Apple's Mac Studio with M5 Ultra chip brings local AI power to the desktop
Wired reviewed Apple's new Mac Studio with the M5 Ultra chip, which can run large AI language models entirely offline on the device. The machine scored 9 out of 10, though its top configuration costs as much as $11,000.
%20092026%20top%20art%20SOURCE%20Amazon.jpg)
Apple's new Mac Studio desktop computer, equipped with the M5 Ultra chip, is capable of running advanced AI language models directly on the device without any internet or cloud connection. A Wired review gave the machine a 9 out of 10 rating.
The Mac Studio comes in two versions. The base model, featuring an M5 Max chip, 36GB of memory and 512GB of storage, starts at $2,499 and targets content creators and game developers. The unit used for testing, however, was the top configuration with an M5 Ultra chip and 256GB of memory, priced at $11,000 — the most expensive computer the reviewer had ever tested.
The M5 Ultra is described as Apple's most powerful chip to date, combining two M5 Max chips into a single unit with 36 CPU cores and up to 80 GPU cores. In benchmark tests, it outperformed an M5 Max-equipped MacBook Pro by 51 percent in CPU performance and 32 percent in graphics performance.
AI and gaming performance
The reviewer tested the system's ability to run large language models locally. A task that took nearly 10 minutes on an M5 Max MacBook Pro using a smaller AI model was completed in just 2 minutes on the M5 Ultra Mac Studio. Even when running a much larger model with 122 billion parameters, the task still finished in roughly the same time, producing more accurate results.
The machine also impressed in gaming tests, running Cyberpunk 2077 at 160 frames per second — 61 percent faster than the M5 Max. On-device AI video generation, however, proved far slower, taking over 11 minutes to produce a 15-second clip.
During testing, the reviewer noted that the Mac Studio runs very hot under heavy load, with fans becoming noticeably loud. Upcoming competition is expected from Nvidia's RTX Spark chips and cheaper mini PCs using AMD processors that also rely on unified memory for local AI tasks.
