI in their business are starting to realize they aren’t saving money nor increasing productivity. In fact. Productivity is dropping in some cases. Companies are dreaming of replacing large swath of workers with AI but are finding they need the same number of employees, or even more, and then pay thousands for AI on top of it. Implementation of AI has been really rushed and is a unicorn that promises to solve all problems, but instead those going all in are guinea pigs for an expensive experiment.

 

[CWallace]

My guess is that production of the modules needed for 512GB M3 Ultra configurations are no longer readily available or presuming an M5 Ultra launches at WWDC in three months, the wholesale price of the module has risen to the point that it is not worth ordering and warehousing new product as it would not sell-through. I could see any existing product being held for warranty coverage.

This is Apple's halo car. Insanity to stop selling it. Ok to do a price correction — all buyers understand the reason.

Apple should instead release M5 Ultra with 1TB RAM. Does not matter how much it costs — $20k would be much better than non-available. It is a significant AI narrative Apple needs.

Trick is, Apple has to be able to ship the product. If they no longer are an important enough customer to secure the memory modules needed for 512GB+ configurations, there is no point in offering them.

 

[1105]

AI funding is in a bubble. AI itself is here to stay. If we don't kill ourselves or let AI do the job for us, this will be a very different world in five to ten years.

Exactly, and it's up to everyone to adapt or get left way behind.

 

[jole]

Trick is, Apple has to be able to ship the product. If they no longer are an important enough customer to secure the memory modules needed for 512GB+ configurations, there is no point in offering them.

That would be devastating. But if that is the case, how can they ship lesser memory configurations at all — is it a question of chip size? Is this a speculation or a fact? (source?)

 

[macduke]

AI needs to crawl up its own ass and die. While I'm terrified for what the inevitable AI bubble burst will mean for the economy (and for the thousands that will lose their jobs), it really can't happen soon enough.

Yes you are right it needs to burst. I don't care what happens to the economy in the short-term, because long-term millions will lose their jobs to AI. Hundreds of thousands already have, and many graduates can't find entry level work.

And for what, so we can drive up electronics prices, electricity prices, pollute and warm the planet a lot more for our children? Make college irrelevant so we have an under-educated population that just listens to and accepts whatever bias Sam Altman, Elon Musk, etc put into their AI? This madness must stop. Greed is out of control.

Society is going to collapse because the change is too fast. We have no social safety nets for these people who will lose their jobs by the millions. What few safety nets we do have are being defunded as we speak. We have no system in place to retrain them for skilled labor. And then what, the robots with AI brains take those jobs too? My daughter graduates from college around 2036. My son around 2039. If college is even relevant by then. They are so screwed! I am so angry.

In theory AI should usher in this great utopia. in reality the world is controlled by pedo billionaires and soon trillionaires with unchecked greed. They will ensure that this technology is bent to their will and serves them above all else. Their hope is that this breaks our education system and keeps people reliant on them. They can more easily manipulate a dumb, dying population. I used to think this was fringe conspiracy think, but the signs are all around us. They want to create feudal states for their tech empires and rule over us like kings.

I have to use AI for my job. It's required to keep our proposals competitive with other companies, unfortunately. It can do really cool things, and sometimes it can be incredibly dumb and screw something pretty bad. I would love for it to usher in a Star Trek like utopia, but I also know how mankind works. We won't let that happen until we can let go of the mindset that for someone to win, someone else has to lose.

 

[JSRinUK]

Could be the 512GB stock is all going in to Apple datacenters.

There seem to be an awful lot of YouTubers clustering these things. Maybe Apple have just run out of stock, lol.

Seems odd that those YouTubers were talking about some new technology to have these things connected together, with Studios provided by Apple to show off the new tech, with 2TB combined RAM, and now you can’t buy one.

 

[jole]

The problem with AI has nothing to do with its capabilities (or lack thereof), but the economics of it. Right now companies are pouring billions and billions into AI (mostly In the construction of data centers, but also R&D), but not getting any money out of it. Google, OpenAI, Anthropic, Microsoft, etc. are desperately trying to monetize runs, but thus far have failed to do so. Eventually all of these companies are going to realize that there isn’t enough money in AI and drastically scale things back - at which point the bubble will burst.

Anthropic is killing it:

Key financial details earlier in 2026 from their funding round:

• Revenue Growth: Revenue has grown more than 10x annually over the past three years.
• Enterprise Focus: Over 500 customers spend more than $1 million annually on Claude.
• Key Products: The Claude Code tool is driving significant growth, with a run-rate exceeding $2.5 billion.
• Projections: Anthropic aims for positive cash flow by 2028.
  

Trajectory:

• January 2025: ~ $1B
• May 2025: ~ $3B
• Mid-2025 (June/July): ~ $4B
• August 2025: > $5B
• October 2025: ~ $7B
• End of 2025 (December): > $9B
• February 2026: ~ $14B
• March 2026: nearing $20B (~$19–20B reported) — I attribute this $5B uplift to Pete Hegseth 🙂

With this, I would guess that Anthropic is way ahead of the game in getting to black due to them planning costs with a significantly lower revenue estimate.

OpenAI might be in pickle through...

 

[mrhick01]

Waiting for the 1000 GB option in the Mac Studio Ultra 😛💰

If the RAMpocalypse continues, $20K low would be my low estimate, up to $40,000 for a 1 TB Studio.

I'm only talking about the RAM upgrade.

 

[visualseed]

If the M5 Ultra is planned for the summer, they might be subtly discouraging the M3 Ultra for this purpose (LLMs) so that they have plenty of RAM inventory for the M5 Ultra.

If the neural engine on the M5 is 4x the M4...imagine the performance of the M5 Ultra. In fact, it would be almost silly to purchase an M3 at this point for running LLMs.

It's only silly if you don't need it. If you have a project that needs it today you can't afford to wait.

 

[Kirkster]

I think it is that they are running down stock of the M3 Ultra. I expect the Mac Studio to get an M5 Ultra shortly. And yes I also run LLMs locally. I thought a 64 gig M4 Max was going to be enough for a while, and it is, but isn’t…

 

[coolfactor]

I’m not a techie like some here, but isn’t 512GB RAM as mentioned in the second sentence a teensie bit overkill?

It's an insane amount of RAM, but the Studio is targeting high-end workflows, and some do require that amount of runtime memory.

Two examples:

- movie production
- local development of AI model training and inference tech

The Studio is Apple's "mini Pro".

 

[erthquake]

There probably weren't that many people buying it with 512GB memory configuration, so this was an easy decision for Apple to make.

In the days before I finally pulled the trigger in late February on my M3 Ultra w/512GB, the ship dates kept extending further and further out. Apple was projecting mid-May for mine when I ordered it. Last week or so, they adjusted the ship date to mid-April.

 

[JSRinUK]

I’m not a techie like some here, but isn’t 512GB RAM as mentioned in the second sentence a teensie bit overkill?

Even for hobbyist purposes, I find my 128GB a bit of a squeeze at times. If I was a professional earning money, I’d most certainly be justifying the max RAM. It’s true what they say that you can’t ever have enough RAM (but also that you never seem to have enough money to pay for it! 😂).

 

[Kirkster]

Even for hobbyist purposes, I find my 128GB a bit of a squeeze at times. If I was a professional earning money, I’d most certainly be justifying the max RAM. It’s true what they say that you can’t ever have enough RAM (but also that you never seem to have enough money to pay for it! 😂).

Ain’t that the truth,looked at a 128Gig M5 Max and was like, nope can’t even remotely justify this, I can’t write it off, and I can’t bite on 5K for a laptop 🫠

 

[Corefile]

You could probably virtualize the 0.5TB Mac Studio in a shiney new Neo.