Global smartphone shipments grew 4.2 percent year-over-year in the latest quarter, driven by consumer interest in hardware-native artificial intelligence tools. Equipment manufacturers are incorporating dedicated Neural Processing Units (NPUs) directly into system-on-chip designs to run large models locally.
Memory Constraints and Component Costs
Running multi-billion parameter models locally requires a minimum of 12GB of DRAM, inflating bill-of-materials costs for mid-range devices. Component suppliers are benefiting from higher average selling prices despite elevated semiconductor input costs.