Apple Evaluates PrismML Technology to Run Large AI Models on iPhones

As demand for on-device AI computing surges, Apple is actively seeking ways to reduce reliance on the cloud and strengthen privacy. Running large language models locally on iPhones is key to enhancing Apple Intelligence. That has made PrismML, a California Institute of Technology spinout specializing in model compression, a focus of Apple’s evaluation. Its technology can sharply reduce model sizes while preserving computational performance.

PrismML emerged from stealth on March 31, 2026, after raising $16.25 million in seed funding. On July 14, the company unveiled technology that compressed Alibaba’s open-source, 27-billion-parameter Qwen 3.6 model from 54GB to less than 4GB, allowing it to run on an iPhone 17 Pro. CEO Babak Hassibi confirmed that the company had held preliminary discussions with Apple.