Qualcomm's next mobile chip hits 5GHz, built for agents

Qualcomm's next Oryon CPU runs at 5GHz, split across two Prime Cores and six Performance Cores. That's the clock speed and core count. The design choice underneath it is what's new.

The chip uses something Qualcomm calls FlexCache: both core types draw from one shared, dynamically allocated cache pool instead of separate fixed allocations. Under load, a single Prime core can claim the entire pool and keep a large working set resident instead of spilling to slower memory.

Who benefits from that depends on what's running. Qualcomm's own list puts agentic AI first, ahead of gaming and video editing, the two workloads that have driven mobile chip marketing for a decade.

A large resident working set is exactly what an on-device agent needs: model weights, conversation state, and tool-call context all held in fast cache instead of getting evicted every time the phone's foreground app changes. Games and video editing have predictable, bursty memory patterns that fixed cache splits already handle reasonably well. Agent workloads are the opposite: long-running, state-heavy, and easy to starve if the OS treats them like a background process.

What's still unclear is the actual cache size, and how FlexCache behaves when a game and a background agent both want the pool at once. Qualcomm hasn't published either number. That's the detail worth watching for at the Snapdragon Summit in September, since a shared pool only helps agentic workloads if the allocator actually favors them when a phone is doing more than one demanding thing at a time.