The new Mac Studio holds 512GB of memory the GPU can address directly — a rack's worth of model, under a desk
Apple announced the M6 and the M5 Ultra on 25 August, ahead of its usual autumn event. The M6 — Apple's first 2nm chip, with a 12-core CPU, a 12-core GPU and two 16-core Neural Engines — goes into a Mac mini starting at $899 that Apple says is up to 4x faster than the M4 machine on AI work. The M5 Ultra, a quad-die part with 80 GPU cores, goes into a Mac Studio starting at $5,499 with up to 512GB of unified memory at 1.2TB/s, and Apple claims up to 4.3x the peak AI compute of the M3 Ultra. Pre-orders opened the same day and both machines ship on 22 September, though the 512GB configuration is not due until late October.
Why it mattersOn a PC a model has to fit inside the graphics card's own memory, and consumer cards stop far short of this. Apple's unified memory is addressable by the GPU in full, so a box you can carry in one hand can hold weights that would otherwise need server hardware — and run them without a single token leaving the room. Apple is now selling the machines in those words: it calls the Mac mini a desktop for always-on agentic computing, and says four clustered Mac Studios infer up to 3x faster than one.
Read in the app — free, in 9 languages
Related stories
One prompt could take over every AI agent in an Amazon account.2026-10-09The maker of the agent used on Korean banks deleted its code.
2026-10-09A new scoreboard grades how often agents do what you never asked.
2026-10-09OpenAI's cheap model now has a mode costing six times more.
2026-10-09One message can run code as root on an unpatched LLM cache server.
2026-10-09