CES巨幕上，老黄的PPT已成中国AI的「封神榜」。DeepSeek与Kimi位列C位之时，算力新时代已至。 万众瞩目的2026 CES科技盛宴上，一张PPT瞬间燃爆AI圈。 老黄主旨演讲上，中国大模型Kimi K2、DeepSeek V3.2，以及Qwen赫然上屏，位列全球开源大模型前列，性能正在逼近闭源模型。 这一刻，是属于中国AI的高光时刻。 另外，OpenAI的GPT-OSS和老黄自家的Nemotron，也做了标注。 而且，DeepSeek-R1、Qwen3 和 Kimi K2 代表着MoE路线下顶级规模的尝试，仅需激活少量参数，大幅减少计算量和HBM显存带宽的压力。 在下一代Rubin架构亮相的核心环节上，老黄还选用了DeepSeek和Kimi K2 Thinking来秀性能。 在Rubin暴力加成下，Kimi K2 Thinking推理吞吐量直接飙了10倍。更夸张的是，token成本暴降到原来的1/10。 这种「指数级」的降本增效，等于宣告了：AI推理即将进入真正的「平价时代」。 另外，在计算需求暴涨这页PPT上，480B的Qwen3和1TB的Kimi K2成为代表性模型，验证了参数规模每年以十倍 ...

被称为科技春晚的国际消费电子产品展览会(CES)于1月6日正式开幕。 作为英伟达新年战略和新品发布的重要窗口，本次英伟达CEO依然身着去年同款亮面鳄鱼皮衣登场，发表了他在2026年的首场演讲。 在CES开始前，英伟达在社交媒体发布称"(CES2026)不会发布新款GPU。"这也是英伟达五年来首次不在CES发布新款GPU产品。 相比去年CES官宣了RTX50系产品，本次演讲中，黄仁勋把重点放在了新一代计算平台和英伟达在物理AI领域的进展，包括自动驾驶和机器人，相关开源 模型和工具等。 演讲开始。黄仁勋表示过去十年投入的约10万亿美元计算资源，正在被彻底现代化。 但这不仅仅是硬件的升级，更多的是软件范式的转移。 他特别提到了具备自主行为能力(Agentic)的智能体模型，并点名了Cursor，彻底改变了英伟达内部的编程方式。 随后，黄仁勋对2025年开源社区给予了高度评价。他表示，去年DeepSeek的突破让全世界感到意外，它作为第一个开源推理系统，直接激发了整个行业 的发展浪潮。 而在介绍开源生态时，黄仁勋的PPT后出现了三家中国模型的名字，分别是月之暗面的Kimi K2，深度求索的DeepSeek V3.2和 ...

Market Outlook - The A-share market in 2025 experienced a structural trend led by industrial trends, with sectors like humanoid robots, new consumption, innovative pharmaceuticals, artificial intelligence, and commercial aerospace performing well, contributing to the theme of "high-quality development" [1][11] - For 2026, the AI industry is expected to remain a core narrative in the market [1][11] AI Models - AI models are anticipated to enter a new phase of innovation, with Google’s Gemini 3.0 demonstrating the potential of pre-training technology, while China's DeepSeek V3.2 showcases unique value in post-training paths [2][12] - In 2026, collaboration among domestic and international large model developers is expected to intensify, focusing on computing power deployment and algorithm optimization [2][12] AI Infrastructure - Since 2023, global tech giants have maintained strong investments in AI, despite concerns about a potential "bubble," with a focus on accelerated computing as the core of AI infrastructure investment [3][13] - The investment in AI infrastructure is characterized by long-term planning, and it is essential to allow sufficient time for development and innovation [3][13] AI Applications - AI applications have made progress over the past three years but remain below the optimistic expectations of the market, leading to discussions about a "bubble" [4][14] - Companies like OpenAI and Anthropic are showing strong potential for AI commercialization through rapid growth in annual recurring revenue (ARR) [4][14] Product Management Strategy - The company will closely track industry trends and continuously optimize its product management system to create long-term sustainable value for investors in 2026 [6][15] Global Perspective - While focusing on the Chinese stock market, the company will maintain close monitoring of global tech companies and enhance its research framework based on global supply chains [7][16] Value Investment Perspective - The company prefers investing in companies with sustainable growth, particularly those with global competitiveness and the ability to create social value [8][17] Adaptability - The company aims to embrace new market opportunities and challenges with an open and pragmatic mindset, believing that the market will continue to present structural opportunities in 2026 [10][18] Risk-Return Balance - In 2026, the company will prioritize risk control while pursuing investment returns, focusing on the performance realization capability of enterprises [19]

Group 1 - The report emphasizes that global liquidity is continuously improving, which is expected to enhance equity asset returns, particularly in the context of AI-related software and edge applications [1][7][11] - The report suggests that the growth space for small-cap and growth sectors is likely to expand due to synchronized global monetary policy easing [7][11] - The report highlights the significant advancements in AI models, particularly with the release of Gemini 3.0 and GPT-5.2, which are expected to drive substantial commercial opportunities in AI applications [21][28][39] Group 2 - The report identifies key companies to watch, including Blue Sky Technology and Spring Wind Power, which are expected to benefit from AI-related growth [8][65] - It also mentions companies like Jieshun Technology and Kaige Precision Machinery, which are positioned to capitalize on AI-driven product releases and innovations [7][65] - The report notes that the valuation of small-cap stocks remains low, with PE ratios for the National Index 2000 and the Growth Enterprise Market Index at 57.1x and 41.21x respectively, indicating potential for further valuation recovery [65]

当年初DeepSeek一夜爆红，打破原有大模型市场的格局，这一年就注定不平凡。2025年的中国大模型市场经历了 一场深刻的"价值回归"，技术突破的边际效应减弱，一场围绕真实需求、可持续商业模式与产业深度的"生存进 化"全面展开。"2025年是全球化AI应用的创业之年。"顺福资本创始人、行行AI董事长李明顺总结道。 在此背景下，国内"AI六小虎"加剧赛道分化，零一万物和百川智能放弃超大模型训练，在更加务实的商业化应用 赛道越走越远，阶跃星辰将智能终端Agent作为⼤模型技术落地的关键发⼒点，在终端Agent领域取得突破，月之 暗面开始重视商业化，任命曾经的投资人为总裁，智谱和MiniMax则作为商业化的佼佼者率先成功闯关二级市 场。 DeepSeek的"起伏" 2025年初，一场由东方掀起的AI浪潮席卷全球应用市场。1月27日，来自中国的人工智能公司DeepSeek一举登顶 美国苹果商店免费应用下载榜首，将长期盘踞头部的ChatGPT暂时拉下王座，之后又迅速演变为一场全球性的现 象级传播——DeepSeek的名字随之刷屏各国社交网络，成为开年最受瞩目的科技焦点。 热度并未止步于年初的榜单登顶。整个上半年，Dee ...

文/石飞月 当年初DeepSeek一夜爆红，打破原有大模型市场的格局，这一年就注定不平凡。2025年的中国大模型 市场经历了一场深刻的"价值回归"，技术突破的边际效应减弱，一场围绕真实需求、可持续商业模式与 产业深度的"生存进化"全面展开。"2025年是全球化AI应用的创业之年。"顺福资本创始人、行行AI董事 长李明顺总结道。 在此背景下，国内"AI六小虎"加剧赛道分化，零一万物和百川智能放弃超大模型训练，在更加务实的商 业化应用赛道越走越远，阶跃星辰将智能终端Agent作为⼤模型技术落地的关键发⼒点，在终端Agent领 域取得突破，月之暗面开始重视商业化，任命曾经的投资人为总裁，智谱和MiniMax则作为商业化的佼 佼者率先成功闯关二级市场。 DeepSeek的"起伏" 2025年初，一场由东方掀起的AI浪潮席卷全球应用市场。1月27日，来自中国的人工智能公司DeepSeek 一举登顶美国苹果商店免费应用下载榜首，将长期盘踞头部的ChatGPT暂时拉下王座，之后又迅速演变 为一场全球性的现象级传播——DeepSeek的名字随之刷屏各国社交网络，成为开年最受瞩目的科技焦 点。 热度并未止步于年初的榜单登顶。整 ...

Group 1 - ByteDance's Doubao model has surpassed 50 trillion daily tokens usage, ranking first in China and third globally, with over 100 companies using more than 1 trillion tokens on the platform [1][24][30] - Tencent has announced a restructuring of its AI model development architecture, establishing new departments to enhance its AI capabilities, with former OpenAI researcher Yao Shunyu appointed as Chief AI Scientist [1][9][32] - The competition among major internet companies in the AI model sector is intensifying, indicating a sustained demand for computing power [1][10][30] Group 2 - Google's Gemini 3 Pro has made significant advancements in multimodal understanding and planning capabilities, excelling in tasks involving text, images, and other data types [2][25] - OpenAI's GPT-5.2 focuses on professional knowledge work, showing improved performance in complex document handling and data analysis, with a new evaluation system introduced to measure economic value [2][11][12] - The DeepSeek V3.2 series has achieved notable improvements through innovations like sparse attention mechanisms and extensive post-training, although it acknowledges limitations in pre-training [2][12][14] Group 3 - Micron Technology reported better-than-expected earnings, with all HBM production capacity sold out for 2026, and anticipates the HBM market to reach $100 billion by 2028 [3][15][26] - The demand for AI-driven storage solutions is surging, leading to a structural shift in production priorities, with data centers consuming significant memory resources [3][16][26] - The launch of ByteDance's Doubao mobile assistant marks a significant breakthrough in AI application, transitioning towards an agent-based interaction model [4][17][27]

Core Insights - Chinese technology stocks, including Tencent, Bilibili, NetEase, and PDD Holdings, have gained momentum following the U.S.-China trade agreement, with China meeting commitments such as terminating semiconductor investigations and resuming agricultural purchases [2] - SMIC achieved volume production of 5nm chips, marking a significant advancement in China's semiconductor manufacturing capabilities [3] - BYD's exports surged 326% year over year, with NEV penetration in China reaching 62% [4] - The humanoid robotics sector saw a 250% increase in investment deals, reflecting growing integration in manufacturing [6] - China's defense budget increased by 7.2% to $249 billion, with significant advancements in military technology [7] - The medical device market in China reached $172.9 billion, showing substantial growth and innovation [8] - China Railway Rolling Stock Corporation maintained a 56% global market share in rail, while Chinese shipyards secured 38% of new global LNG vessel orders [9] - The Politburo's announcement of a "moderately loose" monetary policy and Goldman Sachs raising GDP forecasts to 4.8% indicates a stabilizing economic environment [10] Company Summaries - Tencent Holdings reported record gaming sales of $10 billion internationally, with a 15% revenue growth and 43% surge in international gaming [12] - Bilibili turned profitable with a net profit of RMB469 million in Q3 2025, showing a 233% year-over-year increase in adjusted net profit [13] - NetEase's gaming revenues increased by 11.8% year over year to RMB23.3 billion, supported by a strong partnership with Blizzard [14] - PDD Holdings demonstrated a 9% revenue growth and 17% net income expansion, maintaining a strong financial position with RMB387 billion in cash reserves [15]

Group 1 - Xiaomi officially announced the open-source release of Xiaomi MiMo-V2-Flash, a MoE model with a total parameter count of 309 billion (15 billion activated), achieving top 2 in global open-source model benchmarks [1] - The model features innovations such as Hybrid attention architecture and multi-layer MTP inference acceleration, resulting in a code capability comparable to the closed-source model Claude 4.5 Sonnet, but at only 2.5% of its inference cost and with a 2x increase in generation speed [1] - Xiaomi MiMo-V2-Flash outperformed DeepSeek V3.2 and K2-Thinking in most evaluation benchmarks, reducing parameter count by 50% to 67%, and achieving low cost and high speed, with preliminary capabilities to simulate the world [1] Group 2 - The next generation of intelligent agent systems is envisioned not merely as "language simulators" but as true "intelligent agents" that understand and coexist with the human world [2] - There is a shift in agent execution capabilities from merely "answering questions" to "completing tasks," incorporating memory, reasoning, autonomous planning, decision-making, and execution abilities [2] - Unified multimodal perception is essential for understanding the physical world, which will enhance integration with smart devices like glasses [2]

证券研究报告 | 2025年12月14日 AI 赋能资产配置(三十三) DeepSeek 与 Gemini，谁更懂 A 股？ 核心结论：①大模型具备一定的技术分析基本功：不论是 DeepSeek V3.2 还 是 Gemini 3 Pro，都能够在合适的提示词指引下，面对有限的蜡烛图，完成 顶底分型划分、笔和线段的擘画，中枢的构建等。 ②对于"已成型"走势， 大模型具备一定的"技术分析"能力： DeepSeek 在语言组织与长文本生成 上表现良好，Gemini 能精准识别"中枢扩张"与"走势多义性"，对买卖 点的几何定义执行严格。 ③Gemini 在"易用性"上更占优：Nano Banana Pro 能够完成简易图形标注，在实际使用的便利性上略胜一筹。 如何测评 DeepSeek 与 Gemini 的技术分析能力？公平公正下的"四同原 则"。1）数据同源：所有测试均使用同一来源，即上证指数的标准化 OHLC 价格数据，并以相同格式提供给双方模型；2）提示词同构：向两个模型传 达完全相同的任务指令、规则背景与输出格式要求； 3）环境同期：所有测 试任务将在相近的时间段内依次完成，以尽可能减少模型潜在更新带来的版 ...