> 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?
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.
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.
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?