Goldman Sachs: China's open-source AI models have strong overseas potential

Ronald Keung, Goldman Sachs' head of Asia internet research, said in an interview with China Fund News that agentic AI is driving demand for cost-effective and smaller models, and that Chinese open-source models have huge potential for overseas expansion.

Keung noted that as AI model capabilities improve, market attention may shift from model rankings to cost-effectiveness and Harness deployment. Goldman Sachs Research estimates that token usage in China's AI model market will grow 25-fold by 2030.

He also said Chinese companies have performed well in balancing training and inference efficiency despite computing constraints, with many achieving results close to leading global models using relatively limited computing power. The industry increasingly recognizes that Harness is as important as the model itself, and its data can feed back into model iteration.

On overseas expansion, Keung said some Chinese model companies are generating revenue through commercial licensing partnerships abroad, where computing power is provided by local cloud vendors. Open source offers greater flexibility for training and deployment, and more Chinese AI model providers may combine open-weight releases with commercial licensing in the future.

Based on Goldman Sachs' competitive positioning framework for AI model companies, Keung said independent AI model companies generally stand out in pricing power and cost advantages. Companies with the largest ARR, gross margin advantages, and financial strength are likely to be long-term winners.