You can turn a cluster of Macs into an AI supercomputer in macOS Tahoe 26.2 (www.engadget.com)

Apple’s macOS Tahoe 26.2 adds a low‑latency Thunderbolt 5 clustering feature that lets multiple Macs act as a single compute system, enabling researchers and developers to stitch together Mac Studios, M4 Pro Mac minis, and M4 Pro/Max MacBook Pros with standard Thunderbolt 5 cables (up to 80 Gb/s). In a demo using ExoLabs’ EXO 1.0, four Mac Studios (each capable of up to 512 GB unified memory) jointly loaded and ran the 1‑trillion‑parameter Kimi‑K2‑Thinking model while consuming under 500 W — roughly an order of magnitude less power than comparable GPU racks. This capability removes the prior practical bottleneck of slower Thunderbolt links (and hub‑induced 10 Gb/s drops) and requires no special clustering hardware. Technically, Tahoe’s support makes it easier to scale large local models on Apple Silicon by leveraging unified memory across nodes and high‑speed PCIe/TB5 links; macOS 26.2 also exposes M5 neural accelerators to Apple’s open‑source MLX stack, promising faster on‑device inference. Limitations remain: M5 machines shipping today (the 14″ MacBook Pro) only have Thunderbolt 4, so they can’t join TB5 clusters yet, and high‑end Mac Studio configurations are expensive. Still, for labs and companies with existing Macs, this is a practical, energy‑efficient on‑prem route to run massive models without power‑hungry GPU farms.