Core Viewpoint - The introduction of Google's "TurboQuant" algorithm is expected to significantly impact the demand for memory chips due to its ability to compress data during AI inference, potentially reducing memory usage to one-sixth and increasing processing performance by up to 8 times [2][3]. Group 1: TurboQuant Algorithm - Google Research announced the "TurboQuant" algorithm, which compresses data during inference, leading to a reduction in memory usage and an increase in processing performance [3]. - The algorithm is described as "Accelerator Friendly," making it easier for GPUs to perform AI processing efficiently [6]. - TurboQuant is a universal method applicable to any data type or format, maintaining computational accuracy while improving data conversion efficiency [6]. Group 2: Impact on Semiconductor Industry - The demand for high-bandwidth memory (HBM) has surged with the rise of AI, leading to supply shortages from major semiconductor companies like SK Hynix, Samsung Electronics, and Micron Technology [7]. - If TurboQuant is widely adopted, the expected memory requirements for AI applications may be lower than anticipated, causing a decline in stock prices for memory manufacturers by approximately 10% following the announcement [7]. - The market is closely monitoring whether Google's new technology will alter the AI inference landscape, reminiscent of past fluctuations in AI-related stocks [9].

谅解备忘录备忘录记录战略合作的主要意向，据此，公司将于香港成立一间全资附属公司，以资助、建 设及营运采用数维科技授权或转授权之kodon技术的新一代数据中心。Kodon技术被描述为一种数据表 示技术，能够将各种格式的档案压缩为一个紧凑的「Kodon」(字元序列)，目标压缩至原始档案大小的 1%以下，惟受限于技术参数及验证结果。该技术可在有限实体空间内实现极高逻辑储存容量。订约双 方拟迅速展开技术、法律及商业尽职调查，并就最终协议进行磋商，以记录数据中心的授权、建设及营 运事宜。 格隆汇11月3日丨中国供应链产业(03708.HK)宣布，于2025年11月3日，其已与数维科技有限公司(「数 维科技」)签订谅解备忘录。 根据备忘录，公司计划于香港成立一间全资附属公司，以资助、建设及营运有关设施，而数维科技则拟 就项目授予或安排长期使用该技术的权利，惟须待技术尽职调查完成及最终协议协商后，方可落实。 ...