不找人聊，就能知道人在想什么？一群00后正在用AI改写调研行业。 2024年，一群平均年龄18岁的年轻人用约5000次AI对话（每次仅需30-90秒），便以接近零的成本，成功预测了美国纽约州民主党初选的结果，票数误差 不到400张。 不到两年，这群年轻人所创立的AI调研公司Aaru，已拿下埃森哲、安永和IPG等顶级合作伙伴，并在2025年底以10亿美元估值完成5000美元A轮融资。 这一切的背后，是一个简单到近乎狂妄的理念——用"无限模拟"取代"有限样本"。 Aaru的核心不是让AI变得更会"问问题"，而是让AI学会"当人"。他们训练了成千上万个AI智能体，每个都被赋予复杂的人口属性和行为认知模式，像一个 微缩版的真人。 当这些"合成人"在数字世界里相互作用，就能回答以前无法回答的问题，如人群面对新产品、新政策或新广告时的集体反应。 Aaru所代表的"合成行为"处于技术栈顶层，它正与其他"合成互动" （如Keplar、Outset） 与"合成数据" （如Gretel、YData） 的探索者，重塑价值800亿美 金的调研市场。 01 当AI Agents像人一样思考 当市面上大多数AI竞争者还在围绕"如何更高效 ...

近日，美国AI合成研究公司Aaru完成新一轮融资，融资规模超过5000万美元（约合人民币3.5亿元），本轮由红点创投领投，AngularVentures、 GeneralCatalyst等跟投。此次融资采用分层估值方式，部分股份以10亿美元（约合人民币70.7亿元）的名义估值成交，实际估值略低于该数值，旨在兼顾 对外宣传与投资者优惠条款。 Aaru成立于2024年3月，核心技术是基于真实人口与行为数据训练数千个能够模拟人类行为的AIAgents，用于预测特定人群或地域对事件的反应。其产品 线覆盖企业（Lumen）、政界（Dynamo）和公共部门（Seraph）三大场景，帮助用户在选举、民调、市场洞察等领域实现高效、低成本的数据分析。 | Ned Koh Simulating the future @ Aaru. | Cameron Fink | | --- | --- | | | Predicting everything @ Aaru | | 工作经历 | 工作经历 | | Co-Founder | Co-Founder and CEO | | Aaru · 全职 | Aaru · 全职 | | 2024年 ...

Core Viewpoint - The report from Guotai Junan Securities highlights the challenges faced by large language models (LLMs) due to the "memory wall" issue, proposing SSD-based storage offloading technology as a new pathway for efficient AI model operation [1][2]. Industry Perspective and Investment Recommendations - The massive data generated by AI is straining global data center storage facilities, leading to a focus on SSDs as traditional Nearline HDDs face supply shortages. The industry is rated "overweight" [1][2]. - The growth of KV Cache capacity is surpassing the capabilities of High Bandwidth Memory (HBM), necessitating the optimization of computational efficiency and reduction of redundant calculations through KV Cache technology [2]. KV Cache Management and Technological Innovations - The industry is exploring tiered cache management technologies for KV Cache, with NVIDIA's Dynamo framework allowing for the offloading of KV Cache from GPU memory to CPU, SSD, and even network storage, addressing the memory bottleneck of large models [3]. - Samsung's proposal at the 2025 Open Data Center Conference suggests SSD-based storage offloading to enhance AI model performance, achieving significant reductions in token latency when KV Cache size exceeds HBM or DRAM capacity [3]. Market Dynamics and Supply Chain Adjustments - The demand for AI storage is driving a shift from HDDs to high-capacity Nearline SSDs, with NAND Flash suppliers accelerating production of ultra-large capacity SSDs (122TB and 245TB) in response to the supply gap in the HDD market [4].

Core Viewpoint - The article discusses the rapid growth and investment interest in AI inference chip companies, particularly focusing on Groq, which has recently raised significant funding and aims to challenge Nvidia's dominance in the market [3][4][5]. Investment and Funding - Groq has raised a total of over $3 billion, with its latest funding round bringing its valuation to $6.9 billion [2][11][13]. - The company has seen a dramatic increase in its valuation, from $2.8 billion in August 2024 to $6.9 billion in a recent funding round, indicating strong investor confidence [3][13]. - Groq's funding rounds have included significant investments from major firms such as BlackRock and Tiger Global Management, highlighting its appeal to institutional investors [3][12]. Market Dynamics - The global AI chip market is experiencing rapid growth, projected to increase from $23.19 billion in 2023 to $117.5 billion by 2029, with a compound annual growth rate (CAGR) of 31.05% [4]. - The shift in focus from training to inference in AI applications is creating new opportunities for companies like Groq, which specializes in inference-optimized chips [4][5]. Competitive Landscape - Groq, founded by former Google engineers, aims to disrupt Nvidia's monopoly by offering specialized chips designed for AI inference, known as Language Processing Units (LPUs) [7][8]. - The company emphasizes its ability to provide high-speed, low-cost inference capabilities, which are critical for interactive AI applications [5][15]. - Despite Groq's advantages, Nvidia maintains a significant lead in the market, holding an 80% share of the global AI cloud training market, and has a well-established ecosystem with its CUDA platform [16][18]. Business Model - Groq's business model differs from Nvidia's by focusing on providing cloud-based inference services without requiring customers to purchase hardware, thus lowering entry barriers for developers [9][8]. - The company has launched GroqCloud, a platform that allows developers to access its chips and services, further enhancing its market position [8]. Future Prospects - Groq's ambition to surpass Nvidia within three years reflects a strong market aspiration, but challenges remain, particularly in establishing a developer community and supporting large-scale models [11][16]. - Other competitors, such as Cerebras, are also emerging in the AI chip space, indicating a growing trend of new entrants aiming to challenge established players like Nvidia [17][18].

Core Viewpoint - The traditional Scaling Law for AI models is facing significant bottlenecks, particularly in China, where infrastructure investment is lagging behind the US, leading to challenges in AI inference performance and commercial viability [1][4][9]. Group 1: AI Inference Challenges - AI inference has become a critical area, with current demand for inference computing power exceeding that for training, as evidenced by GPT-5's API call volume exceeding 20 billion calls per minute [4][6]. - Chinese enterprises face a "push not moving," "push slow," and "push expensive" dilemma, with domestic models outputting less than 60 tokens per second compared to over 200 tokens per second for foreign models [7][9]. - The increasing complexity of AI applications, such as long text processing and multi-turn dialogues, has intensified the demand for improved inference performance [1][4][6]. Group 2: Huawei's UCM Technology - Huawei has introduced the Unified Cache Manager (UCM), a breakthrough technology designed to enhance AI inference performance by optimizing memory management and overcoming HBM capacity limitations [1][11]. - UCM employs a tiered caching strategy that allows for the efficient storage and retrieval of KV Cache data, significantly reducing inference latency and costs [10][11][18]. - The technology has demonstrated substantial improvements in inference speed, with a reported 125-fold increase in processing speed for specific applications in collaboration with China UnionPay [19][21]. Group 3: Industry Implications and Future Prospects - The introduction of UCM is seen as a pivotal move for the Chinese AI industry, potentially leading to a positive cycle of user growth, increased investment, and rapid technological iteration [18][24]. - Huawei's open-source approach to UCM aims to foster collaboration within the AI ecosystem, allowing various stakeholders to integrate and enhance their frameworks [28]. - The technology is expected to be applicable across various industries, addressing the challenges posed by the increasing volume of data and the need for efficient inference solutions [23][24].

Core Viewpoint - Nvidia's recent financial reports exceed expectations, and its growth prospects may surpass market predictions, particularly in AI infrastructure projects valued conservatively at over $1 trillion [1][2]. Group 1: AI Infrastructure Potential - UBS analysts estimate that Nvidia's AI infrastructure projects, conservatively assessed at "tens of gigawatts," could lead to annual data center revenues of approximately $400 billion within 2-3 years, nearly double the current market expectations of $233 billion for fiscal year 2026 [1][2]. - The construction boom in AI data centers is expected to manifest in the real economy by the second quarter of 2026, indicating a shift towards exponential infrastructure expansion rather than a cyclical concept [1][6]. Group 2: GB200 Shipment Insights - Nvidia reported that major hyperscale customers are deploying nearly 1,000 NVL72 racks weekly, equating to 72,000 Blackwell GPUs, with expectations for further capacity increases this quarter [3]. - UBS clarifies that Nvidia's communication regarding GB200 shipments aims to assure investors that rack issues have been resolved, rather than providing specific revenue run-rate figures [3]. Group 3: Network Business Growth - Nvidia's network revenue surged to approximately $5 billion in the first fiscal quarter, a 64% quarter-over-quarter increase, largely driven by NVLink revenue growth [3]. - The NVL72 system, which includes 72 GPUs, significantly enhances the network's performance compared to previous configurations, leading to tighter tracking of network revenue with NVL72 rack shipments [3]. Group 4: Gaming Business Recovery - The first fiscal quarter saw a nearly 50% quarter-over-quarter increase in gaming revenue, raising investor concerns about the potential repurposing of RTX 50 series graphics cards [4][5]. - UBS argues that any such repurposing is minimal due to limited supply of Blackwell-based RTX GPUs in the gaming channel, with growth primarily driven by channel replenishment after severe supply shortages [5]. Group 5: Gross Margin Recovery Path - Improvements in Blackwell profitability and cost reductions are expected to drive gross margins back to around 75% by the end of fiscal year 2026 [6]. - The GB300 is anticipated to play a crucial role in revenue recognition, with significant volume expected in the third fiscal quarter, while value pricing remains a key factor for Nvidia's gross margins [6].

Core Insights - The interview with Jensen Huang highlights the evolving challenges Nvidia faces in the context of geopolitical dynamics and AI advancements, particularly regarding collaborations with Saudi Arabia and the UAE, and the implications of U.S. chip control policies on Nvidia's market position [1][10][14] - Huang emphasizes the transformative potential of AI in driving GDP growth and reshaping industries, indicating a shift towards AI-driven factories and the need for substantial computational resources [6][36][37] Group 1: Nvidia's Strategic Positioning - Nvidia aims to redefine itself as a comprehensive computing platform provider, moving beyond traditional tech roles to become a key player in AI infrastructure [5][36] - The company is focusing on a dual customer strategy, targeting both OEMs and large-scale cloud service providers, which necessitates a flexible sales approach [2][39] - Huang argues that the U.S. chip control policies may hinder Nvidia's competitive edge, suggesting that a more integrated approach across the AI technology stack is essential for maintaining leadership [14][18][19] Group 2: AI and Economic Implications - Huang predicts that AI will significantly contribute to economic expansion, potentially alleviating labor shortages and creating new job opportunities through automation [36][37] - The concept of AI factories is introduced, where the demand for computational power will drive the creation of new industries, fundamentally altering economic models [6][36] - The interview discusses the importance of engaging with the Chinese market, highlighting the risks of missing out on substantial opportunities if U.S. companies do not participate in global AI advancements [19][23][29] Group 3: Technological Innovations - The introduction of the Dynamo system is presented as a critical innovation for optimizing AI processing tasks across data centers, enhancing efficiency and performance [42][45] - Huang elaborates on the need for a robust architecture that can handle diverse AI workloads, emphasizing the importance of balancing throughput and interactivity in system design [41][42] - The discussion includes the significance of Nvidia's gaming division, GeForce, as a foundational element for its broader technological ecosystem, underscoring its relevance in the company's overall strategy [63][67]

Core Insights - The interview with Jensen Huang, CEO of NVIDIA, highlights the company's pivotal role in AI computing and the challenges it faces due to geopolitical factors and chip control policies [2][4][12] - Huang emphasizes the transformation of NVIDIA into a data center-scale company, focusing on AI as a new industry that requires extensive computing resources [7][8][35] - The discussion also touches on the implications of the AI Diffusion Rule and the necessity for the U.S. to remain competitive in the global AI landscape, particularly against China [14][15][19][23] Geopolitical Challenges - Huang discusses NVIDIA's collaborations with Saudi Arabia and the UAE, emphasizing the importance of these partnerships in building AI infrastructure [12][13] - The conversation addresses the U.S. government's chip export restrictions, particularly the ban on H20 chips, and how these policies could undermine U.S. and NVIDIA's long-term leadership in AI [4][27][29] - Huang argues that limiting U.S. technology access to other countries could lead to a loss of competitive advantage, as other nations develop their own ecosystems [18][19][23] AI as a New Industry - Huang describes AI as a new industry that enhances human labor capabilities and will drive significant economic growth in the coming years [7][35] - The concept of AI factories is introduced, where data centers are seen as essential for the production of AI technologies [8][35] - Huang predicts that the integration of AI into various sectors will lead to a rapid increase in GDP and the emergence of new job opportunities [35] NVIDIA's Strategic Positioning - The company is positioned as a full-stack solution provider, aiming to maximize utility for both technology and manufacturing sectors [4][8][56] - Huang emphasizes the importance of flexibility in NVIDIA's offerings, allowing customers to choose components based on their needs while still encouraging the adoption of complete systems [56] - The discussion highlights NVIDIA's commitment to innovation and maintaining a competitive edge in the rapidly evolving AI landscape [57][58] Economic Implications - Huang notes that the global market for AI technology is vast, with the potential for significant revenue generation if the U.S. engages effectively with international markets, particularly China [29][30] - The conversation underscores the economic model of AI factories, where the efficiency of architecture directly impacts profitability and operational costs [53] - Huang stresses that the future of AI will not only transform existing jobs but also create new roles, driven by advancements in robotics and digital labor [35]

Group 1: Market Challenges and Competition - NVIDIA's market share in China has dropped from 95% to 50% due to local competitors gaining ground [1] - Huang expressed concerns that U.S. chip export controls could lead to the development of a non-U.S. AI ecosystem in China, threatening NVIDIA's CUDA platform [2][1] - Huang criticized the zero-sum logic of trying to win by suppressing competitors, arguing that it would only make them stronger [2] Group 2: Economic Impact of AI and Robotics - Huang believes that the integration of AI and robotics will expand the global GDP, despite concerns about job displacement [3][4] - The IT industry is expected to transition from a $1 trillion market to a $50 trillion global capital and operational expenditure market due to AI advancements [3] - Huang anticipates the emergence of new industries related to token manufacturing systems, which will enhance productivity [4] Group 3: Technological Innovations and Infrastructure - NVIDIA is positioning itself as an "infrastructure company" and is focused on optimizing data center energy usage to generate high-quality tokens [5] - The introduction of Dynamo aims to intelligently distribute processing tasks within data centers, enhancing efficiency in token generation [7][8] - Huang emphasized that large model inference is a complex process requiring a sophisticated operating system to manage diverse demands effectively [8]

图片来源：NVIDIA NVIDIA 的芯片产品组合涵盖了中央处理器（CPU）、图形处理器（GPU）以及网络设备（用于纵 向扩展和横向扩展）。 NVIDIA 发布了其最新的 " Blackwell超级AI工厂" 平台 GB300 NVL72，与 GB200 NVL72 相比，其 AI性能提升了 1.5 倍。 NVIDIA 分享了其芯片路线图，这样一来，行业内企业在现在采购 Blackwell系统时，便可以谨慎 规划其资本性支出投资，以便在未来几年内有可能从 "Hopper" 系列升级到 "Rubin" 系列或 "Feynman" 系列。 "Rubin" 和 "Rubin Ultra" 两款产品分别采用双掩模版尺寸和四掩模版尺寸的图形处理器（GPU）， 在使用 FP4 精度运算时，性能分别可达 50 petaFLOPS（千万亿次浮点运算）和 100 petaFLOPS，分 别搭载 288GB 的第四代高带宽存储器（HBM4）和 1TB 的 HBM4e 存储器，将分别于 2026 年下半 年和 2027 年推出。 全新的 "Vera" 中央处理器（CPU）拥有 88 个基于Arm公司设计打造的定制核心，具备更大的 ...