B[Lyon: Optimistic about Korean semiconductors’ key position in Agentic AI supply chain Rating SK Hynix “highly confident of outperforming the market”] On September 22, Lyon published a research report stating that agent-based AI transforms a single inference of a large AI model into a long-term tool usage cycle that requires multiple rounds of execution for each task, putting pressure on memory capacity, bandwidth and rack-scale integrity, far beyond simple Computational problems require the cooperation of high-bandwidth memory (HBM), low-power double data rate (LPDDR)/small compressed add-on memory module (SOCAMM), enterprise-class solid-state drive (eSSD)/high-bandwidth flash memory (HBF), advanced packaging, testing, and front-end shrinkage. The bank pointed out that South Korea has an extensive end-to-end layout in key AI semiconductor processes such as AI memory, bonding, substrate, testing and front-end processing. The bank believes that with the support of structural Agentic AI demand and tight supply, K-enablers' fundamentals have improved and their positioning has been strengthened. SK Hynix, Samsung Electro-Mechanics and EO Technics have all been given a "highly confident outperform" rating, and Samsung Electronics, Hanmi Semiconductor, HPSP Eugene and ISC have been given an "outperform" rating. B[Lyon: The rise of China’s Agentic AI, with Tencent and Alibaba as the top choice] On September 21, Lyon published a research report stating that Agentic AI is moving from experiments to enterprise workflows and becoming a new generation of infrastructure layer. Chinese AI models are rapidly narrowing the performance gap and seizing global share with cost-effectiveness; enterprises are shifting from purely pursuing model capabilities to taking into account performance, cost and deployment flexibility. Multi-model adoption will become a future trend. It is expected that ultra-large-scale cloud players in the chip, model, computing power and application layers can seize the greatest opportunities. Lyon expects that the surge in AI applications will ignite the next round of global cloud super cycle. The global public cloud market will double to US$2.3 trillion by 2028, bringing annual revenue opportunities of more than US$1 trillion. The market is more optimistic about full-stack cloud and AI providers because they can form a strong moat by combining their own chips, models, cloud infrastructure, development tools and applications. It is expected that areas such as software, video streaming, in-store services, recruitment and online travel will be relatively fragile, with Tencent and Alibaba being the first choices. The bank gave Tencent a "highly confident outperform" rating and Alibaba an "outperform" rating.
— All loaded —
