Xingyun Technology shifts from selling compute to selling tokens, and its valuation logic is being repriced.
The artificial intelligence industry is approaching a new inflection point.
On July 31, Jiang Yi, director of the Policy Research Office of the National Development and Reform Commission, stated at a press conference that during the "15th Five-Year Plan" period, China's computing power network construction is expected to add 4 trillion yuan in direct investment. As the national integrated computing power network construction has been included as one of the 109 major projects in the "15th Five-Year Plan" outline, computing power infrastructure is transitioning from being the foundational resource supporting the digital economy to becoming strategic infrastructure for the AI era.
The importance of computing power is being redefined. In recent years, market attention on the AI industry has focused primarily on large model capabilities, chip breakthroughs, and algorithmic innovation. However, as large models enter the commercialization stage, computing power is no longer just a resource behind model training, but has become key infrastructure connecting models, applications, and commercial value.
At present, China has established an "8+10+3" national computing power spatial layout, comprising 8 national computing power hubs, 10 national computing power clusters, and 3 computing-electricity collaborative development zones. The national integrated computing power network monitoring and scheduling platform has also connected to approximately 70% of the country's intelligent computing resources. As the national-level computing power network gradually improves, the computing power industry is entering a new phase of large-scale construction.
Against this backdrop, the evaluation logic that capital markets apply to computing power companies is also undergoing change.
Previously, computing power leasing companies were largely viewed as capital-intensive cyclical industries, with the core profit model being the procurement of GPU servers followed by revenue generation through leasing computing power. This model relies essentially on equipment scale, utilization rates, and depreciation cycles, which imposes certain limitations on growth potential.
But with the acceleration of AI commercialization, a new direction is emerging in the industry: shifting from "selling computing power" to "selling Tokens."
And this is precisely the new model that Xingyun Technology is exploring.
On the evening of July 29, Xingyun Technology (300209.SZ) announced that its wholly-owned subsidiary, Shenzhen Xingyun, had signed a supplementary computing power services agreement with a leading domestic long-context large model service provider, increasing the original contract amount from 1.014 billion yuan to 3.053 billion yuan, representing an increase of 201.14%.
The reason this contract has attracted market attention is not merely the increase in value, but more importantly, the changes in its business model.
According to the announcement, fixed service fee arrangements related to Token revenue appeared for the first time in the cooperation model between the two parties. This also signifies that computing power services are evolving from the traditional resource leasing model toward the TaaS (Token as a Service) model, which is more closely aligned with the commercial value of AI applications.
This is also why the market has dubbed Xingyun Technology "A-share's first Token stock."
Tokens, in essence, are the unit of measurement in the operation of large models. As enterprises deploy large models across scenarios such as customer service, office work, R&D, and marketing, model invocation volumes continue to rise, and Tokens are becoming an important metric for measuring AI commercial value.The relationship between traditional computing power companies and their clients has historically been more of a one-off resource procurement arrangement, where clients purchase GPU usage time and computing power companies earn fixed leasing revenue. However, under the TaaS model, computing power service providers begin to participate in the model commercialization process, generating more sustainable revenue through the value created by Tokens.
This means that the revenue logic of computing power companies is shifting from "equipment utilization" to "AI application growth."
Looking at the global market, similar business models have already been validated by cloud computing giants.
Take Amazon AWS, the global cloud computing leader, as an example. Its business has long since moved beyond simply providing server leasing services and instead builds a complete business ecosystem through cloud infrastructure, large model platforms, and AI application ecosystems.
On one hand, AWS provides underlying computing resources; on the other hand, it connects large model providers and enterprise customers through AI service platforms such as Bedrock, allowing developers to directly call upon model capabilities and pay based on actual usage.
The core value of this model lies in the fact that cloud providers are no longer just infrastructure providers, but have become value distributors within the AI industry chain.
The direction Xingyun Technology is exploring bears a certain resemblance to AWS's development logic.
Based on publicly available information, the company has built a comprehensive system covering hardware, software, and services around AI computing power. On the hardware side, the company's business covers complete computing power equipment, prefabricated liquid cooling system integration, computing power leasing, and full-lifecycle operations and maintenance services. On the software side, the company has brought in Dr. Tang Bo, Chief Scientist and researcher at Southern University of Science and Technology and founder of AlayaDB.AI, along with his team, to develop the AlayaJet inference engine, which improves the Token output capacity per unit of computing power by optimizing large model inference efficiency.In simple terms, in the past, computing power companies competed over "how many GPUs they own," but in the future, the core of AI infrastructure competition will become "how many Tokens the same GPU can produce." This is also the key distinction that sets the TaaS model apart from traditional computing power leasing.
In terms of industrial layout, Xingyun Technology is attempting to complete a business model upgrade.
In the past, the company was largely regarded by the market as a computing power infrastructure enterprise. Now, with the gradual implementation of the TaaS model, the company is transitioning toward becoming an AI infrastructure platform.
On the evening of July 30, the company released its semi-annual report for 2026, which further validated this transformation process. According to the data, during the reporting period, the company achieved operating revenue of 254 million yuan, a year-over-year increase of 497.11%; net profit attributable to shareholders of the listed company reached 12.0154 million yuan, a year-over-year increase of 540.15%, turning from losses to profits.
Excluding the impact of share-based incentive expenses, the company's net profit attributable to shareholders reached 20.4585 million yuan, a year-over-year increase of 989.99%.
At the same time, the company's capital support capabilities have continued to strengthen. As of July 2026, Xingyun Technology and its subsidiaries had applied for a total of 18.024 billion yuan in credit facilities from financial institutions, of which 10.264 billion yuan had been approved. Currently, the company has disclosed an order backlog exceeding 15.4 billion yuan.
Improved performance, order growth, and the TaaS business model are collectively driving the market to reassess the company's value proposition.
For the computing power industry, the 4 trillion yuan investment is only the starting point for industrial development.
As large models move from technological exploration into the stage of commercial deployment, the companies that will truly be competitive in the future are not necessarily those that simply own the most computing resources, but rather platform-based enterprises that can connect computing power, large models, and application demand while sharing in the growth dividends of AI commercialization.
From this perspective, what Xingyun Technology is experiencing is not just business expansion, but a shift in its valuation framework. If traditional computing power companies rely more on hardware assets for their valuation, then companies with TaaS capabilities have the opportunity to be repriced according to the logic of AI infrastructure platforms.
The 4 trillion yuan computing power network construction opens up industrial space, while the Token business model unlocks new growth potential for computing power companies. For Xingyun Technology, the key going forward is not just how much computing power it owns, but how much new value in the AI era that computing power can create.