Which means nothing. Apple is extremely good at calculating how much inventory they should produce. They’re not going to make more than they’ll sell and have them sit on shelves. They’re also not going to cut into their profit margin too much by paying their way out of the memory shortage.
If a new premiere high-margin product sells out "within a month" that means there's a disruption in sales and an opportunity for competitors to meet the demand they just generated while manufacturing/distribution tries to catch up.
It would absolutely represent either a mis-estimation of demand or compromise to some supply/manufacturing bottleneck that they couldn't overcome without sending their release calendar into chaos.
I think I agree with the economic sentiment in general but who is the competitor that can even make something comparable? And by that I mean: Matches the memory amount, the memory bandwidth, the form factor, the power draw, the compute and the OS
> Apple is extremely good at calculating how much inventory they should produce.
That doesn't really matter though for a physical product, unless you are implying they are 1) selling it at a loss or 2) not attempting to maximize profits from the product. If they sell out, it means they hit the target profit from the sale of the product.
What's the argument here? Are you saying that Apple is manufacturing scarcity? All you're saying is that Apple is anticipating demand, which everybody tries to do, but it kinda sounds nefarious or at least artificial.
When the M3 Ultra 256GB+ models got discontinued the secondary market prices went through the roof. That's not manufactured demand.
All the M5 Ultra Mac Studios have had 3-4+ month delivery estimates within a short time of launch until now. I occasionally check and noticed that the 96GB model has moved up from February to December where I live (pickup or shipping). There's clearly a demand for these things and it's widely believed it's largely AI.
Getting an equivalent amount of ram in GPU's is substantially more expensive than this. A single RTX 6000 Blackwell with 96gb of ram is like $16,000 now. Someone please correct me if I'm wrong but essentially with the Mac Studio you're getting the ability to run larger models, at slower speeds, for a lot cheaper.
- Larger MoE models only need compute proportional to the number of active parameter, so Qwen3.8-Flash-Next (180b with 6b active) runs more tok/sec than Qwen3.8-27b on the same hardware.
- Larger models typically have higher quality output and require fewer thinking tokens to get to an answer.
Consequently, measuring time to complete a task with a wall clock, the speed difference is less than you would expect.
Large mixture-of-experts models, like Muse Spark or Kimi K3.
MoE models want a different ratio of memory/compute/$ than dense models like a Gemma4 31B or Qwen3.8 27B.
Since only some of the experts are active each round, you don’t need as much compute to get good tok/sec, but you do need a lot of memory to avoid streaming weights from SSD or system RAM.
You could run a full Muse Spark on one of these for $24k at probably around 50 tok/sec.
You’d need 5x RTX 6000 Pros at $80k for Muse Spark (albeit at much higher tok/sec).
On a $24k budget, you’re looking at a maximum of 96GB from Nvidia or 384GB from Intel or AMD. So you’re bottlenecked by PCIe bandwidth, and all those TFLOPS go to waste.
Other configurations of the Studio have similar logic for other MoE models. For instance, Qwen3.8 flash-next runs at a very acceptable 75 tok/sec on a 128GB M5 Max for about 1/3rd the price of an RTX 6000 or 40% less than 4x B70’s.
1.3TB/s of Unified Memory. I'm not sure today how much distributing dense models across GPUs is ok or bad these dats but you don't have to think about it by having all the memory pool available on the one GPU you have.
Which means nothing. Apple is extremely good at calculating how much inventory they should produce. They’re not going to make more than they’ll sell and have them sit on shelves. They’re also not going to cut into their profit margin too much by paying their way out of the memory shortage.
If a new premiere high-margin product sells out "within a month" that means there's a disruption in sales and an opportunity for competitors to meet the demand they just generated while manufacturing/distribution tries to catch up.
It would absolutely represent either a mis-estimation of demand or compromise to some supply/manufacturing bottleneck that they couldn't overcome without sending their release calendar into chaos.
I think I agree with the economic sentiment in general but who is the competitor that can even make something comparable? And by that I mean: Matches the memory amount, the memory bandwidth, the form factor, the power draw, the compute and the OS
1 reply →
Sounds like a great way to generate interest and FOMO.
Did you know the McRib is back?
But only for a limited time so you better act fast!
> Apple is extremely good at calculating how much inventory they should produce.
That doesn't really matter though for a physical product, unless you are implying they are 1) selling it at a loss or 2) not attempting to maximize profits from the product. If they sell out, it means they hit the target profit from the sale of the product.
It could also mean they're supply-constrained, and cannot set or reach future targets without accounting for surge pricing.
What's the argument here? Are you saying that Apple is manufacturing scarcity? All you're saying is that Apple is anticipating demand, which everybody tries to do, but it kinda sounds nefarious or at least artificial.
When the M3 Ultra 256GB+ models got discontinued the secondary market prices went through the roof. That's not manufactured demand.
All the M5 Ultra Mac Studios have had 3-4+ month delivery estimates within a short time of launch until now. I occasionally check and noticed that the 96GB model has moved up from February to December where I live (pickup or shipping). There's clearly a demand for these things and it's widely believed it's largely AI.
The 256 GB is backordered by 13–15 weeks, if you click through to that.
What would be the use case for this over a dedicated 8xgpu server?
Getting an equivalent amount of ram in GPU's is substantially more expensive than this. A single RTX 6000 Blackwell with 96gb of ram is like $16,000 now. Someone please correct me if I'm wrong but essentially with the Mac Studio you're getting the ability to run larger models, at slower speeds, for a lot cheaper.
Right. Couple things to note though:
- Larger MoE models only need compute proportional to the number of active parameter, so Qwen3.8-Flash-Next (180b with 6b active) runs more tok/sec than Qwen3.8-27b on the same hardware.
- Larger models typically have higher quality output and require fewer thinking tokens to get to an answer.
Consequently, measuring time to complete a task with a wall clock, the speed difference is less than you would expect.
Exactly, optimizing for VRAM where you don’t need the RTX 6000 performance.
Large mixture-of-experts models, like Muse Spark or Kimi K3.
MoE models want a different ratio of memory/compute/$ than dense models like a Gemma4 31B or Qwen3.8 27B.
Since only some of the experts are active each round, you don’t need as much compute to get good tok/sec, but you do need a lot of memory to avoid streaming weights from SSD or system RAM.
You could run a full Muse Spark on one of these for $24k at probably around 50 tok/sec.
You’d need 5x RTX 6000 Pros at $80k for Muse Spark (albeit at much higher tok/sec).
On a $24k budget, you’re looking at a maximum of 96GB from Nvidia or 384GB from Intel or AMD. So you’re bottlenecked by PCIe bandwidth, and all those TFLOPS go to waste.
Other configurations of the Studio have similar logic for other MoE models. For instance, Qwen3.8 flash-next runs at a very acceptable 75 tok/sec on a 128GB M5 Max for about 1/3rd the price of an RTX 6000 or 40% less than 4x B70’s.
Qwen3.8 flash-next runs at 110 t/s on the Olares One using Strata. Cost USD$2.8K as an early backer. Now costs double.
<https://www.kickstarter.com/projects/167544890/olares-one-th...>
Data centers outfitted with Ikea furniture instead of the usual 19" racks.
1.3TB/s of Unified Memory. I'm not sure today how much distributing dense models across GPUs is ok or bad these dats but you don't have to think about it by having all the memory pool available on the one GPU you have.