In this week's Sunday Reboot, Apple's interest in Prism ML's model-shrinking technology could yield massive benefits for on-device AI. The startup compressed a 27 billion-parameter Qwen model from 54GB to 4GB, a breakthrough that fits large language models onto consumer devices.

Prism ML CEO Babak Hassibi confirmed Apple and other firms are evaluating the technology, despite Apple's typical NDA secrecy. With John Ternus reportedly shifting Apple's acquisition strategy toward billion-dollar deals, an acquisition remains possible, though Hassibi may prefer licensing agreements.

For Apple, the technology addresses a core constraint: iPhones lack the memory for sizable models, forcing cloud offloading. Current Apple Intelligence features require 12GB RAM, limiting them to iPhone 17 Pro, M4 iPad, and M3 Macs. Prism ML's compression could shrink these models further, enabling on-device processing on older hardware.

This represents a prime opportunity for Apple to dominate on-device AI. If the tech proves sound, it could expand Apple Intelligence reach while freeing memory for faster inference—a potential billion-dollar advantage in the AI race.