The Next Valuation Driver for Mobile AI Might Be Memory

Just saw OPPO partnering with OpenKG and several universities to launch MobileMem, an on-device AI memory evaluation benchmark. My first reaction was about defining the entry point.

Over the past two years, AI phones mainly solved "can it chat." I've seen many projects last month covering foundation models, on-device compute, and power consumption, with many business plans still competing on benchmark scores. But an internal OPPO saying is more interesting: Data is dead, memory is alive. If phones truly become 24/7 online personal agents, beyond model size, long-term retention of user contexts like photos, schedules, chats, and locations will become competitive moats.

This affects valuation logic. Hardware makers talk about shipment volumes and ASP; platform makers talk about services and retention. If on-device memory becomes an evaluation standard, phones will look more like containers for personal AI. Whoever turns context into task execution holds distribution power. Where the ceiling is for this track depends on how much memory users cannot take with them when they leave the phone.

However, since the benchmark is just out, whether it becomes vendor propaganda for self-validation remains to be seen, depending on third-party reproducibility. Whether app developers buy into standards set by phone manufacturers also needs verification.