HAIZHI TECH GP (02706) made its trading debut on the Hong Kong Stock Exchange. According to the listing announcement, the final offer price was set at HKD 27.06 per share. The company issued a total of 28,030,200 shares, with a board lot size of 200 shares. The net proceeds from the global offering amounted to approximately HKD 655 million. A notable highlight was the significant investor interest, with the Hong Kong public offering portion being oversubscribed by approximately 5,065.06 times. At the time of writing, the stock price had surged 225.39% to HKD 88.05, with a turnover of HKD 377 million.
Public information indicates that HAIZHI TECH GP is a leading enterprise in China's graph computing and graph database sector. The company specializes in developing its Atlas Knowledge Graph solutions and industrial-grade intelligent agents through its graph-model fusion technology, providing industrial-level artificial intelligence solutions. The Atlas Knowledge Graph solution, one of the company's core products during its track record period, is built upon its industry-leading graph computing and data analysis capabilities. It comprises the DMC data intelligence platform, the Atlas Knowledge Graph platform, and the AtlasGraph graph database.
With the gradual expansion of large AI model applications, HAIZHI TECH GP has experienced rapid growth in recent years in its related hallucination-reduction revenue, specifically from its "Atlas Intelligent Agent" product. Revenue from this segment saw an increase of nearly nine times in 2024 compared to 2023, followed by a further increase of nearly five times in the first half of 2025. According to data from Frost & Sullivan, HAIZHI TECH GP is recognized as the first company in China to effectively reduce hallucinations in large models through knowledge graphs. The company has established core competitiveness and technological barriers across three key dimensions: factual accuracy, hallucination rate, and complex logical reasoning capability, successfully mitigating hallucination issues in large language models.