After the Hong Kong stock market AI revaluation heats up, who is the true "Chinese version of Palantir"?

Recent trading sentiment in the Hong Kong AI sector is very direct: on one side, pure model companies are being lifted by capital, with Minimax (00100) seeing a nearly 90% surge in March and Knowledge Atlas (02513) also rising 72% and hitting a new high, quickly breaking away from traditional valuation frameworks. On the other side, enterprise-level AI application companies are seeing a recovery after performance disclosure, with one AI data governance company with revenue over 1 billion RMB seeing its stock price triple in a month. Looking at the short-term stock price performance, the former seems to be trading on imagination, while the latter appears to be trading on fundamental improvement. However, a more forward-looking question in the capital market is: who is most likely to take AI from "able to answer" to "able to execute" and form long-term barriers within complex organizations? This is why Zhizhi Technology (02706) deserves to be re-discussed in a higher framework. Looking at market sentiment itself, a report in March from the UK's Financial Times pointed out that the two best-performing AI new stocks in Hong Kong this year have seen their share prices rise by over 400%, driving IPO and refinancing volumes in Hong Kong to highs not seen since 2021. Jason Lui, head of stock and derivative strategies at BNP Paribas in the Asia-Pacific region, mentioned in the report that in 2025 investors will be buying components of large Chinese tech indices, but by 2026, the market will start actively seeking companies with "pure AI exposure." This indicates that the pricing logic of capital towards Chinese AI has shifted from general tech stocks to more vertical and pure AI assets. This sentiment is first reflected in basic model and intelligent body platform companies. Their advantages are simple logic, clear themes, and international funds that are easy to understand: big models, Agents, autonomous evolution, platform entry. These keywords naturally fit the market environment of risk preference enhancement. From the perspective of the secondary market, the rise of these companies is reasonable because they occupy the imaginative space of the "ability ceiling" of Chinese AI. Another type being repriced is enterprise-level AI application companies. Public data shows that a certain Hong Kong enterprise-level large-model AI application company achieved revenue of 415 million RMB in 2025, a year-on-year increase of 70.8%; adjusted net losses narrowed by 71.4% year-on-year. Among these, AI solution revenue reached 254 million RMB, a significant increase of 181.5% year-on-year, which has become the largest revenue source. In early April, the stock rose by over 40% in a single day, driven by the market's re-expectation of its "gradually clear profit turning point." From the perspective of the capital market, these companies are also worthy of recognition. Because they prove an important point: enterprise customers are willing to pay for AI, the key is whether AI can enter real processes, take on specific responsibilities, and not just stay at the display and question layer. In other words, the commercialization path of enterprise-level AI is being validated, the market is giving higher valuations for it, and this is not a problem. However, the problem lies precisely here. Whether it is a pure model platform or an enterprise AI application, the majority of companies are currently closer to the "reasoning layer" or "interaction layer". They can generate content, assist in decision-making, connect parts of processes, but the real difficulty lies in how to make AI understand the world in complex organizations according to rules, reason within boundaries, and execute according to processes. This step determines whether enterprise-level AI is just a smarter interface or can become a true productive system. The scarcity of Zhizhi Technology lies in being closer to the latter. According to Zhizhi Technology's prospectus, the company's core product is not a single model or Agent, but a combination of the DMC Data Intelligence Platform, Atlas Knowledge Graph Platform, AtlasGraph Graph Database, and the Atlas Intelligent Body Solutions built on top of these. In 2025, the company achieved revenue of 621 million RMB, a year-on-year increase of 23.4%; adjusted net profit of 24.15 million RMB, a year-on-year increase of 42.6%; overall gross margin increased to 43.3%. Among these, Atlas Intelligent Body Solutions revenue reached 145.7 million RMB, a year-on-year increase of 68.4%, with a gross margin of 53.2%. Looking at revenue growth alone, Zhizhi may not be the most aggressive in the sector, but when measured by "technical depth" and "control over execution layers," it is closer to the type of company that the capital market is willing to assign long-term high premiums to. The reason is that the core logic of Zhizhi Technology is not just a simple "graph + large model," but ontology + graph model integration. These two words may seem technical, but they actually correspond to the most critical value of enterprise-level AI. The strength of the general large model lies in generalization, induction, and language reasoning; its problem is that it is easy to lose boundaries in complex business environments. The internal environment of an enterprise is not an internet language corpus but a system composed of roles, objects, permissions, relationships, processes, time sequences, and business rules. Who reports to whom, which customer corresponds to which product line, which approval path can be taken, which conclusions must be verified back in the database, which actions can truly be executed by the systemthese are not stable problems that a model "good at talking" can solve. The significance of graph databases, knowledge graphs, and ontology modeling lies precisely here. They first structurally express entity relationships and business rules in the enterprise world, and then allow models to reason within this delimited, explained, and traceable scope. In other words, Zhizhi Technology is not trying to remedy after the conclusion of the large model but is setting boundaries, providing structure, and establishing paths for the model before reasoning occurs. The result of this approach is not only reducing illusions but more importantly enhancing the controllability and accuracy of execution. This is where Zhizhi is "deeper" than many upper-level AI application companies. Many enterprise-level AI products are essentially doing process encapsulation, enhanced question answering, or task arrangement, which have value, but the control points mainly remain at the front-end interaction and light collaboration layers. On the other hand, Zhizhi's graph ontology and graph model integration bring the control point deep into the enterprise knowledge structure and task execution logic itself. It does not just make AI "smarter" but allows AI to "know what to believe, what to connect, which step to take, and ultimately what to achieve" in a complex environment. This is very close to the core methodology of Palantir. Palantir has been given a high strategic valuation by the capital market not just because it has the concept of AI, but because it can truly connect heterogeneous data, relationship networks, task processes, and decision execution to form an enterprise- or institution-level operating system. Following this standard, what Zhizhi Technology holds is not just a single application entrance but a more scarce infrastructure of the execution layer that truly integrates graph ontology, knowledge structure, reasoning boundaries, and execution loops. Public data also indirectly supports this point. According to Zhizhi Technology's prospectus citing Frost & Sullivan data, based on revenue in 2024, the company is ranked first in China as a "graph-centric AI intelligent body provider," with a market share of approximately 50%; the market size of "industrial-grade AI intelligent bodies integrated with knowledge graphs" is expected to grow from 200 million RMB in 2024 to 13.2 billion RMB in 2029. This means that Zhizhi is occupying not a crowded AI application track but a segment of the market space that is still opening up, with clear technological barriers. Therefore, if the first type of AI companies represents the elasticity of Chinese modeling capabilities, and the second type of enterprise AI companies represents the beginning of commercialization, then what Zhizhi Technology represents is the direction where AI can truly yield long-term value in the next stage: not making models more human-like but making models truly function as systems within organizations. The capital market always chases after the brightest concepts first, then gradually returns to the strongest barriers. This is also the case for the AI sector. Pure models are worth valuation, enterprise-level applications worth reconsidering, but if one were to find a Hong Kong stock that is more like a "Chinese version of Palantir," then perhaps it's worth seriously looking at the company that truly integrates graph ontology, knowledge structure, reasoning boundaries, and execution loops. Along this line of thought, the value of Zhizhi Technology may not have been fully realized by the market yet.