要理解CPU为何能再次站到"C位"，首先要澄清一个误区：CPU从未不重要。在传统的云计算和数 据中心架构中，CPU始终是绝对的核心。无论是虚拟化技术的实现、容器编排、网络流量转发， 还是数据库与中间件的运行，无一不依赖CPU强大的计算与调度能力，高性能CPU是现代云计算 体系大厦的地基。 然而，在过往若干年的AI时代，大语言模型（LLM）的成功几乎完全建立在大规模并行矩阵运算 之上。在这一计算范式下，无论是训练还是推理，核心任务都是对海量矩阵进行运算——这正是 GPU的绝对强项。而CPU则退居幕后，负责数据预处理、任务调度与结果后处理等调度性工作！ 但看似退居幕后的CPU，实则扮演着"总指挥"的关键角色。 受此范式驱动，资本市场对GPU创业公司掀起空前追捧，到了2025年底，摩尔线程、沐曦、壁 仞、天数智芯等GPU企业引发一轮上市热潮，CPU的战略价值在喧嚣的算力竞赛中被边缘化， CPU作为AI稳定运行的底座，其被视为"理所当然"——人们产生了一种错觉，仿佛在这一轮AI浪 潮中，GPU决定了一切，而CPU只是配角。 这是一种由特定技术阶段导致的认知盲区。我们将"AI=大模型=GPU"划上等号，却忽略了：真正 ...

Core Viewpoint - The article discusses the significant price increase of server CPUs by Intel and AMD, driven by structural changes in AI computing demand and global supply chain adjustments, which may lead to a revaluation of domestic CPUs in China [2][4]. Group 1: Price Increase and Market Dynamics - Intel and AMD plan to raise server CPU prices by 10%-15% in Q1 2026, with their annual production already largely pre-sold, indicating a severe supply-demand imbalance [2][3]. - The demand for AI infrastructure is expected to drive global AI server shipments to grow by over 28% year-on-year in 2026, with overall server shipments increasing by 12.8% [3]. - The supply constraints are exacerbated by high demand for advanced manufacturing processes, with Intel's production capacity reportedly overloaded at 120%-130% [4]. Group 2: Shift in CPU Role - The emergence of Agentic AI has transformed the role of CPUs from auxiliary computing units to central components responsible for complex scheduling and resource management [6]. - The new architecture allows for a shift from "compute-intensive" to "scheduling-intensive," with CPUs managing vast amounts of parameters and states previously reliant on expensive GPU memory [7]. - The number of active intelligent agents is projected to surge from tens of millions in 2025 to hundreds of billions by 2030, significantly increasing CPU demand [7]. Group 3: Domestic CPU Market Opportunities - The price increase of CPUs reflects their strategic value, particularly in the context of China's push for domestic semiconductor production [8]. - The domestic market is expected to seek alternatives due to international price hikes, creating a historic opportunity for domestic CPUs to fill market gaps [8]. - Key selection criteria for domestic CPUs include compatibility with existing X86 environments, security, and stability, especially in critical infrastructure sectors [9][10]. Group 4: Potential Domestic Players - Companies such as Haiguang Information, Loongson Technology, and China Great Wall (Feiteng) are positioned to benefit from the market overflow due to the global CPU price revaluation [10].

Group 1: Core Insights - Vivo's mobile internet business serves over 400 million users with applications, short videos, and advertising, processing daily data volumes in the range of hundreds of billions [1] - The transition from Kafka to Apache Pulsar addresses scalability and performance issues, enabling effective management of massive data traffic and improving operational efficiency [3][4] Group 2: Business Challenges - Vivo's original Kafka-based messaging system faced limitations due to increasing topic and partition numbers, leading to performance degradation and high operational costs [3] - The inability of Kafka to dynamically scale and the reliance on partition numbers for performance created significant challenges during traffic spikes [3] Group 3: Technical Selection - Apache Pulsar was chosen for its advantages, including a stateless broker architecture that supports rapid scaling and a unique bundle mechanism that manages large numbers of topics effectively [4] - Pulsar's support for multiple consumption modes enhances its ability to handle varying traffic demands and ensures message order [4] Group 4: Implementation and Optimization - Vivo optimized Pulsar's bundle management and data retention strategies, improving data distribution and monitoring capabilities [5][6] - Adjustments to load balancing and client performance parameters significantly enhanced the system's ability to handle high message volumes [6] Group 5: Didi's Big Data Operations - Didi's big data team adopted Apache Pulsar in 2021, replacing the DKafka system and resolving long-standing operational challenges [7][9] - The transition to Pulsar improved performance, cost efficiency, and reliability, addressing issues such as disk I/O bottlenecks and complex load balancing [8][9] Group 6: Didi's Implementation and Optimization - Didi optimized hardware configurations and utilized Pulsar's ensemble mechanism to ensure balanced data distribution and efficient resource utilization [10] - The system's design allows for quick scaling and fault recovery, ensuring continuous service during peak loads and failures [10][12]