Qualcomm's new flagship chips can run 30B AI models on your phone

Most of the AI features running on your phone right now are basically just wrappers sending requests to a distant server farm. If your signal drops in an elevator, your smart assistant suddenly becomes completely useless.

Qualcomm wants to change that. At its annual Snapdragon Summit on Tuesday, the company unveiled its next-gen mobile platforms: the Snapdragon 8 Elite Extreme Gen 6 and the Snapdragon 8 Elite Gen 6. The Extreme variant is destined for ultra-premium flagships, while the standard version will power more accessible devices.

Qualcomm CEO Cristiano Amon framed the launch around what he calls ambient agentic experiences. In plain English, that means your phone orchestrating complex tasks across different apps without you having to micromanage every tap.

Running 30B models in your pocket

The marquee stat from the keynote is on-device computing power. Thanks to an overhauled Qualcomm Hexagon NPU, the new chips can run models with up to 30 billion parameters, specifically MoE architectures, directly on the device.

That is a massive leap forward. Right now, most flagship phones tap out around 4 billion parameters for purely local processing. Moving to 30B brings significantly more reasoning capability into an environment that requires zero internet connection.

The platform ties together Qualcomm Oryon CPU cores, the Hexagon NPU, and an updated Sensing Hub inside its AI Engine. It also pairs with a newly designed, AI-powered Qualcomm X105 5G Modem-RF for rock-solid connectivity.

Beyond raw intelligence, the chips introduce camera upgrades with deeper real-time scene understanding. The Extreme platform also features AI Voice Bubble technology, which can isolate the speaker voice from chaotic background noise during calls or voice commands.

A long roster of Android makers is already signed up, including Samsung, Xiaomi, Honor, OnePlus, Oppo, Vivo, and Motorola. One of the first devices out of the gate will be the Motorola Signature 27 running the top-tier Extreme chip.

Why it matters

For everyday users, jumping from 4B to 30B parameters on-device means two things: latency drops to almost nothing, and your personal data stays on your hardware. If you want an AI assistant that actually reads your schedule, cross-checks your messages, and books an appointment without sending your private life to a cloud provider, this is the hardware foundation you need.

That said, we think you should wait before rushing to upgrade. Running heavy models locally produces heat and drains batteries fast. While Qualcomm promises big architectural efficiency gains, real-world throttling under heavy agentic workloads remains to be proven once commercial phones ship.

Strategically, Qualcomm is making a brilliant play. Instead of trying to fight cloud computing giants, it is cementing its position as the engine room for physical edge devices. Qualcomm silicon is already driving Meta Ray-Bans, Snap Specs, and the Samsung Galaxy Watch 9.

By dragging the on-device baseline all the way to 30B, Qualcomm is forcing competitors like MediaTek and Apple to drastically beef up their local neural processing units. The era of treating on-device phone AI as a toy looks like it is officially over.