股价飙升近8%！百度昆仑芯被曝计划赴港IPO，估值近30亿美元

Core Viewpoint - Baidu's AI chip subsidiary Kunlun Core is accelerating its preparations for an IPO in Hong Kong, with plans to submit its application by Q1 2026 and complete the IPO by early 2027, following a recent valuation of approximately 21 billion RMB (about 2.97 billion USD) [1][4] Group 1: Company Developments - Kunlun Core has initiated its IPO preparations, aiming for a listing on the Hong Kong Stock Exchange by early 2027, with a recent funding round valuing the company at around 210 billion RMB [1][4] - The company has shifted from being an internal department of Baidu to an independent entity, focusing on reducing reliance on its parent company and expanding its customer base [5] - In the past two years, Kunlun Core has increased its external sales efforts, targeting clients beyond internet companies, including state-owned enterprises and local government data center projects [5] Group 2: Financial Performance - Kunlun Core is currently in a phase of "investing for scale," with projected revenues of approximately 2 billion RMB and a net loss of around 200 million RMB for 2024 [5] - The company anticipates that over half of its revenue will come from external clients by 2025, with expectations to exceed 3.5 billion RMB in revenue this year and achieve breakeven [5] Group 3: Market Trends and Opportunities - The enthusiasm in the capital market for AI chips is increasing, as evidenced by the recent IPO of domestic GPU company Moore Threads, which saw its stock price surge significantly on its debut [4] - The demand for high-performance chips in China is rapidly growing, providing a favorable growth window for companies like Kunlun Core [6] Group 4: Product Development - Kunlun Core is expanding its product lineup, having recently launched two new AI chips: M100, focused on inference scenarios, and M300, designed for both training and inference tasks, with launches planned for early 2026 and 2027 respectively [6] - The company aims to capture a share of the high-value AI training computing market, indicating its ambition beyond just edge inference [6]