Jul. 14 at 7:18 PM
Apple is evaluating PrismML's AI compression technology, which can shrink large language models from roughly 54 GB to less than 4 GB, allowing them to run directly on iPhone 15 devices and newer while using up to 15 times less memory. The breakthrough could significantly improve Siri by enabling faster, more private on-device AI processing, reducing latency, cloud-computing costs, and dependence on remote servers.
PrismML claims its compressed models generate responses six to eight times faster while consuming three to six times less energy than conventional versions. Apple is reportedly testing the technology as part of its broader AI strategy following the public beta release of iOS 27. Analysts note that while AI efficiency gains could alter memory requirements for certain workloads, they are unlikely to reduce overall semiconductor demand as AI capabilities continue expanding across smartphones, PCs, and datacenter infrastructure.
$AAPL $NVDA $MU $GOOGL