Lates News

Recently, the technology track represented by optical modules and domestic computing power has once again become the focus of capital. Yang Lingfeng, manager of Rongtong Technology Selected Fund, believes that the continuous expansion of the application scenarios of large AI models is driving a significant increase in Token consumption, and the industry's computing power supply has entered a relatively tight situation. The high prosperity of the AI industry chain has a solid industrial foundation and clear growth logic. In his view, the AI industry consists of three closely related links: computing power, models (algorithms), and applications. From the industry perspective, AI is still in the investment phase, with strong demand for computing power, and the pace of application landing and performance realization is the fastest in history. Yang Lingfeng analyzed that early investment in training computing power by model manufacturers was relatively high, and subsequent investment growth is relatively stable. With the rapid increase in AI penetration rate, inference computing power is showing exponential growth. According to NVIDIA's calculations, by 2024, the demand for inference computing power will account for 40% of the total computing power demand, and it is expected that by 2025, inference computing power will officially surpass training computing power. (China Fund News)