News at a Glance
On September 22, 2026, Qualcomm launched two new chips for smartphones, with the flagship chip capable of running a 30-billion-parameter mixture-of-experts model locally. This move aims to strengthen on-device AI capabilities, address the growing demand for intelligent processing on the device, and is expected to reduce latency and enhance privacy protection, driving mobile AI applications to migrate to the device side.
Background
Smartphone makers are currently competing to use AI features as a differentiating selling point. Running large models on-device can reduce reliance on the cloud, lower latency, and protect user privacy. Qualcomm’s emphasis on locally running mixture-of-experts models shows its desire to consolidate its compute advantage in the mid-to-high-end market and compete with AI chip solutions from rivals such as MediaTek. This technology may accelerate the rollout of native AI applications, but actual power consumption and thermal performance still need to be observed.
In-Depth Analysis
Liu Gong believes this move is Qualcomm’s attempt to seize the high ground in on-device AI. Compared with cloud AI, running a 30-billion-parameter model locally is not easy, but once achieved, the benefits in response speed and privacy are tangible. If other chipmakers fail to keep up, Qualcomm could further consolidate its grip on the high-end market. However, do consumers really need such a large parameter count? Liu Gong is more concerned about whether this capability can translate into everyday functions, rather than remaining just a launch-event talking point. The next step is to see which smartphone maker can use this chip to deliver an AI experience that users rely on every day.
Perspectives
Further Thoughts
- New challenges that on-device AI large models pose to smartphone hardware power consumption and memory design
- How Qualcomm’s AI chip strategy will respond to the catch-up efforts of competitors such as MediaTek
- What locally running mixture-of-experts models means for the application ecosystem and developers
Source and Original Article
This news item is based on TechCrunch AI (published on September 22, 2026, at 20:00:00). This site provides Chinese summaries and commentary on overseas AI news; the original article is owned by its original author.
Daily aggregation of the latest overseas AI news and in-depth perspectives. Bookmark this site so you never miss an important signal; return to homepage for more.