智谱通过港交所聆讯并正式递交招股书，有望成为“全球大模型第一股”

Core Viewpoint - Beijing Zhipu Huazhang Technology Co., Ltd. (Zhipu) is set to become the "first global large model stock" by submitting its IPO application to the Hong Kong Stock Exchange, marking a significant milestone for China's independent large model industry [1][2]. Financial Performance - Zhipu's latest pre-IPO valuation stands at 24.377 billion yuan. The company has shown impressive revenue growth, with projected revenues of 57.4 million yuan, 125 million yuan, and 312 million yuan for the years 2022 to 2024, reflecting a compound annual growth rate (CAGR) of 130% [3][4]. - The gross profit margin has remained above 50%, with specific margins of 54.6%, 64.6%, and 56.3% for the years 2022 to 2024, and 50% for the first half of 2025 [3][4]. Research and Development - Zhipu has made substantial investments in R&D, with expenditures of 84.4 million yuan, 529 million yuan, and 2.195 billion yuan from 2022 to 2024, and 1.595 billion yuan in the first half of 2025, totaling approximately 4.4 billion yuan in R&D investments [4][5]. - The company has achieved significant technological advancements, with its GLM series models undergoing iterations every 3 to 6 months, and ranking first in coding capabilities alongside global leaders like OpenAI in the Code Arena competition [4][5][6]. Market Position - Zhipu has been recognized as a major competitor by OpenAI, highlighting its growing influence in the global AI landscape [6][7]. - The ChatGLM model was acknowledged as an outstanding representative of China's foundational large models by the prestigious journal Nature in May 2024 [5][6]. Shareholding Structure - The major shareholders of Zhipu include Beijing Chainpai and key individuals such as Liu Debing, Tang Jie, and Zhang Peng, collectively holding approximately 33.03% of the company's shares [7][8]. - Liu Debing, the co-founder and chairman, has played a pivotal role in the company's strategic direction and technological innovation, while Zhang Peng focuses on knowledge graphs and large-scale pre-training models [8][9].