Big Tech’s AI spending explosion: Microsoft commits $25 billion through 2026 as the industry pours hundreds of billions into AI amid bubble concerns
Regulatory battles intensify: Meta faces EU fines and Australian levies, Utah passes controversial VPN liability law, while China makes AI job displacement illegal
Critical security threats: severe Linux vulnerability and active cPanel exploits affecting millions of websites
AI policy shifts in entertainment, military drone training for Ukraine, and Sam Altman’s GPT-5.5 launch
One of the big selling points for Mac mini and Studio is EXO and RDMA over Thunderbolt 4/5, which was introduced in macOS 26.2. This allows the equivalent of an Infiniband connection between multiple Macs, sharing RAM over TB4/5 at 40-80gbps, allowing for larger inference models (E.g. if you have 4 Studios with M3 Ultra and 512GB RAM each, you can do inferencing on a 2TB (less OS overhead) MLX models. Given the 512GB model is no longer available, that drops to 1TB with 4x256GB Studios today.
nVidia offers something similar with the Spark series, but they start at around $32,000 a node. Otherwise you are talking about x64 servers with Infiniband and multiple nVidia NPU cards. 4 Studios make a very cost effective option, in comparison.