Podcasts

In this episode, '42 Chapters' invites Zhang Li Yang, the founder of Yunqin Capital, to deeply discuss the complex world of Mergers and Acquisitions (M&A). The podcast begins by clarifying that the essence of M&A is the change of corporate control and distinguishes between strategic M&A (aimed at compensating for business shortcomings) and fund M&A targeting financial returns. Through recent cases such as Zhiyuan Robotics, the value of listed companies' 'shell corporations' and related regulatory considerations are analyzed. The program details the core differences between M&A funds and VC/PE in terms of investment strategies, risk preferences, return expectations, and post-investment management, emphasizing that M&A funds focus more on a strong sense of ownership and achieving steady returns through financial engineering. Zhang Li Yang highlights how M&A can unlock greater productivity by reshaping organizational structures and incentives, and deeply explores management incentives, coordination of multiple parties' interests (including sellers, new shareholders, management, employees, government, etc.), and complex negotiation skills in M&A. Finally, combined with geopolitical trends, it provides AI entrepreneurs with practical advice on self-value recognition, valuation judgment, and communication with existing shareholders when facing M&A, emphasizing the importance of M&A in China's economic transformation.

This podcast features Zhu Yibo, co-founder of StepUp AI, providing an in-depth analysis of all aspects of artificial intelligence infrastructure (AI Infra). Zhu Yibo begins by defining AI Infra as a three-layer architecture comprising hardware (AI chips, network interface cards) and software (basic operation and maintenance, platform services, framework optimization). He points out that the advent of the large model era has moved AI Infra from behind the scenes to the core, making its importance comparable to Google's in the search engine era. It is key to processing massive data and high computing power demands. The podcast compares AI Infra with traditional Internet information processing, highlighting AI Infra's GPU-centric nature and its focus on achieving ultimate performance and customization. The discussion also delves into how AI Infra can improve model training efficiency and reduce inference costs through optimization, emphasizing the importance of shifting the optimization goal from training Model FLOPS Utilization (MFU) to decoding speed. The podcast also explores the AI Infra team's status within the company, its collaboration model with algorithm and hardware teams, and the organizational structure challenges currently faced. Finally, it looks forward to the future development of AI models, including breakthroughs in multimodality and opportunities for Chinese AI chips, emphasizing that AI Infra practitioners need a deep understanding of models and hardware to achieve vertical integration of hardware and software to cope with industry changes.