China's AI providers are poised to invest a staggering $70 billion in data centers as they expand overseas, according to Goldman Sachs Research. Since the January headlines surrounding China's DeepSeek AI chatbot, the country's cloud computing and internet industries have been in a frenzy, racing to develop AI for both consumers and businesses. Our analysts predict a 25% increase in power demand from China's data centers this year alone. Chinese AI hyperscalers are currently catering primarily to domestic customers, relying on homegrown AI chips and hardware. However, they are also generating subscription revenue from AI applications and eyeing further growth opportunities among foreign customers.
The growth of AI in China is exponential. A key indicator is the power consumption of new data centers to support AI. Electricity capacity for data centers in China is set to surge by 30% this year, reaching 30 gigawatts. One prominent cloud computing company and major AI player plans to expand its data center capacity tenfold by 2032. We forecast that China's cloud service providers will boost their capital expenditures by around 65% in 2025.
Next year, top internet firms in China are expected to invest over $70 billion to support AI initiatives. This is only 15-20% of what US hyperscalers are projected to spend, but the Chinese sector is in a 'build it and they will come' phase. These initiatives span from data centers to novel chip types to AI functions, creating a broad and steep growth curve.
The Chinese AI infrastructure is becoming more self-sufficient as hyperscalers adopt a multi-pronged approach to developing and leveraging domestic chips for AI. This marks the growth of China's AI supply chain, where companies have traditionally spent 50-75% of their capital expenditures on foreign chips. Now, the ratio is shifting in favor of domestic producers.
Chinese AI hyperscalers generate revenue through subscription fees for AI applications and charging enterprises for 'post-training' and 'inference' services. Post-training involves companies customizing AI models to behave in specific ways with their proprietary data to serve their customers and grow their businesses. Inference refers to the application of AI models in day-to-day tasks, taking in new user inputs to generate responses.
While Chinese hyperscalers generate 90-95% of their revenue from domestic businesses, international growth is another key driver. Anticipating future growth, especially from overseas markets, Chinese AI hyperscalers plan to construct data centers across Asia, the Middle East, and Latin America. Currently, AI video generation, picture editing, and object identification are the top monetizable use cases for China's AI applications, with overseas markets already contributing more revenue than domestic markets in these areas.
These services typically monetize through monthly subscription fees, and monitoring the rate of 'annual recurring revenue' (ARR) is a key metric to track business growth for AI model and application providers. However, it's early days in the revenue story, given the scale of capital expenditures in the sector. This article is for educational purposes only, and Goldman Sachs does not provide financial, economic, legal, investment, accounting, or tax advice.
