Local AI hardware scramble
Social posts flagged rising demand, shortages and scam warnings for Mac Studio‑class machines as people hunt local hardware capable of running larger models. (x.com) One post highlighted running GLM 5.1 at roughly 17tok/s as an example of why enthusiasts are chasing those setups. (x.com)
People hunting computers that can run larger artificial intelligence models at home are zeroing in on Mac Studio-class machines, just as many high-memory configurations have slipped into long waits or “currently unavailable” status. (apple.com) (9to5mac.com) Apple’s current Mac Studio line starts at $1,999 for an M4 Max model and $3,999 for an M3 Ultra model, with memory options ranging from 36 gigabytes to 256 gigabytes on the online configurator. Apple says the M3 Ultra version can be configured with up to 80 graphics cores and 819 gigabytes per second of memory bandwidth. (apple.com 1) (apple.com 2) Apple said on March 5, 2025 that M3 Ultra supports up to 512 gigabytes of unified memory, but Apple’s current retail configurator shows Mac Studio topping out at 256 gigabytes. MacRumors reported on April 6 that Apple had removed the 512-gigabyte Mac Studio option and that some 256-gigabyte configurations were showing 4-5 month delivery estimates in the United States. (apple.com 1) (apple.com 2) (macrumors.com) A large language model is a prediction engine that keeps huge tables of numbers in memory while it generates text, code, or images. Machines with very large shared memory pools matter because they can hold bigger models locally instead of splitting work across cloud servers. (apple.com) Apple has leaned into that use case in its own marketing. In the M3 Ultra launch announcement, the company said Mac Studio with M3 Ultra can run large language models with more than 600 billion parameters directly on device. (apple.com) The model cited in the social posts, GLM-5.1, is Z.AI’s latest flagship system for long-running coding and agent tasks. Z.AI says GLM-5.1 can work on a single task for up to eight hours and says its predecessor GLM-5 scaled to 744 billion parameters with 40 billion active parameters. (docs.z.ai) (z.ai) (github.com) That helps explain why buyers are chasing memory-heavy desktops instead of thinner laptops. Apple’s own specifications show the Mac Studio’s M3 Ultra model with 819 gigabytes per second of memory bandwidth and a front pair of Thunderbolt 5 ports, while the smaller M4 Max model tops out at 128 gigabytes of memory. (apple.com 1) (apple.com 2) The supply squeeze is not limited to one exact model. MacRumors reported U.S. delivery estimates of 16-18 weeks for some Mac mini configurations with 64 gigabytes of memory, and 9to5Mac reported on April 11 that some Mac Studio and Mac mini configurations had become completely unavailable in Apple’s configurator. (macrumors.com) (9to5mac.com) Retail listings show why secondhand and gray-market offers are likely to multiply when supply tightens. B and H Photo lists an M3 Ultra Mac Studio with 96 gigabytes of memory and 1 terabyte of storage, a configuration pitched for “massive datasets for large language models,” which puts a concrete retail target on a machine many buyers now cannot get quickly from Apple. (bhphotovideo.com) For now, the scramble is being driven by a simple equation: bigger local models need more memory, and the machines built for that job are getting harder to buy on short notice. Apple is still selling Mac Studio online, but the fastest path appears to be whatever configuration is actually in stock. (apple.com) (macrumors.com)
