In the AI wave that has swept the globe in recent years, the semiconductor industry chain has emerged as one of the biggest beneficiaries. Graphics Processing Units (GPUs) for large model training were the first to catch the wind, followed by memory chips supporting massive data throughput—companies like NVIDIA (NVDA) and Micron Technology (MU) saw both their performance and stock prices surge due to soaring demand. However, just as the market assumed the AI hardware红利 had largely been realized, a segment long considered "mature and even saturated" is quietly rising: the Central Processing Unit (CPU). According to the latest market dynamics, following GPUs and memory chips, CPUs are accelerating their integration into the AI ecosystem, reaching a structural inflection point in demand. The capital market has reacted swiftly—starting in the second half of 2025, global CPU-related concept stocks began to show unusual activity, culminating in a full-scale breakout by early 2026.
The capital market was the first to "jump the gun": CPU concept stocks collectively strengthened. The capital market is most敏锐 to感知 industrial trends. For a long time, the CPU field, with its relatively stable technological iteration pace and solidified competitive landscape, was viewed as a mature track with strong growth certainty but limited elasticity. However, this perception began to shift in the second half of 2025, as global CPU concept stocks陆续 showed unusual activity, leading to a全面爆发 in early 2026. In the U.S. stock market, the performance of the two CPU giants has been particularly亮眼. Data shows that since September 2025, Intel's (INTC) stock price has experienced a strong rally, accumulating a gain of over 120% to date, demonstrating very robust performance. However, Intel's fourth fiscal quarter earnings for the period ending December 27, 2025, announced after the market closed on January 22, surpassed expectations, but its guidance for the current quarter was significantly below expectations, causing the stock to plunge over 12% in after-hours trading. Advanced Micro Devices (AMD) has been even stronger; it is both a core GPU concept stock and has a CPU product line, and its stock price has maintained strong momentum in recent years. Data indicates that since April 2025, AMD has entered a new phase of significant gains, soaring nearly 150% to date, with its stock price approaching historical highs. Reportedly, the market holds high expectations for the potential of AMD's MI300 series APU (CPU+GPU融合 architecture) in AI inference scenarios. China's domestic CPU camp has also recently seen unusual activity, triggering a wave of limit-up gains. On January 21, domestic CPU leader Loongson Technology Corporation Limited (688047.SH) hit the 20% daily limit-up, while another representative company, Hygon Information Technology Co., Ltd. (688041.SH), also experienced a significant price surge, rising over 13% in a single day on January 21, driving strength across the entire domestic CPU sector. In Hong Kong stocks, although companies like SMIC (00981.HK) and Hua Hong Semiconductor (01347.HK) are primarily manufacturers, the significant increase in CPU orders they are fabricating has also contributed to a回暖 in sector sentiment. Looking at capital flows, institutional funds have recently been continuously increasing their positions in the semiconductor chip sector, with CPU-related targets becoming a key focus for allocation, reflecting market recognition of their long-term investment value.
Supply-demand imbalance coupled with AI drive: The core logic behind rising CPU demand. In fact, this round of上涨 is not without basis. Combining institutional views, CPUs are transitioning from "general-purpose computing units" to "AI协处理 cores," and their role in AI infrastructure is being重新定义. For instance, although GPUs handle the core computing power, modern AI servers place increasingly high demands on overall system efficiency. CPUs need to undertake critical functions such as task scheduling and memory management. Especially in large-scale cluster deployments, high-performance CPUs directly impact overall throughput efficiency and energy efficiency ratios. Furthermore, with the rise of AI Agents, the core value of CPUs has been significantly enhanced. Frank Lee, an analyst at HSBC, pointed out that as AI evolves from simple chat assistants to intelligent agents capable of autonomous planning, tool invocation, and executing multi-step tasks, the demand for general-purpose computing capabilities like control scheduling and serial processing grows exponentially—precisely the core strength areas of CPUs. It is worth mentioning that DeepSeek's newly launched Engram architecture, through a "compute-memory separation" approach, transfers static knowledge from GPU memory to CPU memory, upgrading the CPU from a supporting role to a core component for memory and scheduling in AI inference. This significantly increases the demand for CPU utilization, core count, memory capacity, and I/O bandwidth, thereby重构 the hardware configuration ratio of AI servers and boosting CPU demand. During an earnings call, Intel's management also emphasized that the diversification of AI workloads has created significant capacity constraints, which in turn reinforces the CPU's role as the "core commander." From AI inference to orchestration and control, CPUs are becoming increasingly indispensable, driving a robust refresh cycle for traditional servers. Intel's CFO, David Zinsner, added a key trend: the world is shifting from "human-prompted requests" to "computer-to-computer interactions," and this persistent, recursive command flow requires powerful CPUs for orchestration and control. Zinsner stated, "The core functionality of the CPU to coordinate these flows will not only drive the refresh of traditional servers but also new demand to increase the installed base." It is precisely this unexpectedly strong CPU demand,叠加 with Intel's own supply bottlenecks, that has led to the current shortage situation.
Rigid constraints on the supply side exacerbate the supply-demand mismatch. In stark contrast to the爆发 on the demand side, there are明显的刚性约束 on the CPU supply side. In recent years, influenced by weak demand for general-purpose servers, CPU manufacturers like Intel and AMD adopted conservative strategies—reducing capacity, lowering inventory, and slowing the pace of capacity expansion—leading to insufficient industry capacity reserves. However, with the sudden爆发 of AI demand, the expansion speed of key resources like advanced process nodes and advanced packaging cannot match demand growth. Simultaneously, the squeeze on capacity for chips like GPUs further limits the release of CPU capacity,加剧了 the supply-demand mismatch. Furthermore, the CPU industry has high technical barriers, making it difficult for new entrants to form effective supply in the short term. The capacity advantages of leading companies are further solidified, pricing power is beginning to tilt towards sellers, providing support for rising CPU prices. The latest industry commentary from Western Securities points out that in response to the supply-demand imbalance and to ensure stable supply, Intel and AMD have planned to increase server CPU prices by 10%-15%, and the server CPU capacity of both manufacturers for 2026 is "basically pre-sold out."
In conclusion, as the AI wave continues to evolve from training to inference and from the cloud to the edge, CPUs are shedding their old label of "traditional general-purpose chips" and accelerating their transformation into indispensable调度与控制 cores within AI infrastructure. The global supply-demand mismatch, technological architectural innovations, and the rise of new application paradigms like AI Agents are collectively pushing CPUs towards a structural inflection point in demand. For the domestic capital market, this trend is particularly noteworthy. On one hand, server CPUs, as the foundation of AI computing power, are deeply benefiting from the dual drivers of data center expansion and domestic substitution. On the other hand, domestic CPU companies represented by Loongson Technology and Hygon Information have entered a period of performance realization,叠加 with continued policy support for IT application innovation. The sector is experiencing a triple resonance of "installed base upgrades + autonomous controllability + model iteration." After GPUs and memory chips, CPUs may become the next investment theme in the A-share semiconductor track with high certainty and high elasticity, and the investment opportunities therein deserve attention.
Comments