Author: CITIC Securities

On November 12, U.S. time, NVIDIA's board announced that Jensen Huang resigned from his position as President, CEO, and Director of the company, effective immediately, with CFO Colette Kress serving as interim CEO; since November 5, NVIDIA's stock price has cumulatively fallen over 22%. We believe that after management changes at leading overseas companies, the likelihood of changes in their clients, supply chains, and overall strategies increases; meanwhile, according to The Information, the U.S. Department of Commerce is finalizing new AI chip export control rules, planning to include countries such as those in Southeast Asia under stricter export licensing management, making it increasingly difficult for NV series cards to enter the domestic market, and concerns about Chinese CSPs acquiring computing power may lead to short-term performance expectation disturbances. In the short term, stricter export controls are expected to affect the progress of overseas computing power center construction, reducing visibility into CSP growth; in the medium to long term, it is anticipated that domestic CSPs will increase their proportion of procurement for domestic computing power. With the iteration of domestic products and the enhancement of supply capabilities, domestic computing power is expected to become a crucial pillar of domestic AI-related Capex, capturing a significant medium-term share.

Events:

Jensen Huang resigned from all positions at NVIDIA, with CFO taking on the role of interim CEO. On November 12, U.S. time, NVIDIA's board announced: Jensen Huang resigned from his positions as President, CEO, and Director of the company, effective immediately; Executive Vice President and CFO Colette Kress will serve as interim CEO while continuing as CFO; the board initiated the selection process for a new CEO and appointed Harvey C. Jones as the company's first Board Chairman. As a co-founder of NVIDIA, Jensen Huang has led the company for 33 years, and his deep ties to the company's strategy and client relationships mean that the impact of this management change will far exceed typical personnel changes.

Domestic Semiconductors and Computing Power Chain: Replacement Logic Fully Strengthened, Super Nodes Drive Domestic Chain into Performance Release Period

After management adjustments at leading overseas companies and alongside the iteration of domestic products and enhancement of supply capabilities, domestic computing power is expected to become a crucial pillar of domestic AI-related Capex, capturing a significant medium-term share. On the demand side, the difficulties of NV series cards entering the domestic market continue to increase; we estimate that the stockpiled computing power load in China will be 30.4GW in 2026, expected to reach 97.4GW by 2030. The state-owned system, credit costs, and infrastructure carrying conditions are favorable, and China's computing power construction remains in a clear acceleration phase. Under the policy requirements for domesticization rates, breakthroughs in performance and increases in production capacity for domestic AI acceleration cards are crucial. As this situation evolves, domestic computing power chip manufacturers are likely to receive more support for supply. In addition, the ramp-up of advanced process production lines in domestic manufacturing and improvements in yield are also a solid support for the volume increase of domestic AI computing power cards. In terms of industry rhythm, it is expected that from the end of this year to next year, domestic super nodes represented by Alibaba and ByteDance will accelerate their landing, and both the supply and demand for domestic cards are expected to increase significantly. The connection phase of super nodes is expected to see a large explosion in value, and the domestic chain is expected to gradually release performance starting from Q4 2026.

Overseas Computing Power Chain: Both Investment and Financing Already Under Pressure, Management Change Adds Disturbance, Maintain Caution towards Overseas Chain in the Short Term

We estimate that the stockpiled computing power load globally (excluding China) will be 69.3GW in 2026, and at the current pace is expected to reach 322.4GW by 2030. However, under rapid growth, North America's computing power construction faces challenges on both the investment and financing ends: on the investment side, AI capital expenditures are expected to peak as a single industry proportion of GDP at the macro level, while at the micro level, CSP terminal revenues are only 1/7 of the investments, making breakthroughs on the application side key; on the financing side, considering the CSP financing gap and the demand for new cloud vendors, it is expected that funding pressure will remain relatively controllable until 2027, but gaps will become more significant after 2028. In this context, Jensen Huang's departure will weaken NVIDIA's ecological integration and credit guarantee capabilities, significantly increasing the likelihood of changes in NVIDIA's clients, supply chain, and overall strategy, triggering a slowdown in technology iteration across the overseas computing power chain.

Domestic CSP: Increased Difficulty in Acquiring Computing Power Disturbs Short-Term Performance Expectations

According to The Information, the U.S. Department of Commerce is finalizing a new AI chip export control rule, planning to include countries such as those in Southeast Asia under stricter chip export licensing management.

In the short term, stricter export controls are expected to affect the progress of constructing advanced computing power centers, affecting the continuity of growth. The visibility of CSP growth in the short term is reduced, but considering that the demand for computing power has not changed significantly, CSP margins are expected to improve significantly under supply constraints, while the ROIC of previously purchased chips will further increase. Secondly, for model companies, exploring the intelligent limits of frontier models still relies heavily on support from advanced computing power. With limited computing power supply, we believe model companies will further optimize their training and inference AI infrastructure, promoting cost reduction and efficiency enhancement.

In the medium to long term, we believe that the procurement proportion of domestic CSPs for domestic computing power will further increase, and full-stack platforms with strong heterogeneous computing power capabilities are expected to achieve higher efficiency utilization of domestic computing power and capture more market share.

AI Applications: Chip Structure Remains Good, Application Diffusion is the Main Increment for 2027 Model Commercialization

In our annual strategy, we judge that the certainty of AI applications is increasing: AI coding has already crossed the critical point and settled as a universal base for intelligent agents, leading products show explosive growth in ARR, and the daily workload of agents at OpenAI now equates to 3.1 human workloads; on the enterprise AI front, corporate AI spending and LLM application rates continue to rise, with local deployments dominating enterprise AI demand, top vendors driving a "deploy-feedback-iteration" flywheel with front-line deployments (FDE); in the realm of personal intelligent agents, products like Qianwen Go and Meituan Xiaotuan have entered the real task execution stage, and WeChat has opened two AI access modes supporting the use of mini-program services; AI4S and sector-specific AI are also entering an explosive phase.

In the media sector, we remain optimistic about the prosperity of AI content due to the iteration of domestic video model capabilities and cost reductions, especially with AI long dramas and AI movies set to launch a batch of quality content. We are particularly focused on the industry catalyst brought by the December release of the ByteDance world model; on the platform side, WeChat accelerates its entry into the short drama sector, and upstream IP and content providers are expected to enjoy benefits phase-wise, with a focus on companies holding quality IP and content. From the perspective of chip structure and trading volume shares, currently, the media sector has a considerable space compared to the first half of 2023 and early 2026, and the chip structure of the sector remains good. Fluctuations in other directions from a funding perspective are favorable for the subsequent rise of the media sector.

Investment Strategy:

From a配置 perspective, the A-share AI hardware industry chain currently does not provide sufficient odds to hedge against uncertainty, and actively managed public TMT holdings remain at 98.4% of the percentile since 2010. The sector will shift from being driven by incremental funds to a game of stock or even reduction, and the restoration of the funding ecosystem will take time, with momentum and prosperity strategies weakening and rotations accelerating.

Within the technology sector, we recommend capturing structural opportunities in the following order: 1) Domestic computing power direction, adjustment of overseas management and export control strengthens the logic of domestic sourcing, with the domestic chain expected to gradually release performance starting from Q4 2026; 2) Physical AI direction, with giants entering the space and 2027 heralding the year of implementation, focusing on orders from the embodied intelligence supply chain and the mass production verification of L4 and L3 autonomous driving; 3) AI application diffusion direction, where AI coding has already crossed the critical point, and AI film, AI4S, enterprise-level AI, and personal intelligent agents are entering task execution phases one after another, focusing on the diffusion of application scenarios and commercialization realization.

Additionally, we pay attention to the broad structural repair opportunities of non-AI sectors: the non-AI sector has previously endured the pressures of material price increases, rising long-end interest rates, and foreign exchange losses due to AI's negative externalities. The major process of valuation contraction has already been completed this year. As AI investment slows, interest rates and material prices decline, and the pace of renminbi appreciation slows, K-type differentiation is expected to converge. We recommend a robust return strategy represented by PB-ROE (focusing on home appliances, logistics, energy and chemicals, oil transportation, pharmacy, overseas brokers, and financially robust property leaders), as well as a reverse layout and expectation enactment strategy represented by domestic consumption (focusing on food and beverages, catering supply chains, airport duty-free, health products, leading medical devices, etc.).

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。