China’s burgeoning artificial intelligence sector is currently navigating a pivotal shift, with the humble ‘token’ emerging as a critical indicator of competitiveness and a foundational unit for large language models. This transformation is not merely technical; it is fundamentally reshaping the economic landscape of the nation’s AI industry, driving both unprecedented revenue growth and significant cost pressures. The increasing sophistication of AI applications, particularly in conversational agents, has driven exponential demand for these computational units, creating a dynamic, often volatile market.
In the nascent stages of AI chatbot development, typical user interactions often involved only a few hundred Chinese characters, translating to a relatively small number of tokens. However, as reasoning models evolved, this figure quickly escalated to approximately 10,000 tokens. Today, the average user interaction demands more than 50,000 tokens, according to Li Boxun, CTO of Infinigence AI. This dramatic increase underscores the growing complexity and capability of AI models, which now handle more intricate queries and generate more comprehensive responses. The demand for processing power, measured in tokens, has become a direct reflection of an AI model’s utility and performance.
Tokens as Revenue Drivers
The financial implications of this token-driven economy are already evident in the impressive revenue figures reported by leading Chinese AI firms. Zhipu AI, for instance, witnessed its API revenue in the first quarter of this year soar sixty-fold compared to the previous year. Similarly, Alibaba Cloud reported a 15-fold increase in token revenue since the beginning of 2026. The sheer scale of operations is further highlighted by ByteDance’s Doubao model, which surpassed an astonishing 180 trillion token calls per day in June. These figures not only demonstrate the rapid adoption of AI services but also the immense computational resources required to sustain such growth. This financial momentum is crucial for understanding the broader trends in China’s tech landscape, particularly how Chinese tech giants secure over $17 billion in Hong Kong funding amidst the global tech race.
However, this rapid expansion has not been without its challenges. Since August 2025, a massive wave of data center construction, aimed at meeting the insatiable demand for AI infrastructure, has inadvertently tightened the supply of critical hardware components. Memory chips, GPUs, CPUs, optical modules, and liquid cooling systems have all seen their availability constrained, leading to significant price hikes. The cost of some memory chips, for example, has more than doubled, directly impacting the operational expenses of AI service providers. This supply chain pressure is a critical factor influencing the overall economics of the AI token market.
Adjusting Prices
In response to these escalating costs, major cloud providers have begun adjusting their pricing structures. Tencent Cloud, for instance, raised prices for its core models by more than 430% in March. Alibaba Cloud followed suit, increasing prices for some AI computing services by 5-34% in March and boosting prices for its high-performance storage systems by 30%. These adjustments reflect the underlying economic realities of providing advanced AI infrastructure, where the costs of hardware and energy are significant determinants of service pricing. The need for capital infusion to support such expansions is also evident, as seen in reports such as Minimax aiming for $1.9 billion capital infusion amid AI expansion.
Despite these price increases, demand continues to surge. Zhipu AI CEO Zhang Peng noted that while API prices rose by 83% quarter-on-quarter in Q1, call volume still jumped more than four times. This indicates strong market appetite for AI services, even amid higher costs. ByteDance also adjusted its pricing strategy, releasing Doubao 2.1 on June 23 with double the price of its previous version and introducing a Pro subscription at 68 yuan ($10.03) per month. These moves suggest a market confident in the value proposition of its AI offerings.
Conversely, some players are adopting a different strategy. DeepSeek, for example, cut its prices again in May, with DeepSeek V4 Pro now priced at one-quarter of its original level. The company has also announced plans to launch the official version of DeepSeek V4 in mid-July, featuring peak-and-valley pricing, with daytime peak prices twice the normal levels. This approach aims to optimize resource utilization and potentially attract a broader user base with more flexible pricing options. This dynamic pricing environment highlights the competitive nature of China’s AI market, where companies are experimenting with various models to capture market share and maximize profitability.
Industry Outlook
The long-term outlook for AI service costs remains upward, according to industry leaders. Alibaba Group CEO Eddie Wu stated that server costs have more than doubled since January 2025. He anticipates that over the next one to two years, prices for models and cloud services will continue to rise, which is expected to significantly boost Alibaba Cloud’s gross margin. This forecast suggests that the current trend of increasing AI infrastructure costs is not a temporary blip but a sustained development that will continue to influence market pricing and profitability.
According to IDC data from 2025, the market share in China’s Model-as-a-Service sector is competitive. ByteDance’s Volcano Engine held the leading position with over 40% market share by token call volume, followed by Alibaba Cloud with approximately 30%. SiliconFlow ranked fourth in this rapidly evolving market. These market share figures underscore the dominance of major tech conglomerates in AI services, leveraging their extensive infrastructure and technological capabilities.
However, the profits generated within this token economy are not evenly distributed. A significant portion remains concentrated upstream, particularly with hardware manufacturers. In 2025, Nvidia and the world’s three major memory chip makers collectively generated more than $160 billion in operating profit. This highlights the critical role of hardware in enabling the AI revolution and the substantial financial returns companies at the forefront of chip manufacturing and design reap. This concentration of profit upstream raises questions about the long-term sustainability and equitable distribution of value within the broader AI ecosystem.
Enterprise users, who are the primary consumers of these AI services, are particularly sensitive to price fluctuations. Jia Anya, head of SenseTime Group’s Raccoon agent application family, emphasized that “Enterprise users are very sensitive to 20 to 30 percent price increases. They need AI infrastructure with controllable budgets, stability and high cost-performance.” This sentiment is echoed by a China executive at an international cloud company, who told The Business Times, “If tokens are destined to become the next generation of electricity, then data centers are power plants, and large-model and cloud companies are the power grid.
Companies don’t want tokens that are simply easy to use or cheap. They want cost-performance.” This perspective underscores the growing maturity of the AI market, where enterprises are seeking not just cutting-edge technology, but also reliable, cost-effective, and scalable solutions that can be integrated seamlessly into their existing operations. The focus is shifting from raw computational power to the overall value proposition and economic efficiency of AI services.