Apple's New Mac Chips Are Built for AI That Runs on Your Desk
The M6 is Apple's first 2-nanometer chip and doubles the Neural Engine. The M5 Ultra goes quad-die with 512GB of memory. Both landed on August 25 in a new Mac mini and Mac Studio.
Apple announced two new processors on August 25 without any of the usual keynote buildup: the M6, its first chip built on a 2-nanometer manufacturing process, and the M5 Ultra, the first M-series chip made from four connected pieces of silicon instead of one or two. The M6 ships in a new Mac mini, the M5 Ultra in a new Mac Studio. Both announcements are unusually focused on one thing: how much AI work these machines can do without sending anything to a server.
The numbers, briefly. The M6 has a 12-core CPU, a 12-core GPU, and what Apple calls a Dual 16-core Neural Engine, a doubled-up version of the dedicated AI chip block, which Apple rates at up to twice the peak AI compute of the previous generation. Memory starts at 16GB and tops out at 32GB, with bandwidth of 170GB/s. The M5 Ultra is the big one: up to a 36-core CPU, up to an 80-core GPU, and up to 512GB of unified memory running at 1.2TB/s. Apple claims up to 4.5 times the AI GPU compute of the M3 Ultra it replaces. Reported starting prices put the new Mac mini around $899 and the Mac Studio around $5,499, roughly $100 to $200 above the tiers they replace.
What’s actually going on here: two numbers matter more than the rest, and neither is the CPU core count. The first is memory bandwidth, how fast the chip can shovel data around, because running a large language model locally is mostly a memory problem rather than a math problem. The second is unified memory size, since a model has to fit in memory to run at all. 512GB in a desktop you can carry with one hand is the headline: that is enough room for models that until recently needed a rack of server GPUs. “2-nanometer” is a manufacturing label rather than a literal measurement, but it does mean more transistors in the same space and less power for the same work. Apple is not marketing these as AI machines by accident. Its pitch is that the interesting model you want to run should run on the desk, not in someone else’s data centre.
What this means for you: if you have never run an AI model on your own computer, nothing here is urgent, and a current Mac already handles chat, transcription and image tools fine. If you have been curious about local AI, where the model runs on your machine and your data never leaves it, this is the clearest signal yet that the hardware is arriving to make that normal rather than niche. For developers and small teams working with open-weight models, a 512GB Mac Studio is now a serious alternative to renting cloud GPUs by the hour, and worth pricing out against a year of cloud bills. Worth keeping expectations grounded: Apple’s multiplier claims are peak figures under ideal conditions, and real-world speed on any given model will be lower.
Sources
Claude Now Remembers You in One Place Instead of Two
Anthropic merged the memory systems behind Claude chat and Claude Cowork on August 25. It is on by default, sensitive topics are off by default, and you can read and delete everything it keeps.
