编辑部 整理自 MEET2026 量子位 | 公众号 QbitAI 当国产AI芯片接连发布、估值高涨之际，一个尖锐的问题依然悬在头顶：它们真的能撑起下一代万卡集群与万亿参数模型的训练吗？ 一边是市场对替代方案的迫切期待，另一边是英伟达在利润与估值上仍高歌猛进…… 百度智能云 A I计算首席 科 学家王雁鹏 在量子位MEET2026智能未来大会上，基于百度昆仑芯在大规模生产环境中的实战经验指出： 评判芯片实力的标准已从单纯的算力数值，转向了能否稳定支撑从百亿到万亿参数、从稠密模型到MoE架构、从单模态到多模态的完整模型谱 系训练，并能在万卡乃至更大规模集群上实现高效扩展。 在演讲中，王雁鹏系统拆解了大规模训练中集群稳定性、线性扩展与模型生态三大核心挑战的攻坚路径，并回应了行业对MoE时代硬件路径的 关切。 他认为，即便是参数激增的MoE模型，"小芯片搭大集群"的路径依然可行，其关键在于极致的通信优化与系统级协同设计。 为完整呈现王雁鹏的思考，在不改变原意的基础上，量子位对演讲内容进行了整理编辑，希望能给你带来更多启发。 MEET2026智能未来大会是由量子位主办的行业峰会，近30位产业代表与会讨论。线下参会观众 ...

Investment Rating - The report provides a positive outlook for the Hang Seng Technology sector, indicating a "slow bull" market trend for 2026, driven by improved liquidity and AI advancements [19][29]. Core Insights - The Hang Seng Technology Index has shown significant volatility since its inception, with a notable recovery in 2025, where it outperformed major indices like the Shanghai Composite and Nasdaq [6][10]. - The report emphasizes the importance of earnings growth as a foundation for index performance, with a projected revenue growth of 13.6% and a non-GAAP net profit growth of 21.7% for 2025 [16][17]. - Key investment opportunities for 2026 are identified in sectors such as AI applications, autonomous technology, and consumer internet, with a focus on companies that can demonstrate earnings upgrades and AI progress [27][33]. Summary by Sections Review and Retrospective: What Happened in 2025? - The Hang Seng Technology Index experienced a 24% increase in 2025, outperforming the Shanghai Composite Index, which rose by 18% [10]. - Major contributing factors included a favorable liquidity environment due to the Federal Reserve's easing policies and strong earnings from leading companies [19][22]. Industry and Trends: Which Sub-sectors Are Worth Watching? - AI applications are highlighted as a key area of focus, with advancements in technology and commercialization expected to drive growth [33]. - The report also discusses the importance of the gaming sector and the autonomous driving market, indicating a shift towards head-to-head competition in L2+ and L4 technologies [33]. Views and Strategies: What Certainty Opportunities Can Be Seen in 2026? - The report suggests that identifying companies with upward earnings revisions and significant AI developments will be crucial for investment strategies in 2026 [27][29]. - The Hang Seng Technology Index's performance is influenced by both micro and macro factors, with a strong emphasis on the earnings potential of constituent companies [29]. Key Stocks: Structurally Viewing Stocks and Grasping Quality Targets - The report identifies top-performing stocks such as Hua Hong Semiconductor, JD Health, and Alibaba, which have shown significant price increases due to positive earnings revisions and AI advancements [23][27]. - The focus for 2026 will be on finding stocks with potential for earnings surprises and strong AI progress [27].

▲ 戳蓝 色字关注我们！ 垄断者最大的敌人不是竞争对手，而是时间。 —— 克莱顿·克里斯坦森 开场 一场精心编排的科技"变脸"大戏 2025年12月8日，美国东部时间下午，全球科技圈被一条消息炸开了锅。 美国总统特朗普在其自创的社交平台"真实社交"（Truth Social）上发文宣布： 批准英伟达向中国出口 H200人工智能芯片——但美国政府要从每笔销售中抽取25% 的分成。 这条消息的冲击力，不亚于在平静的湖面投下一颗深水炸弹。要知道，就在一周前，美国国会还在激烈讨论《SAFE法案》，试图全面禁止向中国出口高端AI芯 片。而现在，特朗普却突然来了个180度大转弯。 这就像是一个曾经把你拉黑、删好友、在朋友圈发誓"此生不再相见"的前任，突然给你发消息说："我们可以复合，但你得每个月给我上交四分之一的工资，而且 我只能给你用18个月前的旧款手机。" 荒诞吗？非常荒诞。但这就是当下美国科技政策的真实写照—— 在 "国家安全"的遮羞布下，赤裸裸地玩着利益算计的游戏。 更有意思的是，特朗普在声明中特意提到："我已经通知中方领导人这一决定， 中方对此表示积极回应 。"这个"积极回应"四个字，耐人寻味。是真的积极？还是 ...

Group 1: Market Performance and Company Overview - Moer Technology, known as the "first domestic GPU stock," saw its share price increase by 425.46% on its first trading day, closing at 600.5 yuan per share, with a market capitalization of 282.3 billion yuan [2] - The initial public offering (IPO) price was 114.28 yuan per share, indicating a significant rise in value and a potential profit of 240,000 yuan for investors holding one lot [2] - The company focuses on the research, design, and sales of GPUs and related products, targeting AI, cloud and data centers, high-performance rendering, and video acceleration [2] Group 2: Competitive Landscape - Moer Technology's market valuation at the IPO was 53.715 billion yuan, with a projected 2024 diluted static price-to-sales ratio of 122.51 times, higher than the industry average of 111.23 times [2] - The domestic AI chip market, particularly for GPUs, faces intense competition, with Nvidia holding a dominant position globally [2] Group 3: AI Developments and Innovations - DeepSeek launched two new models, DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, which reportedly outperform Google's Gemini3 Pro in inference capabilities [3] - Lenovo introduced the "Lenovo AI Factory" solution and upgraded its heterogeneous computing platform, indicating a shift towards deeper integration of AI in industry applications [8] - Nvidia's CFO highlighted a shift in large model vendors seeking direct collaboration with Nvidia, moving away from reliance on cloud service providers [9] Group 4: Industry Trends and Future Outlook - UBS analysts noted that the likelihood of an AI bubble in China is low, attributing this to limited domestic financing and cautious capital expenditure [10] - Micron Technology announced its exit from the consumer storage business to focus on providing storage products for AI applications, reflecting a strategic pivot towards higher-growth segments [14] - Amazon launched its custom AI chip, Trainium3, which reportedly offers four times the computational speed of its predecessor and can reduce costs by up to 50% compared to equivalent GPU systems [15]

Core Insights - The release of Google's Gemini 3 marks a new leap in large model technology, showcasing significant advancements in reasoning, multi-modal understanding, and code generation capabilities [8][9][10] - Alibaba has launched the "Qianwen App," an AI assistant based on the Qwen model, aiming to compete directly with ChatGPT [5][28] - The Chinese government has issued guidelines to accelerate the cultivation and application of high-value AI scenarios, indicating strong support for AI development [5][24] Group 1: AI Applications and Market Trends - The AI application theme is gaining traction, with a focus on domestic consumption, infrastructure in Xinjiang, and AI applications in various sectors [5][8] - The Chinese government aims for over 70% penetration of new-generation intelligent terminals and agents by 2027, and over 90% by 2030, highlighting the urgency of AI integration [5][8] - The financial sector is identified as a prime testing ground for AI applications, with significant potential for transformation through AI technologies [13][14] Group 2: Technological Advancements - Gemini 3 has achieved a significant leap in reasoning capabilities, scoring 37.5% in the Humanity's Last Exam, a notable increase from its predecessor [8] - The model's multi-modal understanding has set new benchmarks in complex scientific chart analysis and dynamic video comprehension [9] - Innovations in code generation and front-end design have positioned Gemini 3 as a leader in programming competitions, enhancing its commercial viability [9][10] Group 3: Competitive Landscape - Alibaba's "Qianwen App" is positioned to leverage its strong model performance to create a high user engagement platform, potentially accelerating the profitability of large model vendors [28][29] - The competition in the AI space is intensifying, with major players like Google and Alibaba making significant strides in consumer-facing AI applications [34][35] - The AI ecosystem is evolving rapidly, with companies like OpenAI and Nvidia pushing the boundaries of AI capabilities and applications [51][56] Group 4: Policy and Regulatory Environment - The Chinese government's recent policies are designed to foster AI development, emphasizing the importance of high-quality data governance and infrastructure [24][25] - The issuance of guidelines for AI deployment in government sectors indicates a structured approach to integrating AI into public services [24][25] - The focus on AI in governance is expected to create substantial market opportunities, with projections estimating the AI+ government service market to reach 2.54 billion RMB by 2024 [25]

Core Insights - Baidu's Q3 financial report highlights the significant growth of its AI business, with total AI revenue reaching 9.6 billion yuan, a year-on-year increase of over 50% [1][9] - The report marks the first disclosure of AI business revenue, showcasing the company's strategic pivot towards AI as a growth engine amid pressures on traditional business [1][9] AI Business Performance - AI cloud infrastructure revenue was 4.2 billion yuan, up 33% year-on-year, with high-performance computing subscription revenue increasing by 128% [9] - AI applications generated 2.6 billion yuan, reflecting a 6% year-on-year growth, while AI-native marketing services surged to 2.8 billion yuan, a remarkable 262% increase [9] - The overall AI business is positioned as a transformative force for Baidu, with expectations for continued revenue growth and improved profit margins [9][12] Autonomous Driving Developments - Baidu's autonomous driving service, "Luobo Kuaipao," reported 3.1 million global ride services in Q3, a 212% increase year-on-year, with total rides exceeding 17 million [1][3] - The company aims to expand its autonomous driving operations both domestically and internationally, with plans to increase vehicle deployment in existing cities and enter new markets [3][4] - Baidu's Robotaxi business is expected to focus on expanding order volume, city coverage, and global market penetration in 2026 [6][4] Financial Overview - Baidu's total revenue for Q3 was 31.2 billion yuan, a 7% decline year-on-year, with core revenue also down 7% to 24.7 billion yuan [12][14] - Online marketing revenue fell by 18% to 15.3 billion yuan, while non-online marketing revenue grew by 21% to 9.3 billion yuan, driven by the AI cloud business [12][14] - The company reported a net loss of 11.2 billion yuan for the quarter, compared to a profit of 7.6 billion yuan in the same period last year [14][16] Cost and Investment Strategy - Revenue costs increased by 12% to 18.3 billion yuan, primarily due to rising costs associated with AI cloud and content [13] - Baidu plans to maintain high investment levels in AI, having already invested over 100 billion yuan since the launch of its Wenxin large model [16] - The company anticipates that improvements in asset structure and capital efficiency will enhance profitability as AI business scales [16][17]

Group 1: Global Market Trends - Global stock markets experienced a significant decline, termed "Black Tuesday," with the Nikkei 225 index dropping 3.22%, marking its largest single-day drop since April, closing below 49,000 points. The KOSPI index fell by 3.32%, led by declines in chip stocks [1][3] - Gold and silver prices also plummeted, while Bitcoin fell below $90,000 for the first time in seven months, erasing all gains made in 2025 over the past month [2] - A-shares and Hong Kong stocks opened lower and continued to decline, with the Shanghai Composite Index experiencing its first three consecutive days of decline since mid-September [3] Group 2: Economic Factors - The expectation of a Federal Reserve interest rate cut has cooled, with the probability of a 25 basis point cut in December now below 50% according to the CME FedWatch Tool [4][5] - Japanese government bonds faced significant selling pressure, with the 10-year bond yield rising above 1.75%, nearing its highest level since 2008. This increase in yield is closely linked to global liquidity and capital flows [6] Group 3: AI Market Concerns - There are growing concerns about a potential AI bubble, particularly ahead of Nvidia's upcoming earnings report, which follows a previous strong performance but resulted in a significant stock price drop [7][8] Group 4: Chinese Market Outlook - UBS forecasts a prosperous year for the Chinese stock market in 2026, driven by favorable factors including innovation. The MSCI China Index is projected to reach a target of 100 by the end of next year, indicating a 14% upside from current levels [10] - Earnings per share for Chinese companies are expected to grow by 10% in 2026, supported by measures against "involution" and a decrease in depreciation expenses. The focus remains on internet, hardware technology, and brokerage sectors, while high-dividend stocks are being removed from the watchlist [11] Group 5: Baidu's AI Revenue - Baidu reported a total revenue of 31.2 billion yuan for Q3 2025, with core revenue at 24.7 billion yuan. The company disclosed its AI business revenue for the first time, showing a year-on-year growth of over 50% [12] - AI cloud revenue grew by 33%, while AI application revenue reached 2.6 billion yuan, and AI native marketing service revenue surged by 262% to 2.8 billion yuan [12]

Investment Rating - The report maintains a "Recommended" rating for the computer industry [1] Core Insights - The report highlights the increasing consensus on supernodes as a foundational infrastructure for AI, driven by significant capital expenditures from major cloud service providers (CSPs) and advancements in GPU/ASIC technologies [5][8] - OpenAI has secured substantial power agreements totaling over $1 trillion, indicating a robust demand for AI computing power [5][14] - The report emphasizes the growth potential in various segments including servers, liquid cooling, power supply, and storage, with clear growth momentum identified [7][8] Summary by Sections Demand Side - CSPs are raising their capital expenditure forecasts, with Google increasing its 2025 capex to $91-93 billion, Meta to $70-72 billion, and Amazon to $125 billion, primarily for AI infrastructure [5][29] - OpenAI has signed significant power agreements with NVIDIA, AMD, and Broadcom, totaling 26GW and over $1.1 trillion in value [5][15] - Sovereign AI investments are projected to reach $1 trillion over the next five years, with major projects in the US, EU, and Saudi Arabia [5][25] Supply Side - The report notes the continuous iteration of chips, with supernodes becoming a consensus in AI infrastructure, as evidenced by new product launches from various manufacturers [34][5] - NVIDIA's upcoming GPU architectures are expected to enhance computational capabilities significantly, with the Blackwell Ultra and Rubin architectures set to launch in the coming years [36][39] - Major companies like AMD and Huawei are also advancing their AI chip offerings, with AMD's Helios solution and Huawei's Atlas series expected to drive further innovation [41][44] Growth Segments - The server market is experiencing substantial growth, with companies like Hon Hai and Wistron achieving significant increases in AI server shipments [7] - Liquid cooling technology is becoming essential in AI data centers, with companies reporting high double-digit revenue growth [7] - The storage chip industry is entering a period of severe shortage, driving up DRAM prices and increasing demand from CSPs [7] Multi-modal and Application Ecosystem - The report outlines the rapid evolution of AI models, with major updates from OpenAI and Alibaba, and a significant increase in token usage across platforms [7][8] - OpenAI's new applications and models are enhancing its ecosystem, with a notable increase in daily token usage [7][8]

Group 1 - SoftBank expressed acquisition interest in Marvell, but terms were not agreed upon, leading to a significant increase in Marvell's stock price [1] - Marvell specializes in custom chips, with ASIC services similar to Broadcom, offering comprehensive solutions from network interfaces to memory and packaging [1] - Google plans to launch its seventh-generation Tensor Processing Unit, Ironwood, which boasts four times the performance of the sixth-generation Trillium TPU for training and inference tasks [1] Group 2 - ASICs demonstrate significant advantages in energy efficiency and cost, with Google's TPU v5 being 1.46 times more efficient than NVIDIA's H200, and Amazon's Trainium2 reducing training costs by 40% and inference costs by 55% [2] - Broadcom's AI ASIC revenue is projected to reach $12.2 billion in 2024 and $13.7 billion in the first three quarters of 2025, with quarterly growth rates surpassing NVIDIA [2] - The global AI ASIC market is expected to reach $125 billion by 2028, according to AMD, while Broadcom anticipates the ASIC service market for major clients to be between $60 billion and $90 billion by 2027 [2] Group 3 - Domestic cloud service providers are achieving results in self-developed ASICs, with Baidu's Kunlun chip reaching its third generation and securing a 1 billion yuan order from China Mobile [2] - Alibaba's PPU surpasses NVIDIA's A800 in memory and bandwidth, with a signed order for 16,384 computing cards from China Unicom [2] - ByteDance initiated chip self-development in 2020, planning to achieve mass production by 2026 [2]

Group 1 - The Hang Seng Tech Index rose over 3%, reaching a nearly four-year high, driven by gains in major tech stocks such as Kuaishou, Baidu, Alibaba, JD.com, and Tencent [1] - The Hang Seng Tech Index increased by 9.2% in September, with valuations of Hong Kong internet giants converging rapidly with those of their overseas counterparts [1] - Analysts believe that domestic tech giants are entering a phase of increased AI investment, similar to the significant capital expenditures seen from Microsoft and Meta in mid-2023 [1] Group 2 - Morgan Stanley raised Alibaba's target price to HKD 240, citing a continuous acceleration in Alibaba Cloud's revenue growth, driven by demand in generative AI across various sectors [2] - Guotai Junan Securities adjusted Baidu's target price to HKD 176, maintaining an "overweight" rating, as the market begins to reassess the commercial viability of Baidu's AI initiatives [2] - Dongfang Securities noted that Kuaishou could maintain its leading position through model iteration, with the 2.5 Turbo version expected to drive user growth and revenue due to a combination of performance upgrades and a 30% price reduction [2]