China Galaxy Securities has released a research report stating that as the capabilities of domestic large language models continue to improve, the "hundred-model battle" has entered an elimination phase, potentially leading to token inflation. While native large model companies, due to their business purity, hold an advantage over traditional internet giants and have garnered short-term capital favor, the competition for traffic portals in the AI era remains essential for traditional internet behemoths. Traffic is still one of the most crucial assets in the AI era. In the long run, internet companies are expected to leverage their traffic advantages to reclaim their status in the AI era, with native large model manufacturers likely becoming a vital part of their industrial chain. It is recommended to continuously monitor the low-position layout opportunities for Hang Seng Internet technology giants.
The main views of China Galaxy Securities are as follows: Domestic large models are entering a phase of token inflation, and the competition for AI portal traffic among internet tech giants is intensifying. In February, the Hang Seng Tech Index fell sharply by 10.15%, marking its largest monthly decline so far this year. Given that seven of the top ten heavily weighted stocks by market capitalization in the Hang Seng Tech Index are internet technology companies, traditional internet tech firms are currently facing a battle for AI traffic portals. The market is concerned about whether short-term traffic narratives can translate into actual performance, leading to uncertainty about the future profit expectations for these internet tech companies. Coupled with the U.S. Federal Reserve's more hawkish-than-expected stance, raising concerns about a turning point in global liquidity, this triggered a significant sell-off in Hong Kong-listed internet companies, resulting in the Hang Seng Tech Index's substantial February decline.
In contrast, recent performances of Hong Kong-listed native model companies like KNOWLEDGE ATLAS (02513) and MINIMAX-WP (00100) showed substantial gains of 154.2% and 64.42% respectively. According to data from OpenRouter, model calls for Chinese AI experienced explosive growth in February, surpassing those from the United States for the first time. A weekly ranking from February 16 to 22 showed that four of the top five models by platform calls were from Chinese manufacturers: MiniMax M2.5, Moonshot AI's Kimi K2.5, KNOWLEDGE ATLAS's GLM-5, and DeepSeek's V3.2. These four models collectively accounted for 85.7% of the total calls among the top five. Concurrently, KNOWLEDGE ATLAS announced price adjustments for its GLM Coding Plan packages, eliminating acquisition discounts while retaining quarterly and annual subscription discounts, resulting in a structural price increase starting from 30%. Prices for existing subscribers remain unchanged.
China Galaxy Securities believes that as domestic large model capabilities continue to advance, the competitive landscape will lead to token inflation. Native model companies currently hold an edge, but the necessity for traditional internet giants to compete for AI traffic portals remains. Traffic is a key asset. Long-term, internet companies are poised to use their traffic strength to regain prominence, with native model makers becoming integral to the ecosystem. Continued attention on low-position entry points for Hang Seng Internet giants is advised.
Agent capabilities are making significant leaps, potentially reshaping the SaaS industry landscape. AI startup Anthropic launched 11 new plugins in early February, including a legal plugin capable of assisting teams with contract review, compliance risk flagging, and tracking legal clauses, demonstrating capabilities akin to a junior lawyer. This has led investors to question the moats of traditional legal tech companies, contributing to declines in SaaS stocks both domestically and internationally. China Galaxy Securities posits that AI's current greatest advantage lies in programming, which may impact traditional software companies in terms of feature and interface development, potentially leading to an industry reshuffle. The report suggests focusing on three types of companies: 1) Those possessing specialized, proprietary data resources in vertical industries; 2) Companies deeply embedded in physical systems, production processes, or highly complex scenarios and their corresponding environments; 3) Companies actively advancing software agentification.
Risk warnings include: technology iteration falling short of expectations; intensified competition among tech giants; legal and regulatory risks; supply chain risks; and downstream demand failing to meet expectations.