Guohai Securities·2026-03-08 15:10Investment Rating - The report maintains a "Recommended" rating for the computer industry, indicating a favorable outlook for the sector [1]. Core Insights - The upcoming NVIDIA GTC 2026 conference is expected to unveil a new AI inference chip system, integrating the LPU architecture from NVIDIA's acquisition of Groq, aimed at optimizing inference performance [5]. - The global demand for computing power is shifting from training to inference, with significant implications for the market structure and opportunities for domestic chip manufacturers [8]. - The report emphasizes the acceleration of domestic infrastructure development for AI inference, suggesting a broad growth opportunity for related AI chip, hardware, and service companies [8]. Summary by Sections Recent Performance - The computer industry has shown mixed performance over the past year, with a 1-month decline of 0.6%, a 3-month increase of 4.4%, and a 12-month decline of 1.3% [3]. Upcoming Events - The NVIDIA GTC 2026 conference will take place from March 16 to 19 in San Jose, California, featuring key advancements in AI technologies, including new AI acceleration chips and systems [5]. Technological Developments - The Groq LPU is designed for inference acceleration, utilizing SRAM for model parameter storage, which significantly enhances data processing speed compared to traditional GPU architectures [6][7]. Market Trends - By 2032, the proportion of training costs in large cloud computing companies' data center expenditures is expected to drop from over 40% to around 14%, indicating a clear shift towards inference [8]. - Major companies are securing large-scale orders for inference-optimized chips, highlighting the growing demand for such technologies [8]. Investment Strategy - The report suggests a focus on specific stocks within the AI chip sector, including companies like Haiguang Information, Cambrian, and others, as well as those involved in servers and liquid cooling technologies [8].