Back to today’s briefing

Story perspectives

Apple partners PrismML for compressed AI, may boost iPhone prices

7/16/2026

41 10 Full Breakdown

Summary

Apple Eyes Compression

Apple is in talks with PrismML to run compressed AI on iPhone 15 or newer.
PrismML reduced Alibaba’s 27-billion-parameter Qwen model from 54 GB to under 4 GB, a 10-15× shrink.
Analysts estimate Apple’s significant DRAM costs could rise 190 % and substantial NAND costs could rise 180 % by fiscal 2027, potentially adding $200 to iPhone 18 starter prices.

Summary

Apple Tests 1-bit

Apple is testing PrismML’s 1-bit and ternary neural network tech in the iOS 27 public beta.
PrismML, a Caltech startup, pioneered architectures that store each weight in one to three values instead of 16 bits.
The compressed models could power on-device photography, video generation, and health-fitness tools while keeping data local.
PrismML offers a Patreon tier from $2 to $10 a month for AI research content.

History

Historical Context

In 2017 Apple introduced the A11 Bionic with the first Neural Engine, integrated with Core ML for on-device AI in iPhone X. The effort expanded to Apple Intelligence in 2024 and, after a 2026 Google-Gemini partnership, Apple now works with PrismML to compress models for iPhone 15. The M4 Neural Engine reaches 38 trillion operations per second.

1 of 3

Apple Eyes Compression

Apple is in talks with PrismML to run compressed AI on iPhone 15 or newer.
PrismML reduced Alibaba’s 27-billion-parameter Qwen model from 54 GB to under 4 GB, a 10-15× shrink.
Analysts estimate Apple’s significant DRAM costs could rise 190 % and substantial NAND costs could rise 180 % by fiscal 2027, potentially adding $200 to iPhone 18 starter prices.

1 / 3