Chinese AI Token Prices Rise Amid Computing Demand, UBS Analyst Warns

The rising demand for computing power driven by artificial intelligence agents is transforming the competitive landscape within the industry. However, the token pricing offered by Chinese providers remains significantly lower than that of their U.S. counterparts, with the performance gap being relatively small compared to the price difference, according to an analyst at UBS Securities.

This year has seen a surge in the practical application of AI agents, requiring substantially more computational power to complete tasks compared to traditional conversational AI. Xiong Wei, a China internet analyst at UBS Securities, noted at the Asia Investment Conference media briefing yesterday that “the token consumption for performing an agentic task can exceed that of AI chat use by more than 1,000 times.”

According to third-party data from an API aggregation platform, global token consumption increased approximately tenfold in the week ending May 15 compared to the same period last year.

Xiong emphasized that token usage is experiencing exponential growth. He also indicated that as agentic applications expand from tech enthusiasts to more knowledge-based workers, demand is expected to demonstrate increasing resilience.

Despite this, growth in computing power—whether in chip manufacturing or data center construction—tends to be more linear, while demand is escalating exponentially, Xiong explained. “Companies are confronting token demands that surpass their supply, with even their own processing capacities falling short of the potential customer needs they observe,” he said. He added that this supply-demand imbalance “may persist longer than anticipated.”

Smaller Gap in Performance

While competing in the computing power market, Chinese AI model developers are able to provide more cost-effective token prices compared to U.S. companies, despite the narrower performance gap. This advantage is attributed to their ongoing focus on optimizing inference efficiency, which enables them to “drive their cost advantage as low as possible,” Xiong stated.

China’s computing costs are approximately half those in the U.S., supported by a steady electricity supply, according to Kenneth Fong, head of China internet research at UBS Investment Bank.

In terms of price wars, Xiong argued that in a market constrained by supply, lowering prices to capture more token share “doesn’t make sense because even if you win that demand, you might not have enough computing capacity to meet it.”

Leading providers are actually increasing token prices while discontinuing lower-priced subscription options, launching premium plans, and shifting toward usage-based billing models, he explained.

Approximately 78 percent of IT decision-makers in Chinese companies have moved beyond pilot projects into active implementation, based on a UBS EvidenceLab survey of over 100 respondents. Last year, spending on generative AI accounted for about 12 percent of IT budgets, with AI-related expenditures growing faster than overall IT spending—outpacing the growth observed in similar U.S. surveys. However, concerns over unclear return on investment continue to be a significant barrier.

Xiong commented that these findings reflect similar patterns observed among U.S. companies. “Both Chinese and American firms are showing the same trends,” he said.

Strategic Shift for Internet Companies

Fong remarked that the internet sector’s underperformance since the start of the year—due to macroeconomic headwinds, increased AI investment, and uncertainty about disruptions—is fully reflected in stock prices. “Expectations are low, valuations are inexpensive, and fundamentals are gradually stabilizing, suggesting that stock prices may have already bottomed out.”

Among different subsectors, Fong expressed optimism about the resilience of AI model developers and highlighted the low valuations in the gaming industry, where leading firms have a price-to-earnings ratio of just around 12 to 13 times.

He further explained that the fundamental business logic in the industry has shifted. “Previously, revenue was driven by time spent—more traffic meant more advertising or transaction revenue.”

“Now, internet companies monetize by capturing a share of the value they create for users,” Fong added. “This transformation will likely push their revenue potential to new heights.”