(Yicai) Jan. 15 -- The artificial intelligence industry in China will not suffer from a bubble in the same way as the United States, but is poised to face systematic opportunities along model global expansion, application explosion, and computing power substitution, according to an analyst at UBS Securities.

After DeepSeek took the lead last year, China's AI cost-performance advantage has become the core for global capital to reprice Chinese tech assets, Xiong Wei, a China internet industry analyst at UBS Securities, said at the 26th UBS China Conference held yesterday. This competitive edge will fully extend to application and computing power from the model layer this year, he added.

The cost-effectiveness of Chinese AI models will become a competitive advantage for global expansion, Xiong noted. When calculating the intelligence of large language models per US dollar, the average level of Chinese models offers greater competitiveness than US peers, he pointed out.

Regarding concerns about the AI bubble often raised by overseas investors, Xiong said that leading Chinese AI model developers rarely exhibit the so-called "circular financing" seen in the US. Instead, they rely more on the healthy cash flow generated by companies' mature businesses to support AI research and development, with restrained spending, resulting in low leverage ratios, he added.

US models pursue absolute intelligence leadership, but the most advanced models are also very expensive, Xiong pointed out. However, affordability remains a core consideration for companies in practical implementation, he said.

The cost-performance advantage will drive the "going global" of Chinese AI models this year, entering cost-sensitive emerging markets in the form of application programming interface services, with bubble-free, high iteration, and fast monetization being key, according to Xiong.

Leading Chinese internet companies are estimated to have capital expenditure of about CNY400 billion (USD57.4 billion) last year, about one-tenth of that of US peers, according to UBS. However, they have developed nearly half of the global top 15 models, highlighting more pragmatic spending strategies that prioritize return on investment and computational efficiency.

Chinese computing power steadily advanced in single-card performance and supernode technology last year, without the earlier spending-driven approach of blindly expanding computing power clusters, Xiong said. Domestic computing power will likely gain more share in the inference sector and even parts of the training market this year, he added.

According to UBS statistics, the average utilization rate of self-built data centers by major Chinese internet companies remains above 65 percent, significantly higher than that of their US peers, with strict regulatory control on excessive construction.

Editor: Martin Kadiev