端侧智能体不再只缺算力，个人AI还差什么？

TL;DR - An IDC white paper commissioned by Qualcomm argues that personal AI requires distributed, always-on agents spanning devices, edge systems, and the cloud. Compute is no longer the only constraint; memory, power efficiency, privacy, and cross-vendor interoperability are increasingly decisive.

• On-device sensing and memory provide personal context, while cloud models supply heavier computation and public knowledge.
• Memory bandwidth and battery life can constrain sustained agent workloads more than peak NPU TOPS.
• Heterogeneous hardware assigns sensing, orchestration, inference, and connectivity to specialized low-power components.
• Securely transferring user memory across brands, operating systems, and devices remains an unresolved ecosystem and data-governance challenge.