Small and On-Device Models: Phi, Gemma, Apple — 2026-09-21
Samsung's Galaxy S26 series is driving a resurgence for the Exynos chip with on-device AI capabilities, while Qualcomm details next-gen NPU support for 30B parameter models. Meanwhile, Apple's ecosystem continues to refine Core ML and MLX workflows for small language models like Phi and Gemma.
Small and On-Device Models: Phi, Gemma, Apple — 2026-09-21
Top developments
Samsung Exynos Regains Market Share with On-Device AI Focus
Samsung Electronics' mobile application processor, "Exynos," has recorded its highest global market share since 2024, reaching 9%, largely driven by the strong sales of the Galaxy S26 series. The rebound is attributed to simultaneous adoption in both premium and mid-range segments, where on-device AI features are becoming a key differentiator. This shift highlights the growing importance of local inference capabilities in Samsung's silicon strategy as it competes with Qualcomm and MediaTek.
Qualcomm Unveils Next-Gen Hexagon NPU for 30B Parameter Models
Qualcomm has released technical details for its upcoming Hexagon NPU, which features a new "Element Accelerator" and 50% more shared memory. The upgraded architecture supports the execution of 30-billion-parameter Mixture-of-Experts (MoE) AI models directly on smartphones, significantly expanding the complexity of models that can run locally without cloud dependency. This development pushes the boundaries of what is possible for on-device generative AI in the mobile sector.
Googlebook Pre-Orders Open with Gemini Integration
Google's new "Googlebook" laptops opened pre-orders on September 21 at $999, featuring deep integration with Gemini AI and a "Magic Pointer" feature. These devices are positioned as direct competitors to Microsoft's Copilot+ PCs, leveraging on-device Gemini variants to handle tasks locally. The launch marks a significant push by Google to standardize AI-first laptop experiences outside of the Windows ecosystem.
DeepX Partners with US Smart Glass Leader for Low-Power AI
Korean semiconductor firm DeepX is conducting Proof of Concept (PoC) collaborations with leading US smart glass manufacturers to validate its "K-NPU" technology. The partnership aims to overcome power and heat limitations in wearable devices, enabling ultra-low-power physical AI computations. This move signals a broader industry trend toward specialized, low-power NPUs for edge devices beyond smartphones.
Local view
Korean media outlets are closely monitoring the "One UI 9" update rollout for Galaxy devices, which introduces enhanced Galaxy AI Creator tools. The update includes features that allow users to edit video footage to center specific individuals and proactively suggest reservation information during chat interactions. This reflects Samsung's strategy to embed proactive, on-device AI assistance directly into the user interface layer of its devices.
Context & numbers
- Exynos Market Share: Samsung's Exynos processors achieved a 9% global market share in 2026, the highest level since 2024.
- Qualcomm NPU Capacity: The new Hexagon NPU architecture supports up to 30 billion parameter models via MoE structures.
- Googlebook Pricing: The new AI-integrated Googlebook laptops are priced starting at $999 for pre-orders.
On the radar
- One UI 9 QA Checklist: A detailed quality assurance checklist for Korean users updating to One UI 9 was published, focusing on camera, audio, and battery verification post-update.
- DeepX Policy Discussions: South Korea's Strategy Economy Advisory Committee held field discussions at DeepX to explore policies supporting domestic AI semiconductor adoption and material component equipment support.
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
