Apple is in early-stage discussions with AI startup PrismML, which has developed a model-compression technique that reduces large language models to run on smartphones. Founded in March 2026, PrismML holds exclusive commercial rights to patents from the California Institute of Technology. The company compressed Alibaba’s 27-billion-parameter Qwen model from about 54GB to roughly 4GB, allowing it to run locally on an iPhone during demonstrations. PrismML claims its technology enables models to run on devices with as little as 6GB of RAM, and its Bonsai family of models can deliver processing speeds five to eight times faster while reducing energy consumption by as much as 80%. Apple is exploring ways to bring more advanced AI processing onto its devices instead of relying on cloud servers. No agreement has been announced, and the technology is unlikely to appear in the upcoming iPhone 18 lineup.