【小熊团队】英伟达FY4Q26业绩快评：业绩、指引均超预期 FY1Q27指引超市场预期。 公司指引FY1Q27收入为780亿美金（±2%），同比+77%、环比+14%，超市场预期（727.78亿美金）。 根据电话会议内容，该指引未考虑任何来自中国区收入。 毛利率略高于公司指引，符合市场预期。 收入、净利润均超市场预期。 FY4Q26英伟达实现收入681.27亿美金，同比+73%、环比+20%，超市场预期（659.12亿美金）；实现净利润429.60 亿美金，同比+94%、环比+35%，超市场预期（363.02亿美金）；实现EPS为1.76美元，超市场预期（1.48美元）。 FY1Q27指引超市场预期。< 【小熊团队】英伟达FY4Q26业绩快评：业绩、指引均超预期 收入、净利润均超市场预期。 FY4Q26英伟达实现收入681.27亿美金，同比+73%、环比+20%，超市场预期（659.12亿美金）；实现净利润429.60 亿美金，同比+94%、环比+35%，超市场预期（363.02亿美金）；实现EPS为1.76美元，超市场预期（1.48美元）。 从同比来看，增长主要由 NVLink 72 scale-up交换机 ...

As AI evolves toward real-time reasoning, every part of the system is stressed all at once, from compute, memory, networking, storage, and even software. This new generation of AI requires extreme co-design: engineering the entire stack as a single system, in fact, across the entire data center. This shift is especially clear for state-of-the-art mixture-of-expert models like DeepSeek-R1, Kimi K2 Thinking, and gpt-oss.Reasoning, MoE models generate a ton of tokens, creating higher-quality answers for users ...

在1月31日的饭局结束后，黄仁勋在饭店门口接受媒体采访表 示 ， 2026 年 将 是 行 业 " 极 度 吃 紧 " 的 一 年 ， 对 高 带 宽 内 存 （HBM）和先进封装的需求将大幅爆发。算力，在2026年或 许依然会是资本市场的"热词"。 作者：郑晨烨 封图：视觉中国 2026年1月31日晚，一张合影开始在社交平台流传。 照片的背景是中国台湾省台北市的砖窑古早味怀旧餐厅，画面中，英伟达CEO黄仁勋与台积电董 事长魏哲家、鸿海董事长刘扬伟等多位AI供应链高管面对镜头，集体竖起了大拇指。 在这张合影中，黄仁勋坐在第一排正中央，他的身边是台积电董事长魏哲家、联发科执行长蔡力 行、广达电脑董事长林百里，纬创资通董事长林宪铭。在黄仁勋身后的，还有鸿海精密董事长刘扬 伟 、 和 硕 联 合 科 技 董 事 长 童 子 贤 、 工 业 富 联 （ 601138.SH ） 董 事 长 郑 弘 孟 、 胜 宏 科 技 （300476.SZ）董事长陈涛。全球绝大多数的AI服务器，都要在这些人的工厂里完成组装。 当天晚上，黄仁勋在台北市宴请了40多位供应链高管，根据社交平台流传的菜单，这顿饭的菜色 包括豆酥鳕鱼、芋头米 ...

Core Insights - The dinner hosted by NVIDIA CEO Jensen Huang in Taipei included key executives from major companies in the AI supply chain, highlighting the importance of collaboration in the industry [2][3] - Huang emphasized the challenges in the production of NVIDIA's next-generation AI chip, Blackwell, and indicated that the supply chain will be "extremely tight" in 2026 [3][12] - The event showcased the significant market value of the participating companies, collectively exceeding $5 trillion [2] Supply Chain Dynamics - TSMC's chairman, Wei Zhejia, confirmed that TSMC needs to work hard to meet NVIDIA's demand for wafers and CoWoS packaging capacity, with NVIDIA accounting for over 60% of TSMC's CoWoS demand in 2025 [3][4] - Major Taiwanese manufacturers like Hon Hai Precision, Quanta Computer, and Wistron are responsible for assembling NVIDIA's AI supercomputing systems, indicating a clear division of labor within the supply chain [5] - The introduction of liquid cooling solutions by companies like Qihong Technology reflects the increasing power demands of new chips, necessitating advanced thermal management [5][6] Financial Performance - Victory Technology, a PCB manufacturer, anticipates a significant increase in net profit for 2025, projecting a rise of 260.35% to 295.00% year-on-year, driven by high-value product orders related to AI computing [6][7] - Industrial Fulian, a subsidiary of Hon Hai, expects a net profit increase of 51% to 54% for 2025, with a notable growth in cloud service server revenue [7][8] - Companies in the optical module sector, such as Zhongji Xuchuang and Xinyi Sheng, are also experiencing substantial growth, with projected net profits increasing by up to 128.17% [8][9] Emerging Competitors - Domestic chip companies like Cambrian are achieving significant revenue growth, with projections indicating a 410.87% to 496.02% increase in 2025, marking their first annual profit [10][11] - The rise of domestic GPU manufacturers, such as Moer Technology and Tianxu Zhixin, reflects a shift towards local alternatives in the computing power market, driven by capital market support [11][12] - Huang's comments on the increasing demand for high-bandwidth memory and advanced packaging solutions suggest ongoing opportunities in the semiconductor sector [12]

Core Insights - The focus in the AI chip market is shifting from performance to cost efficiency, as commercial pressures mount and the cost of inference becomes a critical factor in determining competitive advantage [1][2][3] Group 1: Shift in Evaluation Criteria - The evaluation criteria for AI chips are transitioning from "who computes faster" to "who computes cheaper and more sustainably" as inference becomes a significant source of long-term cash flow [2][3] - High costs associated with inference are becoming more pronounced as deployment and commercialization of large models progress, leading to a reevaluation of chip performance metrics [3] Group 2: TPU's Cost Reduction - Google/Broadcom's TPU has significantly reduced its inference cost, with the transition from TPU v6 to TPU v7 resulting in a 70% decrease in unit token inference cost, making it competitive with NVIDIA's GB200 NVL72 [1][4] - The cost reduction in TPU v7 is attributed to system-level optimizations rather than a single technological breakthrough, indicating that future cost reductions will depend on advancements in adjacent technologies [4] Group 3: Competitive Landscape - Despite TPU's advancements, NVIDIA maintains a time-to-market advantage with ongoing product iterations, which are crucial for customer retention [5][6] - The investment outlook remains positive for both NVIDIA and Broadcom, with Broadcom's earnings forecast for FY2026 raised to $10.87 per share, reflecting its strong position in AI networking and custom computing [7] Group 4: Industry Dynamics - The report suggests a clearer division of labor within the industry, where GPUs continue to dominate training and general computing markets, while custom ASICs penetrate predictable inference workloads [7][8] - The significant drop in TPU costs serves as a critical stress test for the viability of AI business models, highlighting the importance of economic considerations in the ongoing GPU vs. ASIC competition [8]

Group 1 - The core issue is the increasing electricity demand from AI data centers, which is straining the existing power grid and leading to rising electricity prices [2][4][7] - There are over 4,000 AI data centers in the U.S., and their number is expected to triple in the next four years, significantly increasing electricity consumption [2][3] - By 2035, U.S. data centers' electricity demand is projected to surge from 200 terawatt-hours to 640 terawatt-hours, equivalent to Germany's annual electricity consumption [3] Group 2 - The current power grid is unable to meet the demand from new data centers, with Texas only able to approve about 1 gigawatt of the tens of gigawatts requested monthly [4][7] - The construction of new power lines and plants takes several years, which is not feasible for tech giants needing immediate power solutions [8] - Major tech companies are exploring various energy sources, including natural gas, nuclear, and renewable energy, to ensure stable power supply for their operations [15][22] Group 3 - Elon Musk's xAI has built a data center with 200,000 GPUs and on-site power generation using gas turbines and Tesla batteries, while Google has acquired a power company to secure its energy needs [9][11] - Meta has signed agreements with nuclear energy companies to supply power for its AI supercomputing cluster, aiming for 6.6 gigawatts by 2035 [12][11] - Microsoft has committed to not passing on electricity costs to consumers, although the complexity of the power grid makes this challenging [14] Group 4 - The competition for energy talent is intensifying, with tech companies increasing hiring in energy-related positions by 34% since 2022 [16][18] - Companies like Amazon and Microsoft are aggressively recruiting energy experts to navigate the complexities of energy procurement and grid access [18][21] - The demand for energy professionals is reshaping the job market, with traditional energy sectors facing talent shortages as tech firms offer higher salaries [21] Group 5 - The AI-driven electricity crisis is reshaping the energy industry, benefiting manufacturers of gas turbines and storage devices, while also creating economic disparities in local communities [22][24] - The ongoing "electricity war" highlights the limitations of current energy systems in supporting rapid technological advancements [23][25] - The future of technology may increasingly depend on energy availability, emphasizing the need for sustainable and efficient power solutions [25][26]

近期AI伺服器組裝廠商緯穎公布低於預期的4Q25毛利率 (7.2% vs. 市場預期的8-8.3%)，導致股價下跌，此事又引起投資人關注。雖零組件漲價也會稀釋帳面上的毛利率，但更重要的是從AI伺服器設計的本質上去檢視組裝的獲利能力趨勢。Nvidia持續提升AI伺服器的設計整合程度，以提升每單位空間與電力的Token產出，來因應資料中心有限空間與稀缺電力挑戰。此外，高度整合的設計也有助於改善生產效率，進而有利供應鏈管理與降低維修成本。VR200 NVL72就是設計高度整合的例子，我的產業調查顯示，其Compute tray導入無線化設計（Cable-less），零組件項目顯著減少約40% (vs. GB300 NVL72)。但隨整合程度提高，伴隨而來的是Nvidia指定用料比重提升，與客製化空間也遭到壓縮，這些都不利組裝廠商的獲利能力。我的產業調查顯示，VR200 NVL72 compute tray「不允許客製化規格」的零組件項目佔比提升至20-22%，遠高於GB300 NVL72的5-7%，且VR200 NVL72更直接取消「ODM設計」的零組件項目。事實上，提高整合程度的設計趨勢自GB300 NVL72就 ...

Core Insights - The presentation at CES 2026 highlighted the significant advancements of Chinese AI models, particularly Kimi K2 and DeepSeek, which are now competing closely with closed-source models in performance [1][8] - The introduction of the MoE (Mixture of Experts) architecture has become a mainstream choice, with over 60% of open-source AI models adopting this structure since 2025, leading to a substantial increase in intelligence levels [16][31] Group 1: Model Performance and Advancements - Kimi K2 Thinking's inference throughput increased tenfold, with token costs dropping to one-tenth of previous levels, indicating a shift towards a "price parity era" for AI inference [4][6] - DeepSeek-R1 and Kimi K2 represent top-tier attempts under the MoE architecture, significantly reducing computational load and memory bandwidth requirements [2][12] - The performance of Kimi K2 Thinking was validated in tests, showing a tenfold increase in performance on the GB200 NVL72 platform [9][19] Group 2: Global Recognition and Impact - DeepSeek and Kimi K2 were recognized in a rigorous benchmark test, with Kimi K2 Thinking achieving the title of "best-performing non-U.S. model" due to its low misguidance rate [21][24] - The rapid development of Chinese open-source models is closing the gap with the strongest closed-source models, providing a significant first-mover advantage [31] - The increasing international acceptance of Chinese AI models is evidenced by endorsements from prominent figures in the tech industry, indicating a growing influence in the global market [24][33] Group 3: Trends and Future Directions - The transition from high benchmark scores to practical usability is evident, with models like Qwen evolving from being known for high scores to being recognized for their quality [32] - The emergence of features such as "interleaved thinking" in Kimi K2 Thinking reflects a trend towards more sophisticated model capabilities, enhancing their applicability in real-world scenarios [34] - The rise of open-source models is pressuring U.S. closed-source giants, as the value proposition of paid models becomes harder to justify against the performance of open-source alternatives [35]

Core Insights - The report highlights a significant shift in AI inference economics, where the focus has moved from raw chip performance to the intelligence output per dollar spent [1][4][46] - NVIDIA continues to dominate the market, with its GB200 NVL72 outperforming AMD's MI350X by a factor of 28 in throughput [1][5][18] AI Inference Economics - The key metric for evaluating AI infrastructure has transitioned to "how much intelligence can be obtained for each dollar" [4][6][46] - In high-interaction scenarios, the cost per token for DeepSeek R1 can be reduced to 1/15th of other solutions [2][20] Model Architecture - The report discusses the evolution from dense models to mixture of experts (MoE) models, which activate only the most relevant parameters for each token, improving efficiency [9][11][46] - MoE models are becoming the standard for top open-source large language models (LLMs), with 12 out of the top 16 models utilizing this architecture [11][14] Performance Comparison - In terms of performance, the GB200 NVL72 shows a significant advantage over AMD's MI355X, achieving up to 28 times the performance in certain scenarios [18][24][30] - The report indicates that as interaction rates increase, the performance gap between NVIDIA and AMD platforms widens, with NVIDIA's solutions becoming increasingly efficient [30][37] Cost Efficiency - Despite the higher hourly cost of the GB200 NVL72, its advanced architecture and software capabilities lead to a lower cost per token, making it more economical in the long run [20][41][45] - The analysis shows that the GB200 NVL72 can achieve a performance per dollar advantage of approximately 12 times compared to its competitors [42][44] Future Trends - The future of AI models is expected to lean towards larger and more complex MoE architectures, with platform-level design becoming a critical factor for success [46][47] - Companies like OpenAI, Meta, and Anthropic are likely to continue evolving their flagship models in the direction of MoE and inference, maintaining NVIDIA's competitive edge [46]

Market - On January 3, US stocks closed mixed, with the Dow Jones and S&P 500 rising due to AI-related stocks, while the Nasdaq slightly declined. Nvidia, AMD, and Micron contributed to the gains, with Micron's stock rising approximately 10% to a record high. However, Tesla's stock fell over 2% due to poor delivery numbers [1][2] - The Nasdaq Golden Dragon China Index rose 4.38% on January 3, with notable gains in Chinese stocks such as Alibaba (up 6.24%) and Baidu (up 15.03%). However, UMC saw a decline of 0.51% [3] - Oil prices slightly decreased after recording the largest annual decline since 2020, with Brent and WTI crude oil both dropping by 10 cents. Analysts expect OPEC+ to maintain production cuts in the first quarter of 2026 [4] - European stocks reached record highs at the start of 2026, with the Stoxx Europe 600 index rising 0.7% and the UK FTSE 100 index surpassing 10,000 points for the first time [5] Company - BYD surpassed Tesla to become the world's largest electric vehicle seller in 2025, with a 28% year-on-year increase in pure electric vehicle sales to 2.26 million units. Tesla's deliveries fell by approximately 8% to 1.64 million units, marking a second consecutive year of decline [10][12] - Micron Technology's stock surged over 10% to a record high, driven by strong demand for AI, leading to significant improvements in profitability. The company is accelerating its domestic wafer fabrication plans to capitalize on market opportunities [11] - AQR Capital Management's multi-strategy product Apex achieved a return of 19.6% in 2025, while its market-neutral Adaptive Equities Strategy realized a 24.4% return. The firm's assets grew to $189 billion, with a record increase of $75 billion [15] - Berkshire Hathaway entered the post-Buffett era as Warren Buffett handed over leadership to Greg Abel. The company's stock performance has lagged behind the S&P 500, raising concerns among investors [23] - Saks Global's CEO is expected to step down as the company prepares to file for bankruptcy, following difficulties in meeting debt obligations from its merger with Neiman Marcus [18]