China's AI Ambitions Face Data Shortage as Training Material Runs Low

A New Constraint on AI Development

China's artificial intelligence sector faces an emerging constraint that may prove more difficult to overcome than semiconductor restrictions: a looming shortage of high-quality training data. While export controls on advanced chips have captured international attention, AI researchers in China are increasingly concerned about exhausting the supply of quality Chinese-language text needed to train next-generation models.

This challenge affects the global AI industry, not just China. According to research from US-based Epoch AI, the worldwide supply of high-quality, publicly available human-generated text could be completely depleted within the next six years. The constraint applies across languages and borders, forcing AI labs everywhere to confront fundamental questions about how to continue improving their systems.

Why It Matters

Unlike chip shortages that can potentially be addressed through alternative suppliers, smuggling, or domestic manufacturing investments, data scarcity represents a more fundamental limitation. You cannot manufacture training data the way you manufacture semiconductors. This bottleneck could fundamentally reshape the competitive landscape in AI development, potentially limiting how much further large language models can advance regardless of computing power available.

The Data Wall Ahead

OpenAI co-founder Andrej Karpathy has warned of an approaching "data wall" by decade's end. Beyond this threshold, model capabilities could plateau without access to fresh, reliable information sources. The concern centers on exhausting the corpus of high-quality human-generated content—books, articles, websites, and other text that meets the standards required for training sophisticated AI systems.

For China specifically, the challenge carries additional weight. The volume of publicly available Chinese-language training material is substantially smaller than English-language resources. This asymmetry means Chinese AI developers may hit data constraints sooner than their Western counterparts, even as they work to close the capability gap with leading American models.

Industry Response

Major US AI laboratories are already responding aggressively to the impending shortage. These companies are investing heavily in mining offline human knowledge and exploring alternative data sources, according to the report. These efforts have sparked ethical debates about data acquisition practices and the boundaries of acceptable training material sourcing.

Chinese AI experts note that hardware workarounds—the kind that have helped domestic companies navigate chip restrictions—cannot easily solve data scarcity. While techniques like synthetic data generation and data augmentation offer partial solutions, they cannot fully replace the diversity and quality of authentic human-generated content.

These details were first reported by the South China Morning Post.