รัฐบาลจีน ‘เบรก’ ดีล Meta ซื้อกิจการบริษัท AI 'Manus' มูลค่า 2 พันล้านดอลลาร์ | THE STANDARD WEALTH

Chinese AI is rapidly advancing and challenging Silicon Valley's business models, despite limitations in accessing advanced chips and massive investment budgets. Unlike competitors like OpenAI, Chinese developers such as DeepSeek and Alibaba Group Holding prioritize value and efficiency over sheer computing power. Their strategy focuses on creating AI models that achieve similar results to leading global models but use significantly fewer computer resources. A key technique employed is a "mixture of experts," where only a portion of the neural network is activated for processing, leading to substantial savings in energy and costs. For instance, Fig's latest model operates using less than 3% of all parameters each time, a stark contrast to US models that often consume more resources for advanced capabilities. While US models may still surpass Chinese counterparts in complexity and data support, the cost factor is a crucial differentiator. The operational cost of OpenAI models can be dozens of times higher than that of Chinese alternatives, making Chinese AI an attractive option, particularly in markets prioritizing cost over peak performance. Another strategy for efficiency in China is the "OpenWeight" concept, which grants university developers and even small companies free access to AI model parameters. This open approach, unlike the proprietary nature of US companies, decentralizes development and accelerates innovation. Alibaba Group Holding's Mod Quent, for example, has surpassed 1 billion downloads and spawned over 200,000 sub-models globally. Energy is a significant cost in AI development, and China leverages its government structure to invest in rapid electricity production capacity, providing financial support for data centers in areas with affordable renewable energy. This has resulted in lower AI development costs compared to many other countries. Furthermore, China is applying AI extensively in manufacturing, logistics, and other industrial sectors to boost economic productivity, diverging from the US focus on the service sector. Companies like electric vehicle manufacturer BYD and New Fox are using AI to reduce waste, increase efficiency, and improve product quality. This progress aligns with a national strategy where the Chinese government views AI as a tool for economic growth rather than solely for maximizing private company profits. Government support encompasses funding, infrastructure, and initiatives to drive AI adoption across all societal sectors. China is also expanding its global influence by exporting AI technology to Southeast Asia, the Middle East, and Africa, offering comprehensive, cost-effective AI platforms and cloud options, including related infrastructure. This has led to a rapid increase in China's AI market share. Although US and European markets still have security and privacy concerns regarding Chinese companies, cost pressures may prompt some users to explore Chinese alternatives, which can deliver up to 90% of competitors' performance at a much lower price. Recently, the Chinese government unexpectedly halted the $2 billion acquisition deal of AI startup Minus by Meta, citing concerns over technology transfer. This decision, made by China's National Development and Reform Commission (NDRC), has sent shockwaves through the rapidly growing Chinese AI industry and occurred just weeks before high-level meetings between Donald Trump and Xi Jinping. While the sale was initially seen as a model for startups seeking global expansion, domestic critics viewed it as a loss of key technology to a geopolitical competitor. The founder of Minus noted that the company started in China before moving its main team to Singapore. This incident underscores China's willingness to interfere in transactions, even those initiated outside the country, to protect its strategic interests in AI. Overall, China's AI strategy emphasizes efficiency, price competitiveness, and decentralized development to achieve rapid progress.