TL;DR
Qualcomm's new Snapdragon 8 Elite Gen 6 chips can run a 30-billion-parameter AI model entirely on-device, pushing the smartphone into territory previously reserved for cloud or dedicated AI hardware.
What happened
- Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6 were unveiled at Qualcomm's annual Snapdragon Summit on September 22, 2026.
- The Extreme variant runs a 30B mixture-of-experts (MoE) model locally, activating only a subset of parameters per task to stay within mobile power budgets.
- A new sensing hub handles models up to 200M parameters for always-on tasks: speaker differentiation, personal scribe, usage-based memory, and full voice-in/voice-out agents.
- The Extreme chip supports 8K 60fps and 4K 240fps video, plus the new Advanced Professional Video (APV) codec for pro-level recording.
- Motorola announced the Signature 27 as the first device on the Extreme chip, with availability before year-end.
Why it matters
- Qualcomm's 30B on-device ceiling now exceeds Apple's 20B MoE foundation model (released at WWDC in June 2026), sharpening the silicon rivalry for AI-first smartphones.
- Running large models locally means no cloud latency, no data leaving the device, a direct answer to enterprise and privacy-sensitive consumer demand.
- The sensing hub's persistent memory and task-automation layer moves the smartphone toward a proactive agent, not just a reactive assistant.
- Apple CEO John Ternus and Nothing co-founder Carl Pei have both recently argued the phone remains the primary AI interface, validating Qualcomm's bet over dedicated AI gadgets.
- Qualcomm is already working on more than 40 AI devices, but smartphone volume gives these chips outsized reach compared to any niche AI hardware category.
What to watch next
- Motorola Signature 27 real-world benchmarks will be the first stress test of 30B MoE performance and battery impact under sustained use.
- Watch whether Apple responds with a higher-parameter model in its next foundation model generation, or shifts strategy toward efficiency rather than raw parameter count.
- Adoption by other OEMs beyond Motorola will signal whether the Extreme tier becomes a mass-market standard or a premium niche, determining how quickly 30B on-device AI reaches scale.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.