Qualcomm pushes data-center-class AI toward smartphones Qualcomm试图把数据中心级AI能力推向手机
Qualcomm is pitching smartphone silicon as a way to bring more data-center-style AI capability onto the device, according to the candidate feed. The practical question is memory bandwidth and power, not only TOPS: handset inference has to run within a thermal envelope that server accelerators never face. If Qualcomm can move larger models locally, it shifts some demand from cloud inference toward edge-optimized accelerators and high-bandwidth mobile memory. 候选源称,Qualcomm正在把手机芯片定位为承载更多数据中心式AI能力的载体。真正的约束不只是TOPS,而是内存带宽和功耗:手机端推理必须在服务器加速器无需面对的热设计范围内运行。如果Qualcomm能把更大的模型放到本地执行,部分云端推理需求会转向边缘侧加速器和高带宽移动内存。