Desktops as personal supercomputers

Meta’s Muse Spark Open (816B/42B active, 1M token context) is tuned to run across two forthcoming 512GB M5 Ultra Mac Studios using Apple’s unified memory and MLX, turning high‑end consumer desktops into nodes that can hold and execute agentic, multi‑step models locally. This shifts some AI compute from hyperscale clouds into premium devices, concentrating capability in buyers of expensive hardware and changing who controls, audits, and hosts powerful models. — If powerful open models can run on commodity‑sold desktops, debates about privacy, vendor lock‑in, export controls, hardware supply chains (memory, SSD, GPUs), and equitable access to AI capability will move from datacenters into consumer and regulatory policy arenas.