Qualcomm launched two flagship smartphone processors today, the Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6, built around on-device AI as the core differentiator. The Extreme version can run a 30-billion-parameter mixture-of-experts model locally, a specification that puts a phone chip ahead of the 20-billion-parameter MoE Apple unveiled at WWDC in June as the flagship of its third-generation foundation models. Both chips ship with new sensing hubs capable of running models up to 200 million parameters continuously in the background.
The pitch is memory and personalization. Qualcomm said the sensing hub can run a personal scribe locally, differentiate between speakers in a conversation, and build a usage-based memory that feeds task automation suggestions. The company said the new chip can run a complete voice-in and voice-out agent end-to-end without a network round trip, which is the kind of latency and privacy claim that only holds up if the accelerator hardware is genuinely doing the work.
The Gen 6 gets a redesigned accelerator element aimed at efficient inference for the smaller models most consumer features actually use. The Extreme reserves the 30B MoE for higher-end scenarios — the mixture-of-experts design means only a subset of those 30 billion parameters activates for any given prompt, keeping the memory bandwidth and power draw within a phone's thermal envelope. That's the same architectural bet Apple made with its 20B MoE, and it's now the default approach for on-device flagship models.
Key facts
- 01Snapdragon 8 Elite Extreme Gen 6 runs a 30B-parameter MoE model on-device, versus Apple's 20B MoE announced at WWDC in June.
- 02New sensing hubs on both chips run models up to 200M parameters locally, enabling always-on personal scribe and speaker differentiation.
- 03The Extreme version supports 8K 60fps and 4K 240fps video capture, plus the new Advanced Professional Video codec.
- 04Motorola's Signature 27, launching later this year, is the first announced device using the Extreme chip.
- 05Qualcomm has now worked on more than 40 AI devices, but is betting the phone remains the primary AI form factor.
Motorola announced the first device built on the Extreme chip at the Snapdragon Summit — the Motorola Signature 27, with general availability set for sometime this year. That's a notable design win. Motorola has spent the last several product cycles trying to reclaim premium shelf space, and being first-to-market with a chip that outspecs Apple on raw local-model size is the kind of headline it can build a marketing campaign around.
The camera and video story got a matching upgrade. The new Qualcomm CPU provides pixel-level control for cameras with improved stabilization and motion understanding, and the Extreme version supports 8K video at 60fps and 4K at 240fps for ultra-high-frame-rate slow motion. Qualcomm is also enabling the new Advanced Professional Video codec on the Extreme, aimed at content creators who want a phone that can genuinely substitute for a dedicated camera on a shoot.
Audio gets AI treatment too. Both chips can boost vocals and suppress background noise using neural processing, and Qualcomm introduced a new voice bubble feature that isolates the user's voice during calls in noisy environments. These are the kind of features that show up in the demo reel and then quietly become the feature users notice most when they switch devices.
The broader strategic context matters. Qualcomm has now worked on more than 40 AI devices — smart glasses, wearables, dedicated AI hardware — but the company's public posture is that the phone remains the AI form factor most consumers will actually use. Nothing co-founder Carl Pei has said the same, and new Apple CEO John Ternus reinforced the point at the iPhone Duo launch earlier this month. After a year of ambitious dedicated-AI-device launches, the industry consensus is quietly converging back on the pocket.
The counterweight: raw parameter counts on a spec sheet don't automatically translate to a better user experience. Apple's 20B MoE is tightly integrated with the operating system, the App Intents framework, and a set of curated on-device use cases. Qualcomm is selling a chip; what any given Android OEM does with a 30B on-device model — and whether the software stack, the fine-tuned adapters, and the developer APIs are actually in place to use it — is a separate question that won't be answered until devices ship and users try them.
What Qualcomm has done, though, is remove hardware as the bottleneck. The on-device AI conversation for the next twelve months is now a software conversation: which OEM ships the best assistant, which model developer ships the best fine-tune for the sensing hub, whose agent architecture actually delivers a voice-in, voice-out experience that beats a cloud call to ChatGPT or Gemini. Qualcomm has handed the ecosystem a chip that can run flagship-class models locally at a moment when Apple, Google, and every Android maker are all racing to define what an AI phone actually does.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
