Liang Wenfeng stated China's AI computing power gap with the United States is primarily in resources, not talent, and domestic chip substitution has entered a historic window of opportunity. He estimated training an 800 billion parameter model would require approximately 50,000 NVIDIA GB300 chips or 200,000 Huawei 950 chips.