China Internet Sector: APAC Focus: How China‘s AI models compete on cost efficiency

Research evidence excerpt

China Internet Sector: APAC Focus: How China‘s AI models compete on cost efficiency

Global Research

16 June 2026ab

China Internet Sector Equities

ChinaAPAC Focus: How China's AI models compete

on cost efficiencyPlease Internet Services

Wei Xiong

Analyst

wei.xiong@ubs.com

+86-21-3866 8883

Kenneth Fong

kenneth-kc.fong@ubs.com

+852-3712 3890

Charles Chen

Associate

S1460524020001

charles-za.chen@ubs.com

+86-21-3866 8907Inthe global AI race, model intelligence is key to assessing competitiveness and Sardonna Fong

commercialisation. However, as AI adoption shifts to scaled deployment, we expect Analyst

cost efficiency to become increasingly important. Through targeted R&D, sardonna.fong@ubs.com

+852-3712 3042

architectural innovations, engineering optimisation and open-source collaboration,

we believe China's companies are rapidly closing the performance gap with global

leaders at markedly lower costs, making them well positioned to gain share in the

sizeable, expanding global AI TAM. We highlight MiniMax, Zhipu, Alibaba,

Tencent and Baidu under our coverage as major companies in this theme,

together with private companies DeepSeek, Kimi, ByteDance and StepFun.

Assessing the cost leadership of China's AI models

We estimate the training costs for China's models (using MiniMax and Zhipu as

examples) are less than 10% of those of global leaders such as OpenAI and Anthropic,

while the average API price of major China models is below 20% of comparable global

peers. However, China's frontier models are closing the capability gap with rapid

iteration and improving intelligence, gaining global traction in token usage while

preserving healthy gross margins.

Structural and sustainable efficiency gains in model training and inference