To The Moon

KGI Securities Chairman Andy Chu stated that there are currently significant differences in the investment logic of the AI industry between China and the US, a divergence stemming from their respective industrial foundations and development paths. However, in the long run, the global AI industry is ultimately expected to converge toward Physical AI. KGI Chief Investment Officer Ken Leung supplemented this view, predicting that the US will maintain its leadership in AI hardware and core technologies, while China's AI development will focus more on the application layer. By making technology more "down-to-earth" and reaching consumers, China aims to build clear business models, which may represent its core pathway in AI. The divergence in technological foundations is causing a split in Sino-US AI investment logic at the current stage. Chu indicated that the path divergence is clear: investment in leading US AI companies still follows the "Scaling Law," meaning they continuously increase computing power investment to iterate better models. Consequently, their investment focus concentrates on core technology areas leading to Artificial General Intelligence (AGI), such as Large Language Models (LLMs) and underlying hardware like GPUs, as well as foundational model R&D. The "Scaling Law" is the foundational theory for large model development, providing a framework for quantitative prediction and resource optimization for model improvement. Specifically, it refers to how system performance, features, or behaviors change with scale, quantity, and resource investment, resulting in empirical mathematical规律. In the AI field, it specifically denotes the allocation规律 of model performance relative to resource data like parameter count, training data volume, and computational load. At present, the US primarily focuses on fundamental AI research, concentrating resources on breakthroughs in AGI and multimodality, with massive investments in underlying algorithms, complex reasoning, and basic research. China mainly concentrates on scenario adaptation; constrained by computing power, it emphasizes efficiency and industrial innovation, leading to strong performance in specific segments like autonomous driving, robotics, and embodied AI. KGI Chief Investment Officer Ken Leung anticipates that the US will maintain leadership in AI hardware and core technologies, while China's AI development will be more focused on the application layer, reaching consumers through technology implementation to build clear business models. The root cause of the divergence lies in differences in industrial foundation and development path. Analyzing the reasons, Chu explained that the US investment logic is dominated by the "Scaling Law," where leading advantages in computing power investment and model performance drive increased investment in underlying computing infrastructure. China's core advantage, however, lies in its vast pool of AI talent and enormous market space, which provides natural ground for AI application scenarios and determines China's ability to achieve breakthroughs from the application side. "For example, China possesses unique advantages for development in the robotics field, as the robotics industry requires deep accumulation of industrial production line technology, which is precisely China's core strength in manufacturing," he said. Leung used the development history of the internet industry as corroboration. Around 1995, Netscape launched its web browser, kickstarting the commercialization of the internet. Netscape once led the industry wave with its technological advantage but ultimately exited the stage. In stark contrast, the most profitable enterprises were those that survived market competition, like Facebook, Google, and Amazon, all of which excelled at the internet application layer and were companies that could deeply reach consumers. "Whoever can best apply the technology and get closest to the consumer will become the most profitable enterprise," Leung emphasized. The hype around the AGI concept is fading, and market focus is shifting toward Physical AI. Regarding industry trend judgments, Chu pointed out that the previously high-profile AGI concept is "being mentioned less frequently lately," and market focus in recent months has turned to Physical AI. AGI (Artificial General Intelligence) refers to systems possessing general cognitive abilities comparable to or surpassing humans. AGI is not limited to single tasks or specific domains but aims to simulate humans by autonomously learning, reasoning, planning, solving problems, and adapting to environmental changes in various unfamiliar scenarios, requiring sufficiently powerful and efficient algorithms, data, and hardware support. The relative concept is Artificial Narrow Intelligence (ANI), which is the current mainstream focus of the market. Examples include large language models, facial recognition systems, intelligent voice assistants, robotic arm robots, and financial quantitative trading systems. This category also includes Physical AI such as robots, robotic arms, and autonomous vehicles. Chu believes that the core application scenarios for Physical AI are precisely the robotics and autonomous driving fields that China is currently prioritizing for development, which also highly aligns with China's industrial advantages. The logic of global AI will ultimately converge on Physical AI. "At this stage, the US still has large-scale capital expenditure (CAPEX) in the chip sector," Chu said. Based on comprehensive market data, GPUs account for approximately 40-55% of the total server spending in 2024 by the four major US cloud service providers: Microsoft, Amazon, Google, and Meta. Currently, US GPU capital expenditure is in a phase of exponential growth, with the market expecting a compound annual growth rate exceeding 50% from 2024 to 2026. Jensen Huang predicts that global data center capital expenditure will reach $1 trillion by 2028, with GPU-related expenditure comprising over 50%. As AI applications expand from training to inference, GPU investment will continue to intensify. Over time, the global AI industry will increasingly focus on the Physical AI domain. This judgment also aligns with the major trend of the global AI industry shifting from basic R&D towards penetration into real-world applications. China's first-mover advantage at the application end is expected to become more prominent in the long-term competition. However, Chu believes that "the high proportion of GPUs in CAPEX is a short-term phenomenon, and the development directions of the AI industries in China and the US will ultimately achieve convergence." Currently, the US is also expanding its investment in embodied AI, supported by strategic policy initiatives. Agencies like the US Department of Defense, the National Science Foundation, and the Department of Energy are guiding funding towards key development areas in practical Physical AI applications such as humanoid robots, multimodal perception fusion, smart factories, and autonomous driving. Simultaneously, tech giants are making huge investments. Tesla has invested over $4 billion in its Optimus humanoid robot project, aiming for commercialization by 2027. NVIDIA has invested over $10 billion to build a full-stack Physical AI platform including the Cosmos model, THOR chip, and Omniverse simulation environment. Google's DeepMind has also invested $5 billion in robotics R&D, focusing on humanoid robot motion control and operational precision. The market expects total US investment in the Physical AI field to exceed $50 billion during 2025-2026. In the future, the US is expected to bridge the entire process from basic research to commercial implementation.