CITIC Securities: AI Disrupts the Narrative of U.S. Internet Stocks, Short-term Overinterpretation

China International Capital Corporation (CICC) research notes that the AI-driven narrative about the U.S. stock internet sector is being excessively overplayed in the short term. In consumer scenarios, incremental AI experience is limited; AI replacement is constrained by cost; and model companies themselves have capability boundaries. Therefore, AI is more likely to be a cooperative relationship rather than a replacement relationship with existing internet platforms, and some high-quality companies have been clearly mistakenly sold off. CICC recommends focusing on companies that have competitive moats for the AI era—such as links to the physical world, strong network effects, accumulated data and algorithms, and high-quality content IP—while also looking for directions where demand expands as AI penetrates.

Full text below

Outlook｜U.S. internet investment logic in the context of AI Agents

The AI narrative about disrupting U.S. stock internet companies is being excessively overplayed in the short term. In consumer scenarios, the incremental value of AI experiences is limited; AI substitution is subject to cost constraints; and the model companies themselves have capability boundaries. Therefore, AI is more likely to coexist with existing internet platforms rather than replace them, and some high-quality companies have been clearly and unfairly hit. We suggest focusing on companies that have competitive barriers in the AI era—such as links to the physical world, strong network effects, accumulated data and algorithms, and high-quality content IP—while also finding directions where demand expands as AI penetrates.

▍AI progress is rapid, but it is not all-powerful.

Over the past two years, real breakthroughs have occurred in three major bottlenecks that constrain the large-scale application of Agent models: model reliability, inference costs, and ecosystem interoperability. This has driven deeper productization and deployment of AI across core internet scenarios such as e-commerce, advertising, and content. At the same time, market anxiety about existing internet platforms potentially being upended has been growing. We believe the actual impact of AI on the market will be milder than market expectations. The limiting factors come from three levels:

1）The incremental AI experience in consumer internet scenarios is being overestimated. Unlike the B端, in C-end decision chains such as shopping, commuting, and dining, there are limited repetitive tasks that AI can significantly improve. Moreover, core issues like information completeness, payment security, and platform reliability have not been effectively solved under existing AI frameworks;

2）The comprehensive cost of AI replacement creates constraints. It means current platforms cannot be comprehensively rebuilt. After overlaying implicit costs such as subscription fees, maintenance investments, and interface integrations, the cost-performance ratio is significantly worse than directly connecting to mature ecosystems;

3）Capability boundaries within large model companies themselves and the lack of economies of scale mean the probability of large model companies cooperating with vertical platforms rather than replacing them is higher. The profit pressure after listing will also force them to focus on core capabilities. Therefore, we suggest taking an objective and calm view of AI’s impact on the internet sector. What AI brings is structural differentiation rather than a systemic overthrow.

▍Finding high-certainty, structural incremental opportunities in the AI era.

At the impact level, on the one hand, AI brings quantifiable business increments to internet platforms through three paths: improving user experience, optimizing operational efficiency, and opening up new business scenario opportunities. On the other hand, the substitution effects of AI-native products on traffic entry points such as traditional search and content distribution are becoming visible.

At the barrier level, against the backdrop of a mixed picture of both positives and negatives, whether platforms can extend their competitive moats into the AI era is key to judging investment value. We believe companies with links to the physical world, strong network effects, accumulated data and algorithms, and high-quality content IP will stand out amid the AI wave.

At the opportunity level, through traffic diversion and disintermediation, a considerable portion of commercial value will be captured at the model layer rather than flowing to existing platforms. However, some internet platforms still have opportunities to share in AI dividends. Their common feature is that they are outside the scope of direct substitution by AI, but service demand will expand systematically as AI penetration increases. Advertising platforms, livestreaming/media platforms, and cloud computing are typical examples. We classify the key covered U.S. internet companies into four quadrants—beneficiaries, divergence points, safe havens, and losers—based on two major elements: competitive moats and AI opportunities. Among them, some companies are more likely to benefit from this round of the AI wave.

▍Risk factors:

Risk that AI progress far exceeding expectations will lead to stronger-than-expected shocks; risk that AI infrastructure investment is too heavy and returns are uncertain; risk that AI disrupts the content ecosystem; risk that market competition intensifies; risk that regulatory policies related to data and platform operations are tightened, etc.

▍Investment strategy:

We believe that AI has achieved breakthroughs in key bottlenecks and is accelerating deployment, but the narrative of AI overturning the internet is overly pessimistic. This is mainly driven by limited incremental AI experience in consumer scenarios, cost constraints on AI replacement, and capability boundaries at large model companies. We suggest focusing on companies that have competitive moats in the AI era—such as links to the physical world, strong network effects, accumulated data and algorithms, and high-quality content IP—while also finding directions in which demand expands as AI penetration increases.

（Source: Jiemian News）