CNBC reports Apple has held preliminary talks with AI startup PrismML to assess
model-compression tech that would let large AI models run on iPhones without
cloud compute. PrismML says it can reduce model footprint memory 10–15x,
enabling a 27 billion-parameter model to run on iPhone 15 and later, with 6–8x
faster response and 3–6x lower energy use. If validated at scale, Apple could
shift more Siri and generative-AI functions on‑device, improving latency,
privacy, offline use and cutting cloud costs; real-world impacts on battery
life, stability and long-term performance remain untested.
