华西证券研报表示，近期中国AI大模型迎来"春节档"爆发，技术趋势已从通用聊天工具全面转向垂直生 产力工具与真实Agent落地，重点体现在工业级的视频生成（如Seedance 2.0的多模态叙事与音画同 步）、工程级编程（如智谱GLM-5的跨文件重构与系统工程能力）以及消费办公场景。当前时点，面 临近期的流动性担忧以及海外OpenAI资本开支预期的变化，反映市场对于当前AI发展逐渐聚焦于成本 效益比率，预计市场仍将保持震荡，板块建议相对谨慎，中性配置。目前AI发展阶段仍处于Scale up和 Scale out的关键加速期，伴随相关算力芯片的供应体系逐渐丰富，应用发展对于token需求量仍在加速增 加，相关底层算力基建仍然还在扩张期。仍坚定看好Scale up的柜内光升级，电源、液冷等产业领域向 国产厂商升级机遇，国产算力包括芯片、交换机、服务器等需求放量，以及CSP厂商资本开支加速带来 的算力租赁和AIDC相关市场机遇。 ...

Group 1 - The AI industry is experiencing rapid advancements, with large models continuously iterating and increasing demand for computing power [3][18][29] - Major AI companies are transitioning from a "low-cost/free user acquisition" phase to a "high-quality paid" phase, significantly boosting computing power demand [4][9][24] - During the holiday period, US AI computing stocks performed exceptionally well, with companies like Lumentum, Corning, and Coherent reaching historical highs [12][28][29] Group 2 - Anthropic launched the Claude Sonnet 4.6 model, which shows significant improvements in programming and reasoning capabilities, becoming the default model for its platforms [4][19] - Google's Gemini 3.1 Pro model outperformed several competitors in benchmark tests, achieving a score of 77.1% on the challenging ARC-AGI-2 test, doubling its predecessor's performance [6][21] - Zhizhu AI released the GLM-5 model, featuring a parameter scale of 744 billion and training data of 28.5 trillion tokens, achieving top scores in open-source model benchmarks [7][22] Group 3 - MiniMax introduced the M2.5 model, designed for agent scenarios, demonstrating significant improvements in programming tasks and cost efficiency [9][24] - Alibaba's Qwen3.5-Plus model outperformed similar open-source models, offering competitive pricing and extensive language support [10][25] - The domestic AI market saw a surge in user engagement during the Spring Festival, with significant increases in active users for various AI applications [12][27]

David Luan, who led Amazon's San Francisco-based AGI Lab and oversaw one of its most important agentic AI initiatives, is leaving less than two years after joining the tech giant through an acqui-hire deal involving him and other leaders from AI startup Adept. Luan announced his exit Tuesday in a LinkedIn post, saying he will leave at the end of the week "to cook up something new.†"There's incredible work to be done at Amazon and opportunities for me to take on more areas. But with AGI so close, I decided ...

一场"贼喊捉贼"的年度大戏，正在中美AI竞赛的舞台上拉开帷幕。 2月23日，一则由TechCrunch爆出的新闻，瞬间点燃了全球AI领域的舆论场。 报道称，以"AI安全"为金字招牌的明星创业公司Anthropic，公开指控三家国内顶尖AI大模型公司—— DeepSeek、Moonshot AI和MiniMax，通过系统性的欺诈手段，利用其旗下的Claude模型来训练和优化 自己的模型。 根据Anthropic公布的数据，这次"工业级"的窃取行为规模惊人：三家公司共创建了超过2.4万个虚假账 户，与Claude模型进行了约1600万次交互。Anthropic声称，这些行为旨在通过"蒸馏"，系统性地提取 Claude的核心能力，包括推理、编程和工具使用等，从而"抄近道"提升自身模型的性能。 当一个曾经的"大盗"开始指责别人"偷窃"时，我们不禁要问：这究竟是一场维护知识产权的正义之举， 还是在激烈竞争的焦虑之下，一个"前科犯"试图通过攻击对手来掩盖自身"原罪"的公关策略？ Anthropic这个硅谷的AI宠儿，究竟是技术创新的捍卫者，还是北美科技圈最大的"偷子"？ "捉贼"的Anthropic，忘了自己怎么发的家？ ...

责任编辑：王永生 责任编辑：王永生 埃隆-马斯克的人工智能初创公司xAI指控竞争对手山姆-奥特曼（Sam Altman）的OpenAI窃取其商业机 密，加州一名联邦法官周二驳回了xAI的诉讼。旧金山的美国地区法官Rita Lin表示，xAI可以重新提起 诉讼，但目前未能指控OpenAI有任何不当行为。今年9月提起的诉讼，声称xAI前员工在离职后前往 OpenAI就职时带走了与其Grok聊天机器人相关的源代码和其他机密信息。"值得注意的是，诉讼中没有 对OpenAI本身的行径提出指控，" Lin说。"xAI没有指控任何事实表明OpenAI诱使xAI的前员工窃取xAI 的商业机密，也没有指控这些xAI的前员工在受雇于OpenAI后使用了任何窃取的商业机密。"Lin曾在1 月份的意见书表示她很可能会做出有利于OpenAI的裁决。她要求xAI在3月17日前提交修改后的诉状。 xAI 还另外起诉了前工程师李学勤，指控他将商业机密提供给了 ChatGPT 制造商。在该案中，李学勤 被阻止 与OpenAI分享xAI的技术，但OpenAI表示，李学勤从未为该公司工作过，也从未获取或使用过 xAI的任何机密。xAI的发言人和律 ...

原标题：AI"变冷"风波是镜子更是考题 当前国内AI赛道上，不同企业选择了不同发展路线。有的深耕C端交互体验，契合大众日常沟通需求； 有的长期聚焦算法优化等技术能力，筑牢国产大模型底层技术根基。不同路线并无高下，它们共同构成 国产AI发展的多元生态。在国产AI领域，我们既需要沉下心"死磕硬核技术"的企业，也需要静下心优化 日常体验的产品。专业场景看重效率，日常使用则在意温度，不同定位服务不同需求。 而用户的集体"吐槽"，恰恰反映了产品风格变化与用户原有预期的错位。"吐槽"DeepSeek"变冷淡"的用 户，核心诉求不是让AI只做"会聊天的贴心朋友"，而是别把原本成熟好用的功能改没了。因为见过它的 好，大家才会在意它的变。这些看似负面的反馈，本质上是对产品的高期待。当靠技术创新"横空出 世"的DeepSeek弱化情绪价值时，我们也由衷期待，它能在硬核技术上带来新的惊喜。 这场AI"变冷"风波也给行业提了个醒：技术迭代，终究要以人为本。企业在实施可能影响用户核心使用 体验的调整时，不妨通过官方渠道提前告知用户调整逻辑、周期与预期目标，最大程度减少用户的预期 落差与使用不适感。毕竟，用户对一款产品的依赖和信任来之不 ...

Core Insights - The report by Frost & Sullivan indicates a significant increase in the daily usage of enterprise-level large models in China, with a rise to 37 trillion tokens in the second half of 2025, marking a 263% increase from 10.2 trillion tokens in the first half [1][3]. Group 1: Market Dynamics - The primary drivers for enterprises using large models shifted from enhancing product performance and customer experience in the first half of 2025 to improving operational efficiency and R&D effectiveness in the second half [3]. - The market for enterprise-level large models is rapidly consolidating around leading vendors, with the top three models (Qwen, Doubao, and DeepSeek) accounting for over 70% of daily token consumption in the second half of 2025, compared to less than half in the first half [3][4]. Group 2: Vendor Performance - Alibaba Cloud's Qwen model saw its market share nearly double to 32.1% in the second half of 2025, with a growth rate of 14.4% [3][4]. - The report highlights that enterprises are increasingly favoring fewer vendors to reduce long-term operational costs related to interface maintenance and version management, creating higher migration barriers for leading vendors [4]. Group 3: Open Source Trends - There is a notable increase in enterprises' willingness to adopt open-source models, with the proportion of closed-source users planning to add open-source options doubling from 22.6% in the first half to 48.5% in the second half of 2025 [4]. - Chinese companies accounted for 90.2% of the global new open-source large models introduced in the second half of 2025, significantly outpacing international competitors [4]. Group 4: Competitive Landscape - The competition among domestic open-source models has expanded beyond single-point performance to include size, deployment flexibility, and cost control [5]. - Since its launch in 2023, Qwen has released over 400 models, achieving over 1 billion downloads and extending its language capabilities to 201 languages, establishing itself as the leading open-source model globally [5].

Core Insights - The Chinese AI industry experienced a significant surge during the Spring Festival, with major companies like ByteDance, Alibaba, and Tencent actively engaging in AI-related activities while most workers took a break [2][4] - The competition led to record-breaking user interactions, with ByteDance's Doubao achieving 1.9 billion interactions, Alibaba's Qwen serving over 130 million users, and Tencent's Yuanbao facilitating 3.6 billion lottery draws [4][6] - Despite the impressive metrics, the industry faces challenges related to user retention and operational pressures, highlighting a disconnect between user engagement driven by incentives and genuine usage of AI functionalities [6][7] Company Activities - ByteDance's Doubao served as the exclusive AI cloud partner for the CCTV Spring Festival Gala, managing 1.9 billion interactions and ensuring the stability of 100,000 live lottery draws [2] - Alibaba's Qwen team launched the open-source Qwen 3.5 and integrated AI capabilities into consumer platforms like Taobao and Ele.me, resulting in over 130 million users engaging with the service during the festival [2] - Tencent's Yuanbao distributed 1 billion cash red envelopes and facilitated 3.6 billion lottery draws, requiring continuous technical support to maintain product stability [3] Industry Challenges - The rapid user engagement during the festival exposed operational weaknesses, such as the inability of Zhiyu's GLM-5 to handle user demand, leading to phased access and user complaints [4] - The industry is transitioning from a focus on technical achievements to addressing the challenges of high-frequency user interactions and operational reliability [4][9] - The temporary surge in user activity raises concerns about long-term retention, as users attracted by incentives may not continue to engage with AI services once promotions end [6][7] Future Outlook - The industry is expected to shift from a "money-burning frenzy" to a "commercialization elimination race," emphasizing the need for teams that can balance technology, operations, and business strategies [9] - The true value of AI will be determined by its ability to meet real user needs beyond the temporary spikes in engagement seen during promotional events [9]

The head of Amazon's artificial general intelligence lab is leaving the company less than two years after joining through a so-called acqui-hire deal of his startup Adept. David Luan announced his departure Tuesday in a LinkedIn post, saying he will exit at the end of the week "to cook up something new." Amazon tapped Luan in December 2024 to oversee its newly created AGI lab, based in San Francisco. The lab is focused on "long-term research bets," including developing "useful AI agents." Last year, the gro ...

Core Viewpoint - Tempus AI achieved positive adjusted EBITDA for the first time in Q3 at $1.5 million, with Q4 revenue reaching $367 million, an 83% year-over-year increase. Despite beating revenue consensus, shares declined 22% since January due to widening net losses [1]. Financial Performance - Q4 revenue was pre-announced at $367 million, reflecting an 83% year-over-year growth, with diagnostics revenue at $266 million, up 121% year-over-year [1]. - Data and Applications segment contributed $100 million, a 25% increase, while the Insights segment grew by 68% [1]. - Total contract value exceeded $1.1 billion, and net revenue retention was 126%, indicating strong customer base expansion [1]. Profitability and Guidance - Q3 net loss widened to $80 million from $75.8 million a year earlier, despite higher revenue [1]. - Management guided for slightly positive adjusted EBITDA for the full year, absorbing integration costs from the Paige acquisition [1]. - 2026 guidance is crucial, with expectations for the foundation model to produce its first versions in Q1 2026 and ongoing MRD reimbursement progress [1]. Market Reaction and Stock Performance - Tempus shares have underperformed, down approximately 22% since the preliminary filing, despite strong revenue growth and achieving a key profitability milestone [1]. - Analyst consensus includes 7 Buy ratings and 7 Hold ratings, with an average price target of $86.71, suggesting potential upside if execution remains strong [1].