BiPS团队 投稿 量子位 | 公众号 QbitAI 随着视觉-语言模型 （VLM） 推理能力不断增强，一个隐蔽的问题逐渐浮现： 很多错误不是推理没做好，而是"看错了"。 在复杂视觉任务中，模型往往能正确识别对象、理解问题，甚至给出完整的推理链，却因捕捉了错误的视觉证据，得出自信却错误的答案。 现有方法通常在推理阶段"指路"——例如生成视觉提示或调用外部工具，以临时对齐证据。这类策略虽有效，却面临明显局限：视觉线索形式 受限、高度依赖具体任务，且推理开销大。更重要的是，它引出一个根本性问题： 如果模型始终需要外部提醒才知道"看哪儿"，它是否真的理解了视觉世界？ 为此，微软亚洲研究院与清华大学提出 BiPS （Bi-directional Perceptual Shaping） ，从源头重塑模型的"看图方式"。 BiPS不在推理时临时提示关注区域，而是在训练阶段就教会模型： 面对特定问题，哪些视觉细节必须关注，哪些可以忽略 。通过系统性地对 齐问题与视觉证据，BiPS促使模型内化一种核心能力—— 带着问题去看图 。因此，在推理时无需任何额外提示，模型也能自动聚焦于真正决 定答案的关键区域与细节。 实验表明，这种 ...

Core Viewpoint - Big Tech companies are planning to invest approximately $700 billion in capital expenditures by 2026, primarily focused on AI infrastructure, which includes data centers and related equipment [1][21]. Investment Plans - The five major companies are expected to contribute to 2.1% of current-dollar GDP through these investments [2]. - Other companies are also increasing capital expenditures, indicating a broader economic stimulus as long as this trend continues [3]. Share Buybacks and Funding Sources - Concerns exist that the increased spending may come at the expense of share buybacks, which have already begun to decline [2][7]. - The funding for the $700 billion investment will come from various sources, including reduced share buybacks, new share issuances, and debt issuances [5][9][19]. - Specific companies have already shifted from share buybacks to issuing new shares, such as Oracle, which issued $2.1 billion in new shares in 2025 [5][13]. Financial Performance and Debt - In Q4, share buybacks for the five companies dropped to $12.6 billion, the lowest since Q1 2018, compared to a peak of $149 billion in 2021 [7]. - Companies like Amazon and Meta have significantly reduced their share buybacks to allocate funds for AI investments [6][8]. - Oracle's recent bond offerings have seen high demand, indicating strong investor interest in corporate debt [16][19]. Operating Cash Flow - The operating cash flows for these companies are substantial, with Amazon generating $126 billion and Alphabet $127 billion in 2025, which can help fund the planned investments [20][23]. - Utilizing operating cash flow for investments is seen as a positive contribution to economic growth [23]. Economic Impact - The shift from share buybacks to investments in AI infrastructure is expected to stimulate economic growth, although it may not be well-received by shareholders [21][22]. - The overall investment strategy is viewed as a significant stimulus for the economy, provided that financial markets remain stable [23].

Investment Rating - The report maintains a positive outlook on the communication equipment and services industry, particularly focusing on investment opportunities in CPO and related supply chain companies [3][11]. Core Insights - North American cloud vendors have reported better-than-expected earnings, with significant capital expenditure guidance increases from major players like Google and Amazon, indicating strong growth in the AI computing industry chain [3][11]. - The report highlights the rapid growth of the optical communication industry driven by AI applications, with leading companies achieving record highs in stock performance [6][9]. - The transition from Scale-OUT to Scale-UP in optical applications is emphasized, suggesting a broadening of application scenarios and increased demand for optical modules and components [6][9]. Summary by Sections Weekly Viewpoint - The optical communication sector is experiencing unprecedented growth, with major cloud companies significantly increasing capital expenditures, leading to a strong performance in the optical communication supply chain [9]. - The report notes that while the market is currently experiencing fluctuations, long-term investment opportunities are becoming more apparent, particularly in the context of rising prices for optical fibers and components [9]. Industry News - Major cloud service providers like Google and Amazon have substantially raised their capital expenditure forecasts for 2026, with Google estimating between $175 billion and $185 billion, nearly double that of 2025, and Amazon projecting around $200 billion, a 50% increase from the previous year [11][24][25]. - The report discusses the implications of a recent tax increase on value-added services in China, which may impact the revenue and profit margins of major telecom operators [12]. Investment Highlights - The report indicates that the proportion of holdings in the optical communication sector has increased, reflecting a positive market sentiment driven by AI-related infrastructure investments [11][41]. - The domestic new generation computing infrastructure is set to enter a new cycle, with significant opportunities arising from the ongoing global infrastructure wave [11][41]. - The report suggests that the AI-driven network upgrades will enhance communication capabilities, leading to rapid advancements in network innovation and technology applications [11][41].

AI的故事，正在从"软件吞噬世界"，演变为"硬件被世界卡住"。 在极度撕裂的美国政治环境中，能让极左翼参议员伯尼·桑德斯（Bernie Sanders）和极右翼州长罗恩·德桑蒂斯（Ron DeSantis）站在同一阵线的 议题几乎不存在——除了"遏制数据中心"。 这不仅是华盛顿的政治奇观，更是华尔街面临的冷酷"物理修正"。当硅谷巨头们挥舞着比"阿波罗登月计划"还要昂贵的支票簿，试图通过堆砌 算力延续AI繁荣时，他们撞上了由政治和物理电网极限共同筑就的高墙。 纽约州立法者也已经提出法案，计划对新建数据中心的建设和运营许可实施至少三年的暂停。纽约至少是第六个考虑暂停新数据中心建设的 州。 总之，从佛罗里达州的社区抗议，到德克萨斯州电网的监管急刹车，一个被市场忽视的风险正在急剧升温：如果物理电网接不上、政治环境不 答应，那原本计入估值模型的数千亿美元资本开支，可能根本"花不出去"。 当桑德斯与德桑蒂斯"合谋" 桑德斯和德桑蒂斯在绝大多数议题上势同水火，但在数据中心激增的问题上，他们达成了罕见的共识：必须踩下刹车。 这种跨党派"联合"的敌意，源于美国民众对"AI副作用"的切肤之痛。全美各地，数据中心24小时不间断的低 ...

Group 1 - The core viewpoint of the article highlights a significant sell-off in software stocks, leading to a substantial profit of $24 billion (approximately 166 billion RMB) for short sellers amid a broader market downturn in technology stocks [2][3]. - S3 Partners reported that short sellers have increased their positions in major tech stocks, particularly in the software sector, despite many leading companies experiencing significant price declines [3][4]. - The software sector has seen a market capitalization loss of $1 trillion, with notable short-selling activity in companies like Microsoft, Amazon, Oracle, and Broadcom [3][4]. Group 2 - The article discusses a shift in investor sentiment, with a noticeable move towards defensive strategies as the market faces valuation concerns and a potential withdrawal of funds from previously favored assets like tech stocks and cryptocurrencies [5][6]. - Analysts have pointed out that the current environment reflects a structural sell-off in the software industry, with some investors questioning the relevance of software companies in the age of artificial intelligence [6]. - Despite the prevailing skepticism, some analysts argue that the software industry is not obsolete and can still thrive, as evidenced by companies like Palantir [6].

炒股就看金麒麟分析师研报，权威，专业，及时，全面，助您挖掘潜力主题机会！ 软件股抛售潮，余波未了！ 据最新消息，随着科技股与软件股遭遇猛烈抛售，做空者近期斩获的浮盈已高达240亿美元（约1660亿 元人民币）。 S3 Partners着重提到了几只被空头持续做空的软件类股票，包括微软、亚马逊、甲骨文、博通等。该机 构表示，在上述科技股中，微软尤为引人注目，因为在其股价持续走低之际，空头对其的操作策略发生 了明显转变。"以往，微软走势更像反转股，跌势中空头会选择回补平仓；如今却表现得像一只受情绪 驱动的问题股，空头趁弱势反而加码押注。"S3 Partners写道。 尽管近期人工智能题材备受质疑，但S3 Partners认为当前的普遍看空情绪主要集中在软件板块。追踪纳 斯达克100指数的景顺QQQ ETF以及"美股七巨头"（Magnificent 7）整体的空头兴趣并未明显增加。 不过，S3 Partners强调，对冲基金仍在加大对大型科技股的空头布局，即便其中不少龙头企业股价已大 幅回调。该机构表示："空头仓位蔓延迹象明显，微软今年空头兴趣跃升20%，甲骨文增长10%，博通 和亚马逊的空头兴趣分别上升了15%和 ...

Core Viewpoint - The recent $1 trillion decline in U.S. software giants like Microsoft and Salesforce has raised concerns, but many industry leaders believe the narrative of a software "Armageddon" is exaggerated, despite acknowledging that AI valuations appear inflated [1][5]. Group 1: AI Valuations and Market Sentiment - Arvind Jain, founder of Glean, a $7 billion AI unicorn, believes AI will not render software-as-a-service obsolete, emphasizing the importance of integration for future success [2][5]. - Andrey Khusid, founder of Miro, a $17 billion decacorn, acknowledges that AI valuations are excessive but predicts normalization within the next two years [5]. - Larry Li, founder of Amino Capital, suggests that the AI bubble is deflating, particularly for larger companies, indicating a potential market correction [5]. Group 2: IPO Market and Company Strategies - Discussions at the Web Summit highlighted that AI giants like OpenAI and Anthropic are competing to go public, aiming to attract investor interest in the rapidly growing sector [7]. - Khusid prefers to remain private, citing profitability and operational efficiency without the pressures of public markets [7][9]. - Many AI startups, including OpenAI, are not yet profitable, with OpenAI projected to lose $14 billion this year, yet investment in the sector remains robust, with over $340 billion directed towards global startups in 2025, 65% of which is in AI [9][10]. Group 3: Funding Landscape and Competitive Dynamics - Non-AI startups are facing a tougher funding environment, as they are often compared to AI companies that are experiencing extreme growth rates [10][11]. - The U.S.-China AI race is a significant topic, with the U.S. leading in innovation while China excels in scaling due to its supply chain advantages and a larger pool of AI engineers [13]. - Despite recent stock market volatility, the Dow Jones has surpassed the 50,000 mark, reflecting ongoing optimism in the AI sector, although a valuation reset is anticipated [14].

Core Viewpoint - Wall Street analysts maintain a positive outlook on software businesses despite recent market declines, suggesting that the sell-off may be overdone and presenting potential investment opportunities in select software stocks [1][3]. Software Sector Overview - The iShares Expanded Tech-Software Sector ETF has experienced a decline of over 22% since December 10, officially entering bear market territory as of February 3 [3]. - Analysts believe that certain software stocks could offer significant upside potential, with average price targets indicating increases of 47% to 63% [3]. Company-Specific Insights Datadog - Datadog's stock has fallen from nearly $200 per share in early November to around $120, indicating a potential upside of 61% according to analysts [5][9]. - The company provides cloud monitoring and security solutions, and is expected to grow revenue by 20% by 2026, leveraging AI to enhance operations and create new capabilities [6][8]. - Of the 33 analysts covering Datadog, 30 have a buy rating, reflecting strong confidence in its business model and future growth [9]. Snowflake - Snowflake's stock has an average price target suggesting a 63% upside, despite challenges in convincing investors of its AI strategy and its current lack of profitability [10][14]. - The company has formed partnerships with AI leaders and completed a $200 million deal with OpenAI, indicating its relevance in the AI space [13]. - Analysts remain optimistic, with 30 out of 33 providing buy ratings, highlighting confidence in its long-term potential [14]. Microsoft - Microsoft, while primarily known as a software company, is also seen as a major beneficiary of the AI boom, despite a 23% decline in stock price over the past six months [15][19]. - The company faced a sell-off following lower-than-expected growth in its Azure cloud business, which is critical for its AI-related revenue [16][18]. - Analysts have a strong positive outlook, with 34 out of 35 providing buy ratings, suggesting a 47% upside potential for the stock [19].

Recent data on the U.S. job market has flashed some worrying signs lately, but the construction industry sees greater demand for workers. The Associated Builders and Contractors trade group estimated in a report last month the industry will need to bring in 456,000 new workers in 2027, up 30.7% from the 349,000 needed this year. “Failing to do so will worsen labor shortages, especially in certain occupations and regions, placing further upward pressure on labor costs,” ABC Chief Economist Anirban Basu ...

Core Viewpoint - The technology sector has experienced significant losses over the past week, with major companies facing double-digit declines, marking a three-month downturn in tech stocks [1][2]. Group 1: Market Trends - The slump in U.S. technology stocks has persisted for three months, primarily affecting growth stocks, which are companies that typically increase earnings faster than the market average [2]. - Investors have shifted their preference from growth stocks to value stocks, which are less volatile and often have cheaper valuations relative to their earnings and long-term growth potential [3]. - The Russell 1000 Value index has increased by 8.4% since Halloween, while the tech-heavy Russell 1000 Growth index has decreased by 3.7% [4]. Group 2: Investor Sentiment - There has been a notable decline in investor optimism regarding artificial intelligence, which had previously driven technology stock prices higher [6]. - The recent downturn in tech stocks has been exacerbated by a lack of confidence in AI's ability to significantly enhance corporate financial performance and the broader economy [6]. - The rapid rise in tech stock prices has made them vulnerable to sharp declines upon any signs of disappointment, as evidenced by Microsoft's recent stock drop despite beating Wall Street expectations [7]. Group 3: Company-Specific Performance - Advanced Micro Devices (AMD) has seen a decline of almost 21%, Intuit (INTU) is down more than 17%, Micron Technology (MU) has dropped nearly 13%, Microsoft (MSFT) is down about 7%, Nvidia (NVDA) has fallen 9%, and Salesforce (CRM) has decreased by 12.5% [8]. - Microsoft experienced its largest one-day stock drop since March 2020, falling 11% due to signs of slowing cloud revenue, which is closely tied to AI [9].