Qualcomm’s new chip tackles the AI memory crunch

Your next smartphone may want to do a lot more thinking. The problem is that AI likes to eat memory.

Qualcomm is trying to deal with that tension with two new premium mobile platforms, the Snapdragon 8 Elite Extreme Gen 6 and Snapdragon 8 Elite Gen 6.

The chips are designed for a new generation of high-end Android phones, with a major focus on running AI directly on the device. But the interesting part is not simply that they are faster.

Qualcomm is also changing how the chips handle AI data, at a time when memory is becoming an increasingly important part of the smartphone cost equation.

Why phone memory has become an AI problem

AI features need quick access to large amounts of information. A voice assistant may need to understand what you just said, remember the context and respond. A camera feature may have to process an image while other apps are running in the background.

All of that puts pressure on memory, particularly DRAM. The problem is that memory suppliers are also dealing with huge demand from AI data centres, which use high-bandwidth memory for their workloads.

That can push up costs and make adding more memory to smartphones less attractive, especially in devices where margins are already tight. Qualcomm is not claiming to solve the memory shortage. Instead, it is trying to make the AI workload less dependent on constantly reaching outside the chip for data.

Keeping more AI data closer to the action

The main change sits inside Qualcomm's Hexagon NPU, or neural processing unit.

An NPU is a processor specifically designed for AI workloads. Qualcomm says the new architecture includes a larger shared memory system, with 50% more on-chip memory than the previous generation.

Instead of repeatedly pulling data from external memory, the NPU can keep frequently used information closer at hand. That can include model state, context and intermediate data generated while an AI task is running.

It becomes particularly useful for agentic AI, where an assistant may need to remember what happened earlier, plan several steps and take actions rather than simply answer one question.

The phone becomes more of the AI computer

The bigger ambition is to move more AI processing onto the smartphone itself. That can mean quicker responses, less dependence on cloud servers and more personal information staying on the device. It also gives chipmakers more room to build AI features that work even when connectivity is limited.

The Snapdragon 8 Elite Extreme Gen 6 sits at the top of Qualcomm's new range, while the Snapdragon 8 Elite Gen 6 brings many of the same AI, camera, gaming and connectivity improvements to a broader group of premium devices.

Honor, iQOO, Motorola, Redmi, Oppo, RedMagic, OnePlus, vivo and Xiaomi are among the brands expected to use the new platforms in upcoming flagship phones. But there is a bigger story underneath.

As phones become increasingly dependent on AI, chipmakers cannot simply keep adding more processing power and memory. They also have to figure out how to make AI use those resources more efficiently.

The next smartphone race, in other words, may not just be about who has the fastest chip. It could be about who makes AI the least hungry.