ALITEQ.

apple quietly cut its priciest Mac's RAM by 81% and it might still be the smarter local-AI buy

Apple's own memory shortage just raised Mac Studio prices and gutted its top RAM tier — here's whether it's still worth buying over a GPU PC for running models at home.

Lena FischerUpdated 1h ago8 min readWeb story
Apple Mac Studio desktop computer on a desk

In June 2026, Apple quietly raised Mac Studio prices by up to 33 percent and, over the months before it, deleted the two biggest memory configurations it ever offered — the 512GB tier, then the 256GB tier. What's left is a Mac Studio that costs more and remembers less. And it's still, for one specific job, arguably the smarter buy than building a GPU PC. Here's the real math, not the vibes.

What Apple actually did to its own lineup

Apple's statement, reported by JD Hodges, was blunt: the company said it had 'never seen a component price increase this much, this quickly.' The reason is the same one you've read about on the GPU side — AI data centers buying up memory production capacity at a scale that's dragging retail RAM and flash prices up across the entire industry, the same shortage that delayed Nvidia's own 24GB RTX 50 Super cards to 2027. Apple's response was to raise prices and shrink what's on offer: the M3 Ultra Mac Studio, the only Mac Studio that ever offered more than 64GB, jumped from $3,999 to $5,299 for a config that used to top out at 512GB and now stops at 96GB.

Mac Studio, before and after the June 2026 hike

M4 Max

Config
$1,999
Old base price
$2,499
New base price
64GB

M3 Ultra

Config
$3,999
Old base price
$5,299
New base price
96GB (was 512GB)

The sticker-price comparison everyone stops at

Line the Mac up against a GPU PC and the Mac loses the first round badly. A 96GB M3 Ultra Mac Studio now starts at $5,299. A PC built around two used RTX 3090s — 48GB of dedicated VRAM combined — runs somewhere around $2,000 to $2,800 once you've priced in a compatible motherboard, PSU, and case, with the GPUs themselves at roughly $800-1,100 each on the used market. On paper, the PC is less than half the price for a comparable amount of usable memory. Most comparisons stop right there. That's the mistake.

Mac Studio (96GB) vs a dual-RTX-3090 PC (48GB)

Mac Studio M3 Ultra, 96GB

$5,299

vs

PC, 2x used RTX 3090, 48GB

~$2,000-2,800

$5,299
Upfront cost
~$2,000-2,800
96GB unified (shared CPU/GPU)
Usable memory for models
48GB dedicated VRAM
~$15-25/month
Power at 24/7 inference
~$80-120/month
Near-silent
Noise under load
Loud, needs real case airflow
Slower — memory bandwidth-bound
Raw throughput on models that fit in 24GB
Faster — dedicated GPU bandwidth
96GB wins 3wins 2 48GB

The number nobody puts in the spreadsheet: power

This is the part that flips the math over a few years. Running a Mac Studio M4 Max for local inference around the clock costs roughly $15-25 a month in electricity, against roughly $80-120 a month for a comparable GPU PC running continuously, according to measurements from XDA Developers. That's not a rounding error — over three years it's the difference between about $600 and $3,600 in power alone, and it only gets worse for the PC if your local electricity rate is above the US average, which most of Europe's is.

Where the GPU PC still wins

None of this makes a Mac Studio the right answer by default. For any model that comfortably fits under 24GB, a single RTX 3090 or a current RTX 50-series card is faster in raw tokens-per-second and cheaper outright — our own benchmarking on Mac Studio's M4 Max against Nvidia's GB10 and AMD's Strix Halo found memory bandwidth isn't the whole story, and a dedicated GPU still has an edge in decode throughput at matched precision. If you also want to game on the machine, or you need two GPUs for a 70B-class model and don't mind the noise and power bill, a PC remains the more flexible box.

Custom PC tower with a graphics card installed, side panel open
A GPU PC still wins on raw speed for models that fit comfortably in 24GB of VRAM. · Unsplash

The honest verdict

Verdict

Buy the Mac Studio if quiet, big-model headroom is the actual point

For a machine that sits in a home office running a large model continuously without sounding like a hair dryer or showing up on your power bill, the Mac Studio's total cost of ownership catches up to a GPU PC faster than the sticker price suggests — often inside two to three years of heavy use. If your real requirement is raw speed on a model that already fits in 24GB, skip the memory premium and buy the GPU.

Best for: Buyers prioritizing quiet, always-on capacity for large models over raw tokens-per-second.

Quick answers

Can a Mac Studio actually run a 70B parameter model?
Yes — a 70B model quantized to around 40GB fits comfortably inside the 64GB M4 Max config, and easily inside the 96GB M3 Ultra config, with room left for the OS and context.
Is a single-GPU PC cheaper than a Mac Studio?
For models under about 16-24GB, yes, clearly — a single RTX 5070 Ti or used RTX 3090 costs a fraction of even the base Mac Studio. The Mac's case is specifically for memory-heavy models a single consumer GPU can't hold.
Will Apple bring back the 256GB or 512GB memory options?
There's no official timeline. The removals were tied to industry-wide DRAM shortages driven by AI data center demand, and that pressure hasn't shown signs of easing as of August 2026.

The uncomfortable pattern here is that Apple and Nvidia are being squeezed by exactly the same thing, and both companies chose to pass it straight to buyers rather than eat the margin. If you're shopping the local-AI hardware market right now, that's worth internalizing: this isn't really a Mac-vs-PC decision anymore, it's a bet on which company's supply chain absorbs a memory shortage better. For a deeper look at the PC side of that bet, see our current GPU picks and the cheapest ways in.

AI & Local Compute Editor

Lena Fischer

Lena runs more GPUs at home than she'll admit to and has quantized more models than she's finished reading about. She writes about running AI on your own hardware — what actually fits, what's genuinely fast, and what the polished cloud demos quietly leave out.

Work out the hardware

The Aliteq brief

The tech worth knowing — hardware, AI, gaming, deals. No spam, unsubscribe anytime.

Keep reading