Investment Rating - The report maintains a "Buy" rating for several companies in the AI and computing sectors, including Weike Technology, Nengke Technology, Hehe Information, and Maixinlin [13]. Core Insights - The AI application sector is experiencing significant growth, highlighted by the successful IPO of MiniMax, which raised 5.54 billion HKD and achieved a market capitalization of 898 billion HKD on its first day of trading [6][45]. - Nvidia's new VeraRubin architecture has revolutionized computing power, achieving up to five times the performance in AI inference tasks compared to the previous Blackwell architecture, with a tenfold reduction in inference costs per token [5][34]. - MiroMind's MiroThinker1.5 model has demonstrated superior performance in international benchmarks, achieving significant cost efficiency and redefining the growth path for large models [4][24]. Summary by Sections Computing Power Dynamics - The report notes stable pricing in computing power leasing, with significant advancements from MiroMind's MiroThinker1.5 model, which features 30B and 235B parameter variants that excel in benchmark tests [4][18]. - The token consumption for the week of January 5 to January 11, 2026, reached 6.43 trillion, marking a 15.44% increase from the previous week [18][19]. AI Application Dynamics - NotionAI's weekly traffic increased by 25.06%, indicating strong user engagement in the AI application space [32]. - The report highlights the performance of Nvidia's VeraRubin architecture, which integrates CPU, GPU, and networking components for optimized AI computing, significantly enhancing training and inference capabilities [5][34]. AI Financing Trends - MiniMax's IPO is noted as the largest for an AI model company globally, with a remarkable oversubscription rate of 1837 times for public offerings, reflecting strong investor confidence [6][45]. - The company has successfully completed multiple funding rounds, raising a total of 1.5 billion USD from 30 institutional investors over four years [46][47]. Investment Recommendations - The report suggests focusing on companies like Maixinlin, Weike Technology, Hehe Information, and Nengke Technology, which are positioned to benefit from the expanding AI and computing markets [10][58]. - MiniMax's successful market entry is seen as a significant indicator of the AI application sector's commercial viability and growth potential [57].

当地时间1月5日晚，身穿标志性黑色皮夹克的黄仁勋站在拉斯维加斯CES 2026的舞台上，这位NVIDIA 掌门人正式发布了由六颗全新芯片组成的Rubin平台，宣布将在2026年下半年向合作伙伴供货。更重磅 的是代号"Alpamayo"的自动驾驶AI软件——黄仁勋称之为"机器人技术的ChatGPT时刻"。 关于自动驾驶的安全哲学，老黄也给出了他的答案："最好永远不要有人车交接。所以即使你不承诺 L4，也应该具有L4的能力。" 两天的CES之行，黄仁勋展现的不仅是NVIDIA的技术路线图，更是他对AI时代的宏大叙事：从芯片到 生态，从数据中心到机器人，从当下的几千亿到未来的百万亿。这场百万亿美元的技术革命，NVIDIA 显然不想缺席任何一个战场。 关于AI泡沫 问题：2025年10月你给出了一个当时令人震惊的数字，说未来四五个季度会卖出去大约5000亿美元的数 据中心。过去几天你一直在说需求非常强劲。我们应该期待更高的数字吗？ 黄仁勋：我希望不要定期更新这个数字。 但我要说的是，自那以后发生了许多新进展，应该会提高我们对这个数字的预期。去年我们有一些非常 激动人心的消息。我们一直是模型构建者、AI构建者的好伙伴。 ...

刚刚的CES 2026上，老黄带着Vera Rubin超算架构向全世界走来！Rubin的推理性能比Blackwell提升了5倍，训练性能提升3.5倍，成本降低10倍，已经大 规模投产，将于2026下半年面世。没有新显卡的昨夜，老黄表示all in AI！ 天空一声巨响，全新版本的「皮衣老黄」闪亮登场。 在本次CES演讲中最为激动人心的瞬间，就是英伟达全新一代芯片架构——Vera Rubin正式登场！ 全球AI算力告急？老黄霸气回应：Vera Rubin已全面投产。 这是新一代的算力怪兽，也是对上一代霸主Blackwell的降维打击—— 英伟达在X上的一纸公告，彻底击碎了「攒机党」最后的幻想：CES 2026将没有任何新GPU发布。 这意味着，英伟达自2021年以来连续五年在CES发布新硬件的传统，就此终结。 传闻已久的RTX 50 Super系列，受困于GDDR7显存的「产能地狱」，大概率已经胎死腹中。 Rubin炸裂登场，6颗芯片，10倍推理，AI超算变工厂 推理Token成本直接暴降10倍，算力性能狂飙5倍。 就连训练MoE模型所需的GPU数量，也直接减少了4倍。 曾经，Blackwell终结了Hoppe ...