Goldman Sachs: China AI Competition Shifts to Harness Layer and Cost Efficiency

Goldman Sachs Research said at its second annual Asia Leaders Forum in Hong Kong on October 15 that as competition among Chinese AI models intensifies, investors are shifting focus from the sustainability of model performance leadership toward cost efficiency and the Harness layer.

Ronald Keung, Goldman Sachs Research's head of Asia internet research, noted that in many cases the performance gap between leading AI models, particularly in coding, is narrowing. Competition at the model layer has led investors to question how model developers can differentiate beyond performance, with key focus potentially falling on cost efficiency and the Harness layer—the ecosystem that carries and empowers AI models, enabling functions such as data access and workflow execution. Keung said Chinese AI labs are only beginning to invest in the Harness layer, having previously concentrated on the model layer, and that key investment in the next phase is expected to focus more on this layer.

Patrick Zhang, head of greater China technology research at Goldman Sachs Research, said that although bottlenecks remain, China's semiconductor industry is developing faster than market expectations, with leading domestic chipmakers' capacity expansion plans exceeding analyst forecasts.

Goldman Sachs Research also noted that China's open-weight AI models are being adopted globally and, in some cases, deployed on Western cloud computing systems. The forum attracted more than 1,400 investors and representatives from 400 companies across the AI industry chain.