现在，英伟达研究院的最新研究给出了答案：关键不在于设计多复杂的惩罚，而在于用对强化学习优化方 法。 大模型推理到底要不要「长篇大论」？过去一年，OpenAI o 系列、DeepSeek-R1、Qwen 等一系列推理模 型，把「长链思维」玩到极致：答案更准了，但代价是推理链越来越长、Token 消耗爆炸、响应速度骤降。 如果 scale-up 长链思维是通往 AGI 的路径，那么现有思维链的冗长问题是我们亟待解决的。 那么，能不能让模型「少说废话」，既快又准？ 过去的尝试大多失败：各种复杂的长度惩罚（Length Penalty）要么让模型乱答，要么训练不稳定，结果就 是效率提升了，准确率却掉了。 论文标题：DLER: Doing Length pEnalty Right — reinforcement learning for more concise and efficient reasoning DLER 来了！推理模型的「减长秘籍」 DLER 首先是细致及全面了分析了引入长度惩罚之后出现的新的强化学习训练问题，包括： 对于这些问题，DLER 提出了一套简单却强大的强化学习训练配方： 更令人惊喜的是，DL ...

Core Viewpoint - Microsoft is facing an unprecedented issue with a surplus of GPUs that are idly stored due to insufficient power supply and space for data centers [1][3][4]. Group 1: Power Supply Issues - The primary challenge is not the surplus of chips but the lack of power capacity and the speed at which data centers can be built near power sources [2][5]. - Microsoft has a significant number of NVIDIA AI chips that are currently unused due to power shortages [3][4]. - The overall power demand has surged in the past five years, driven by the AI and cloud computing boom, outpacing utility companies' capacity planning [11][12]. Group 2: Infrastructure Development - The construction of traditional power plants takes several years, while the demand for AI capabilities is growing rapidly, leading data center developers to seek alternative power solutions [13][14]. - Many data center developers are adopting "behind-the-meter" power supply methods to bypass public grids and meet energy needs directly [13]. - The construction timelines for solar energy systems are also lengthy, making it challenging to keep pace with the rapid changes in AI demand [16][27]. Group 3: Strategic Adjustments - Microsoft has decided not to hoard single-generation GPUs due to the risk of obsolescence and depreciation over time [24][25]. - The company emphasizes the need for energy-efficient chips as power constraints become a more pressing issue than chip availability [31][32]. - The industry is shifting focus from peak performance to energy efficiency in chip production as power supply becomes the limiting factor [30][32]. Group 4: Future Investments - Microsoft has received approval to export NVIDIA chips to the UAE for building AI training data centers and plans to invest $8 billion in the Gulf region over the next four years [34]. - This move indicates a shift of AI infrastructure from Silicon Valley to emerging markets with abundant energy resources [34][35].

对冲基金Third Point创始人Daniel Loeb正加倍押注AI赛道，他将投资组合拓展至他认为估值偏低的国际 半导体公司。 11月3日，对冲基金Third Point最新发布的第三季度投资者信函中，披露其建仓了韩国存储芯片制造商 海力士和日本半导体生产设备制造商Ebara。 Loeb在信中表示，这两家公司都是各自行业的领导者，是AI基础设施建设的重要受益者，其估值"绝对 合理"，相较于美国同行甚至存在"显著折价"。 尽管Loeb承认其基金近期在某些领域的表现未达预期，但他对市场前景表达了整体乐观态度。他认 为，持续的AI投资与美联储可能开启的降息周期，预示着有利的市场环境有望延续。 算力依然为王，继续看好英伟达 尽管开拓了新的投资版图，Loeb对AI算力核心资产的信念依然坚定。他强调： 我们仍然生活在一个算力受限的世界里。 他指出，这一主题已使其在台积电和英伟达这两家AI基础设施建设中不可或缺的公司的现有投资中受 益。 Loeb的团队深入研究了AI技术的发展。他反驳了此前因DeepSeek等模型在训练效率上取得突破而引发 的"AI算力需求见顶"的担忧。他写道： Loeb总结道，AI算力已经从单一的预训 ...

太空数据中心的能源成本将只有地面上的十分之一。 11 月 2 日，英伟达首次把 H100 GPU 送入了太空。 作为目前 AI 领域的主力训练芯片，H100 配备 80GB 内存，其性能是此前任何一台进入太空的计算机的上百倍。在轨道上，它将测试一系列人工智能处理 应用，包括分析地球观测图像和运行谷歌的大语言模型（LLM）。 此次测试飞行搭载于位于弗吉尼亚州雷德蒙德的初创公司 Starcloud 的 Starcloud-1 卫星上，是该公司雄心勃勃的计划的第一步，该计划旨在将全球耗能巨 大的数据处理基础设施迁移到太空。Starcloud 是 NVIDIA Inception 创业公司计划的成员。 支持者认为这个想法很有前景：在遥远的太空深处，数据中心不会占用宝贵的土地，也不需要那么多能源和水来冷却，它们也不会向大气中排放温室气 体。 在算力逐渐紧张的 AI 时代，把芯片发射到太空已成为一个新的发展方向。此前，英伟达的 Jetson 机器学习计算板卡曾搭载于多颗实验型和地球观测小型 卫星上。不过相比之下，本次 Starcloud 的行动可谓是建设太空数据中心的重要一步，这将是人类首次把地面数据中心的 GPU 送入 ...

Core Viewpoint - Microsoft is facing an unprecedented issue with a surplus of GPUs that are idly stored due to a lack of power and space, rather than a shortage of chip supply [1][3][4]. Group 1: Power and Infrastructure Challenges - The primary challenge is not the surplus of computing power but the insufficient power supply and the inability to quickly build data centers close to power sources [2][4]. - Microsoft has a significant number of Nvidia AI chips that are currently unused due to power shortages and a lack of ready-to-use data centers, referred to as "warm shells" [3][6]. - The overall demand for electricity has surged in the past five years, driven by the rapid expansion of AI and cloud computing, outpacing utility companies' capacity to meet this demand [15][16]. Group 2: Industry Response and Future Outlook - Data center developers are increasingly opting for "behind-the-meter" power solutions to bypass public utilities and address energy shortages [17]. - Despite efforts to increase power supply, the construction pace of data centers and cooling systems is lagging behind actual demand [18][20]. - There are concerns that if AI demand slows down, the investments in power plants and storage projects may become underutilized [22]. Group 3: Strategic Shifts in Chip Production - Microsoft has decided not to hoard single-generation GPUs due to the risk of depreciation if the chips cannot be powered in time [30][32]. - The industry is shifting focus from peak performance to energy efficiency, as companies now prioritize the most energy-efficient chips due to power constraints [39]. - The CEO of Microsoft has called for an increase in annual power generation capacity by 100 gigawatts, viewing it as a strategic asset for AI [28]. Group 4: Investment and Market Dynamics - Microsoft has received approval to export Nvidia chips to the UAE for building data centers necessary for AI model training, indicating a shift of AI infrastructure to energy-rich emerging markets [41][43]. - The company plans to invest $8 billion over the next four years in the Gulf region for data centers, cloud computing, and AI projects, highlighting the region's financial and energy advantages [42][43].

Cloud Computing & AI Partnerships - Amazon is entering a $38 billion deal with OpenAI to provide cloud computing services [1] - OpenAI is diversifying its cloud infrastructure, utilizing services from Google, Microsoft, and Amazon due to high compute demand [7] - Microsoft has invested $13 billion into OpenAI, experiencing a 10x return after renegotiating the deal [5] - Amazon has invested $8 billion into Anthropic, an OpenAI rival [3] Competitive Landscape - The cloud computing sector is witnessing complex relationships, with Microsoft partly owning OpenAI, a company now partnering with Amazon, a major cloud competitor [1][4][5] - Tech giants are engaging in "proxy wars" through investments in AI startups like OpenAI and Anthropic [5] - Amazon needs to attract significant AI clients like OpenAI to strengthen its AWS strategy [3][8] Hardware & Infrastructure - OpenAI is currently using Nvidia GPUs, even in its Google Cloud partnership, instead of Google's TPUs [9] - Amazon is constructing new data centers specifically for OpenAI, similar to its approach with Anthropic [10] - Amazon's deal with OpenAI is initially using Nvidia chips, but there is potential to incorporate AMD chips in the future [9]

Core Viewpoint - Jensen Huang's lobbying for Nvidia to sell Blackwell chips to China faced significant opposition from U.S. officials, leading to a decision not to discuss the matter during a summit with Trump [1][2] Group 1: Nvidia's Position - Nvidia's potential sales of Blackwell chips to China could reach hundreds of billions, which would allow Chinese AI companies to continue relying on Nvidia's technology [2] - The company is currently in a difficult position and is seeking solutions to maintain its presence in the Chinese market [2] Group 2: U.S. Government Response - Trump's decision not to discuss the approval of Nvidia's chip exports was influenced by strong opposition from senior advisors, including trade representative Jamieson Greer and Commerce Secretary Howard Lutnick [1] - The White House emphasized that Trump's decisions are guided by the best interests of the American people, despite input from top business leaders [2]

机器之心报道 编辑：泽南 太空数据中心的能源成本将只有地面上的十分之一。 11 月 2 日，英伟达首次把 H100 GPU 送入了太空。 作为目前 AI 领域的主力训练芯片，H100 配备 80GB 内存，其性能是此前任何一台进入太空的计算机的上百倍。在轨道上，它将测试一系列人工智能处理应用，包 括分析地球观测图像和运行谷歌的大语言模型（LLM）。 此次测试飞行搭载于位于弗吉尼亚州雷德蒙德的初创公司 Starcloud 的 Starcloud-1 卫星上，是该公司雄心勃勃的计划的第一步，该计划旨在将全球耗能巨大的数据 处理基础设施迁移到太空。Starcloud 是 NVIDIA Inception 创业公司计划的成员。 支持者认为这个想法很有前景：在遥远的太空深处，数据中心不会占用宝贵的土地，也不需要那么多能源和水来冷却，它们也不会向大气中排放温室气体。 在算力逐渐紧张的 AI 时代，把芯片发射到太空已成为一个新的发展方向。此前，英伟达的 Jetson 机器学习计算板卡曾搭载于多颗实验型和地球观测小型卫星上。 不过相比之下，本次 Starcloud 的行动可谓是建设太空数据中心的重要一步，这将是人类首次把地面 ...

Core Viewpoint - Nvidia's $1 billion investment in Nokia signifies a strategic move to regain control over the communication sector, particularly in the upcoming 6G era, as the company aims to establish a dominant position in AI-driven wireless networks [2][4][5]. Group 1: Investment Details - Nvidia's investment in Nokia was announced by CEO Jensen Huang during the GTC conference, leading to a 22% surge in Nokia's stock price and Nvidia's market capitalization reaching $5.05 trillion, making it the first company to achieve this milestone [2]. - Despite Nvidia's substantial investments in various companies, the $1 billion investment in Nokia is seen as a critical step towards controlling communication technology, which is essential for the interconnectedness of AI in the future [4]. Group 2: Strategic Implications - The investment is part of Nvidia's broader strategy to reclaim the U.S. leadership in communication technology, which has been dominated by Chinese companies like Huawei in the 5G era [5][10]. - Nvidia's collaboration with Nokia aims to develop AI-RAN (Radio Access Network) technology, which will enhance the efficiency and capabilities of wireless communication networks, positioning Nvidia as a key player in the AI infrastructure [15][18]. Group 3: Nokia's Position - Nokia has shifted its focus from mobile phones to network infrastructure, generating significant revenue from its extensive patent portfolio, which includes over 3,000 5G core patents, accounting for 12% of the global total [11][13]. - In 2025, Nokia reported approximately €9.3 billion in revenue, with a net profit increase of 31.58%, demonstrating its strong financial performance in the telecommunications sector [11]. Group 4: Competitive Landscape - The competition for 6G technology is intensifying, with companies like Huawei and ZTE already advancing their own AI-driven network architectures, which poses a challenge for Nvidia and Nokia [24][25]. - Nvidia's partnership with Nokia is seen as a strategic move to counterbalance the technological advancements made by Chinese firms in the telecommunications space, particularly in the context of AI and network efficiency [26].

Core Insights - Nvidia's market capitalization reached $5 trillion, marking a historic milestone for publicly traded companies, achieved in just 113 days, showcasing the explosive growth of AI technology [2][4] - In stark contrast, the cryptocurrency market is experiencing a severe downturn, characterized by capital flight, innovation stagnation, and a collapse of confidence [2][10] Group 1: Nvidia's Market Performance - Nvidia's valuation surpasses that of most countries' annual GDP, with its market cap exceeding the total value of global cryptocurrencies by $1.2 trillion [5] - The company's market cap is over 60% of the combined valuations of Apple and Microsoft, establishing it as a new benchmark in the tech industry [6] - The S&P 500 index has shown strong performance, with a return rate that has outpaced Bitcoin this year, positioning it as a safe haven for capital [6][18] Group 2: Cryptocurrency Market Decline - The cryptocurrency market, once thriving, is now facing a significant decline, with Bitcoin's market cap at approximately $2.3 trillion, accounting for 62% of the total crypto market [7] - Altcoins have suffered greatly, with a combined market cap of about $1.5 trillion, down nearly 30% since the beginning of the year, leading to liquidity crises for many tokens [9] - The shift in capital from cryptocurrencies to stocks is evident, particularly in the South Korean market, where the KOSPI index has surged nearly 71% this year [10][11] Group 3: Internal Challenges in Cryptocurrency - The cryptocurrency market is facing internal issues, with a 30% decrease in the number of developers on GitHub, indicating a halt in innovation [16] - Many altcoins lack clear use cases and have flawed economic models, leading to a reliance on speculative trading rather than real value creation [15] - The absence of regulatory frameworks has exacerbated the uncertainty in the crypto market, highlighted by recent events that have shaken investor confidence [22][23] Group 4: Comparative Analysis with Traditional Assets - In the competition with traditional assets, cryptocurrencies have failed to maintain their "high risk, high reward" appeal, as evidenced by the S&P 500's superior returns supported by solid corporate earnings [18][20] - Gold has emerged as a preferred safe-haven asset, with prices surpassing $4,300 per ounce, while Bitcoin struggles to transition from a risk asset to a reliable hedge [21][22] - The current capital migration reflects a preference for assets with visible profitability and controlled risks, making it challenging for cryptocurrencies to attract investment [24]