7 ms·

Annecdotal here but I've been buying studios with as much ram as possible. I probably got in one of the last orders for the m3 ultra 512gb ram officially from a

Annecdotal here but I've been buying studios with as much ram as possible. I probably got in one of the last orders for the m3 ultra 512gb ram officially from apple late February right before they took that config down. I know this because the very next day I went back to order more and couldn't. There's a ton of spam on ebay currently with configs of the 512gb ram variant going for sub $4k. Those are all scams/spam. You can tell instantly because the seller's account is/was created this month 2026. The $20k+ listings aren't selling but what's interesting is technically, $20k for 512gb ram is still less than $40/gb ram on a single device. Compare that to a Nvidia spark with 128gb ram selling at $4k+ or if you're lucky, you snagged one for ~$3500. That's $31.25/gb The retail on the 512gb was $9500 and after tax if you don't ship to a sales tax free state makes it only $20/gb ram. I'm curious what price apple's newest studio variant will be priced at per gb ram but I recently bought a "refurb" from a reputable seller I've purchased many laptops from in the past at $13,750 which still put the 512gb at a $27/gb ram price point. I'm bullish on local models especially due to gemma4. Good luck out there!

embedding-shape 5mo ago
> but I've been buying studios with as much ram as possible Why though? Are you managing a fleet of headless Mac Studios? You mention LLMs, but today's Mac hardware really isn't great as soon as you use bit larger contexts, so I'm guessing it's not that either?
- taylorhou 5mo ago
  We combo frontier coding models with last frontier for more admin stuff.
  - embedding-shape 5mo ago
    But with macOS hardware? What kind of calculations did you do that shows that this makes sense? 6-7 months ago I looked into the same, getting a bunch of Apple hardware to do local LLM inference (for a client), but the bad performance + high pricing just makes it completely infeasible compared to nvidia hardware. Now I'm really curious how you got that to make sense in reality.
    - taylorhou 5mo ago
      250+ employees on the lowest paid subscriptions of claude or openai is $5k/month. their usage of AI is chatbot most of the time which local models/inference can easily handle. so just being able to cancel those subscriptions with local hardware makes my break even on a single $10k mac studio 2 months and honestly the mac studio for their use is overkill.

embedding-shape 5mo ago

> but I've been buying studios with as much ram as possible Why though? Are you managing a fleet of headless Mac Studios? You mention LLMs, but today's Mac hardware really isn't great as soon as you use bit larger contexts, so I'm guessing it's not that either?

taylorhou 5mo ago
We combo frontier coding models with last frontier for more admin stuff.
- embedding-shape 5mo ago
  But with macOS hardware? What kind of calculations did you do that shows that this makes sense? 6-7 months ago I looked into the same, getting a bunch of Apple hardware to do local LLM inference (for a client), but the bad performance + high pricing just makes it completely infeasible compared to nvidia hardware. Now I'm really curious how you got that to make sense in reality.
  - taylorhou 5mo ago
    250+ employees on the lowest paid subscriptions of claude or openai is $5k/month. their usage of AI is chatbot most of the time which local models/inference can easily handle. so just being able to cancel those subscriptions with local hardware makes my break even on a single $10k mac studio 2 months and honestly the mac studio for their use is overkill.