IMO金牌已经"过时"了。 基于Gemini 3 Deep Think的谷歌数学智能体Aletheia在更难的挑战赛FirstProof中拿下的最佳成绩。 在公布的完整成绩单中，10道题Aletheia全程0人工参与解出6道，其中5题专家全票通过，还有一题拿到了5/7的通过率。 | | Aletheia | Expert Evaluation | | --- | --- | --- | | | (best of 2) | (correct/total) | | P1 | N/A | | | P2 | Correct | 4/4 | | РЗ | N/A | | | P4 | N/A | | | P5 | Correct | 4/4 | | be | N/A | | | P7 | Correct | 3/3 | | P8 | Correct? | 5/7 | | P9 | Correct | 4/4 | | P10 | | 2/2 | FirstProof是由来自哈佛、斯坦福等名校的11位顶尖数学家联手打造的一套专门验证AI独立科研能力的数学题集。 10道题全网无迹可循，没法儿背答案作弊，连陶哲轩都转发说这事儿 ...

苹果前基础模型团队负责人、上海交大校友庞若鸣，被扎克伯格天价挖去Meta仅仅半年，再度跳槽！ 最新消息，他已于上周离职并加入OpenAI。此消息获OpenAI发言人证实。 OpenAI对其觊觎已久，过去数月一直在全力挖角。 据悉，庞若明当初加盟Meta时，手握的薪酬方案总价值超2亿美元（约14亿元人民币），分多年兑现，具体金额与完成特定里程碑目标直接挂钩。 他的出走，正值Meta人工智能团队离职潮持续发酵之际。 庞若鸣其人 公开资料显示，庞若鸣本科毕业于上海交大，硕博分别就读于南加州和普林斯顿。 庞若明之外，Meta超级智能实验室开发者平台产品负责人Mat Velloso，同样在短暂任职后宣布离职。他于去年7月从谷歌DeepMind加入Meta，不久前在 领英上公布了离职消息。 Geoffrey Hinton核心弟子、Meta生成式人工智能研究副总裁、卡内基梅隆大学计算机科学教授、苹果首任AI总监Russ Salakhutdinov，在Meta任职已满两 年，就在昨天也官宣离职了。 此外，Meta前首席人工智能科学家Yann LeCun，去年离职事件当时更是闹得沸沸扬扬。 一毕业直接加入谷歌，在谷歌担任了15 ...

在一项涉及物理AI的重大重组中，Alphabet正将其机器人软件子公司Intrinsic重新划归至另一子公司谷 歌旗下。此举结束了Intrinsic近五年的独立运营地位。这一整合正发生在全球科技巨头竞相将AI，集成 到从仓库机器人到制造自动化等物理系统之中的背景下。 在早年间于Alphabet的"登月工厂"X实验室进行了五年半的技术开发后，Intrinsic于2021年7月成为了 Alphabet"其他投注"(Other Bets)部门的一家独立子公司。然而，随着Alphabet周三的一纸公告， Intrinsic将重新被并入谷歌内部，此次重组也凸显了这家搜索巨头对物理AI领域的重视程度。 这一时机的选择显然绝非巧合——随着亚马逊部署成千上万的仓库机器人，以及特斯拉推进其Optimus 人形机器人平台，谷歌似乎也正在整合其机器人资产，以便对企业市场发动更协同的攻势。 缘何"分久又合"？ 然而，作为一家独立的Alphabet子公司运营，也存在不少局限性。 虽然Intrinsic可以动用Alphabet的资源，但它无法充分利用谷歌的云基础设施、企业销售团队，或像微 软通过Azure为竞争对手提供的那样，与谷歌的A ...

据悉，ZOLOZ融合AI、人脸识别和动态风险智能等核心能力，为企业提供AI数字安全验证解决方案，目前已为全球超30个国家和地 区的客户提供服务。 【环球网科技综合报道】随着全球企业级AI市场爆发，中国AI科技公司正加速在海外市场布局。 2月26日，据海外媒体报道，蚂蚁数科旗下旗舰AI产品ZOLOZ在马来西亚正式启动运营枢纽中心，旨在升级本地服务能力、加快响应 速度、增强本地处理能力，以更好地服务马来西亚市场客户。 此次马来西亚枢纽中心落地，是蚂蚁数科全球化布局的关键进展。此前，蚂蚁数科海外总部已经落户香港，并在印尼、新加坡等地建 立了成熟的业务基础。 在加速出海的同时，蚂蚁数科也在企业级AI领域持续发力。据媒体近日报道，蚂蚁数科将推出百灵大模型企业版，并已成立"大模型 技术创新部"，攻坚百灵大模型的toB场景落地。百灵企业版将更关注幻觉抑制、指令遵循、Agentic Engineering以及安全合规能力，以 满足企业级场景的高标准需求。 当前，企业级AI市场正迎来需求大爆发，海外AI公司Palantir2025年第四季度营收同比激增70%，被称为"过去十年科技领域最好业 绩"；Anthropic过去三年年营 ...

Core Insights - The article discusses the launch of VoiceAgentEval, a comprehensive evaluation standard for AI outbound calling, developed by Agora, Meituan, and xbench, addressing the lack of a dedicated assessment system in the AI outbound industry [1][3]. Group 1: Evaluation Framework - VoiceAgentEval establishes a unified and objective assessment standard for AI outbound calling, moving beyond previous academic benchmarks that do not adequately evaluate advanced communication capabilities [3]. - The evaluation framework consists of three main dimensions: benchmark construction, user simulation, and interaction quality assessment, leveraging the strengths of Agora in conversational AI, Meituan in outbound business scenarios, and xbench in AI benchmarking [3][4]. Group 2: Benchmark Construction - The benchmark is based on real-world data covering six business areas: customer service, sales, recruitment, finance, research, and proactive care, with 30 specific sub-scenarios [4]. - Each sub-scenario includes detailed evaluation plans that feature specific process breakdowns and a weighted scoring system [4]. Group 3: User Simulation - Meituan has developed a user simulator with 150 different personas to simulate real business interactions, allowing for large-scale testing of model task completion capabilities in a controlled environment [4]. Group 4: Evaluation Metrics - The evaluation employs a dual-dimensional assessment approach, combining text and voice evaluations, with a two-layer assessment system for text and 15 metrics for voice, integrating expert ratings and objective data [4][5]. Group 5: Leading Models - According to VoiceAgentEval, the top three performing models in AI outbound calling are ByteDance's Doubao-1.5-32k, OpenAI's GPT-4.1, and Anthropic's Claude-4-Sonnet, with Doubao-1.5-32k and GPT-4.1 excelling in voice interaction experience [5][6]. Group 6: Industry Impact - The release of VoiceAgentEval provides a critical reference for AI outbound practitioners and shifts AI model evaluation from idealized academic assessments to more realistic business scenario evaluations, significantly impacting the deployment of generative AI in the industry [7]. - Agora aims to continue enhancing its conversational AI and real-time audio-video cloud services, with several retail and healthcare companies already integrating its outbound calling features [7].

行业研究 证 券 研 究 报 告 计算机行业重大事项点评 政策落地，数据+AI 驱动要素价值释放 事项： ❑ 2026 年 2 月 7 日，国家数据局与工信部等多部门联合发布《关于培育数据流 通服务机构加快推进数据要素市场化价值化的意见》，提出到 2029 年底，数 据流通服务机构能力显著提升，流通交易形态更加多元，数据产品和服务更加 丰富，各类主体供数用数意愿持续增强，全社会数据流通利用水平明显提高。 计算机 2026 年 02 月 26 日 推荐（维持） 华创证券研究所 证券分析师：吴鸣远 邮箱：wumingyuan@hcyjs.com 执业编号：S0360523040001 行业基本数据 | | | 占比% | | --- | --- | --- | | 股票家数(只) | 337 | 0.04 | | 总市值(亿元) | 61,676.90 | 4.81 | | 流通市值(亿元) | 55,806.98 | 5.39 | 评论： 《计算机行业重大事项点评：CPU：供需格局优 化，国产龙头或迎价值重估机遇》 2026-01-29 《计算机行业重大事项点评： Agent ：海外 Clawdbot 引爆市场 ...

Investment Rating - The report provides a "Buy" rating for the industry, indicating an expectation of stock performance exceeding the market by more than 10% over the next 12 months [45]. Core Insights - SRAM (Static Random Access Memory) is identified as a high-bandwidth on-chip storage layer that can significantly enhance AI inference speed by reducing latency and jitter compared to external HBM (High Bandwidth Memory) [3][11]. - The architecture of SRAM is gaining mainstream attention, with significant investments and partnerships, such as Nvidia's $20 billion acquisition of Groq's intellectual property and OpenAI's $10 billion contract with Cerebras [3][32]. - The report emphasizes the growing importance of AI memory-related upstream infrastructure, suggesting that investors should focus on key beneficiaries within the industry chain [3][39]. Summary by Sections SRAM as a High-Bandwidth Storage Layer - SRAM is positioned as an essential component in the multi-tier storage architecture, providing high bandwidth but with limited capacity and higher costs [3][11]. SRAM Enhancing AI Inference Speed - SRAM can improve AI inference speed, with examples such as Groq's LPU chip achieving a bandwidth of 80 TB/s and maintaining stable inference speeds of 275-276 tokens/s, outperforming other platforms [3][15][21]. - Cerebras' WSE-3 chip integrates 44GB of SRAM, achieving over 3000 tokens/s in inference tasks, significantly faster than mainstream GPU cloud inference [3][23][39]. SRAM Architecture Gaining Mainstream Attention - The report notes that major companies are investing in SRAM technology, highlighting Groq's partnership with Nvidia and Cerebras' funding round that values the company at $23 billion [3][32][39]. Investment Recommendations - The report suggests that the ongoing expansion of AI memory capabilities will enhance model performance and accelerate the deployment of AI applications, recommending a focus on core beneficiaries in the industry chain [3][39].

午后算力芯片概念拉升，寒武纪涨近10%。消息面上，国产算力芯片龙头海光信息公告，预计一季度实 现归属于母公司所有者的净利润6.2亿元—7.2亿元，同比增长22.56%—42.32%。海外方面，英伟达最新 发布2026财年第四季度财报，营收、净利润及下一季度指引全面超越市场预期。 中信证券指出，当前中美双方各大云巨头均大幅加大AI相关资本开支：OpenAI 正在向投资者传达，公 司目前的目标是到 2030 年累计投入约6000亿美元的算力支出；字节跳动已初步规划2026年资本开支 1600亿元人民币，高于2025年约1500亿元人民币；阿里于2025年云栖大会同样表示，将在未来三年3800 亿元人民币的资本开支投入基础上，额外增加投入。国产算力中，超节点架构是国产算力建设实现后发 赶超的必经之路，云厂商与设备商正加速推进开放协议的适配，建议重点关注互联密度提升带来的价值 重估机遇，包括光通信、高速线模组、交换芯片及交换机、IDC等环节。 数据显示，截至2026年1月30日，上证科创板50成份指数(000688)前十大权重股分别为海光信息、中芯 国际、澜起科技、寒武纪、中微公司、芯原股份、金山办公、联影医疗、佰维 ...

英伟达最新公布的季度业绩与未来财测全面击溃华尔街预期，高盛在最新研报中明确指出，这家芯片巨 头在未来数月内跑赢大盘的路径已变得异常清晰。 受超大规模云服务商持续强劲的资本支出推动，英伟达第一季度的营收指引大幅超出市场共识。据追风 交易台消息，高盛分析师 James Schneider 及其团队重申对该股的"买入"评级，并维持250美元的目标 价，这意味着该股较当前水平仍有近28%的上涨空间，此举预计将进一步提振市场对整个人工智能基础 设施板块的投资信心。 市场乐观情绪的背后不仅仅是历史业绩的兑现。高盛在报告中前瞻性地指出了推动英伟达持续走强的三 大核心催化剂：超大规模企业资本支出的上调预期、AI初创企业融资完成后的支出能见度跃升，以及 基于新一代架构的AI模型发布将再次印证其技术护城河。 此外，英伟达近期与 Meta、OpenAI 和 Anthropic 等顶尖科技巨头达成的深度战略合作与百亿美元级投 资布局，不仅从根本上锁定了未来的订单基本盘，也为包括存储和半导体设备在内的全球科技供应链带 来了广泛的积极溢出效应。 业绩与指引双双击溃市场预期 英伟达第四季度实现营收681亿美元，不仅高于高盛预期的673亿美 ...

Core Insights - Nvidia reported record revenue of $68.127 billion for Q4, a 73% increase from $39.331 billion year-over-year, and a net profit of $42.96 billion, up 94% from $22.091 billion [1] - For the full fiscal year, Nvidia's revenue reached $215.938 billion, with a net profit of $120.067 billion, equating to approximately $328 million per day [1] - Nvidia's performance serves as a barometer for AI demand, indicating that for leading players, there is no downturn, only a resurgence [1] Financial Performance - Nvidia's Q4 revenue of $68.127 billion is a significant milestone, reflecting the ongoing high costs associated with AI [3] - The data center business contributed $62.3 billion in Q4, a 75% year-over-year increase, accounting for over 91% of total revenue [3][4] - Nvidia's full-year revenue surpassed $200 billion for the first time, reaching $215.938 billion [4] Market Dynamics - Nvidia's CEO expressed confidence in the growth of customer cash flows, attributing it to the recognition of the value of Agentic AI across various enterprises [2] - Major cloud providers like Google, Amazon, Meta, and Microsoft are significantly increasing their capital expenditures, with a projected combined spending of nearly $700 billion by 2026 [3] Strategic Initiatives - Nvidia aims to establish a comprehensive AI ecosystem on its platform, encompassing various sectors such as AI, robotics, and life sciences [5] - The company is nearing an agreement with OpenAI for a potential $100 billion AI infrastructure project and has acquired technology from AI startup Groq for approximately $20 billion [5] - Nvidia acknowledges the competitive landscape in China, where local companies are making significant advancements [6] Industry Trends - A McKinsey survey indicates that over 70% of CIOs at large enterprises plan to double their technology spending between 2026 and 2027, with 70% of budgets redirected towards AI [8] - The ROI of AI remains elusive, with clients demanding significant productivity improvements in exchange for large orders [8] - The emergence of Agentic AI is drastically reducing development costs, allowing single individuals to complete tasks that previously required entire teams [9] Future Outlook - Nvidia's inventory is fully booked until 2027, with seamless transitions between product iterations [10] - The company is set to begin mass production of its next-generation Vera Rubin platform in the second half of the year, anticipating widespread deployment among cloud model builders [10]