Yunnan's transformation: selling tobacco, electricity and Token
AI is starting to be clearly priced just like traffic.
At a telecom business hall in Kunming, 10 million Tokens can be purchased for 9.9 yuan per month.
At the same time, Yunnan's green aluminum industry is expanding rapidly. In the first half of 2026, the added value of industries above designated size across the province increased by 0.7%, while the added value of the green aluminum industry increased by 10.2%, contributing 70.8% to the growth rate of industries above designated size in the whole province.
One sells computing power and the other sells aluminum. The two seemingly vastly different businesses share a similar underlying logic, which is to convert resources into tradable commodities.
If we look back at the industrial changes in Yunnan over the past decades, they have basically followed this path. From tobacco to hydropower, from green aluminum to computing power, Yunnan has been constantly looking for new resource export directions, and continuously transforming resource advantages into industrial scale.
I. From Tobacco to Computing Power
For a long period of time, Yunnan's industrial landscape was dominated by the tobacco industry.
The tobacco industry has brought stable tax revenue, employment and industrial chains, and also shaped the fundamentals of Yunnan's industrial economy.
At its peak in the 1990s, the tax and profit contributed by the tobacco industry once accounted for more than 75% of the province's fiscal revenue. With the development of the energy industry, new pillars began to emerge for industrial growth.
In 2019, the added value of the power-based energy industry surpassed that of the tobacco industry for the first time, becoming the largest pillar industry in the province.
Hydropower is the key to this change.
Yunnan is rich in hydropower resources, with a high proportion of clean power, and has obvious advantages in industrial power consumption costs. In 2021, the total installed power capacity of the province exceeded 100 million kilowatts, and green energy began to become an important condition to attract the manufacturing industry. Yunnan proposed that the total installed power capacity of the province will reach about 180 million kilowatts by 2027.
Hydropower itself will not automatically translate into higher industrial income. Low power cost first changes the cost structure of the manufacturing industry, and then creates opportunities to attract industrial chain transfer.
Around 2018, with the adjustment of the electrolytic aluminum industry layout, the "North Aluminum Moving South" process began to accelerate. Yunnan's advantages are clear: rich hydropower and competitive electricity prices, making it suitable for undertaking the high-energy-consuming electrolytic aluminum industry.
In 2017, the operating capacity of electrolytic aluminum in Yunnan was about 1.68 million tons; by 2025, this figure has increased to about 6.5 million tons. In 2025, the output value of Yunnan's green aluminum industrial chain exceeded 200 billion yuan for the first time.
The industry is also extending to downstream sectors. Products such as automotive lightweight materials and new energy battery aluminum foils make Yunnan's green aluminum no longer just a primary raw material. Some enterprises have begun to enter the supply system of automotive industrial chains such as BMW and Tesla.
Hydropower did not directly turn into income, but first turned into low-cost power, then into aluminum ingots, processed products and industrial chain output value.
The computing power industry develops along a similar logic.
Kunming Wanxichong Intelligent Computing Industrial Park has built multiple data centers, and computing power infrastructure continues to expand. The China-Laos and China-Myanmar international communication optical cables and the Kunming International Communication Service Exit and Entry Bureau provide network conditions for cross-border data transmission. Power, computer rooms and communication networks together form the foundation of the computing power industry.
When these conditions are combined, electricity can drive servers, servers can provide computing services, and computing services can be priced by Tokens.
From selling tobacco to selling electricity, and then to selling Tokens, the form of commodities has changed, but the underlying idea is continuous. Yunnan keeps looking for resources with comparative advantages, and then transforms the advantages into market-purchasable products through industrial investment and infrastructure construction.
This method has created new industrial increments for Yunnan one after another.
The problem is that resource advantages can attract enterprises and industrial scale can be rapidly expanded through investment, but whether a company can form the ability of long-term operation and continuous innovation in the local area requires another set of methods.
II. Large Projects Can Be Relocated, But How to Retain Core Capabilities?
The rise of Yunnan's green aluminum industry is inseparable from large enterprises and major projects.
Electrolytic aluminum requires huge investment, as well as stable power supply, mature production management and complete supply chain. For Yunnan, introducing powerful enterprises is easier to form scale effect than cultivating the entire industrial chain from scratch.
Aluminum industry clusters in Wenshan and other regions are expanding continuously, and enterprises such as Chinalco and Yunnan Aluminum are continuing to make layouts, driving the development of smelting, processing and supporting industries.
The computing power industry also has similar characteristics.
Telecom operators build data centers, cloud service enterprises provide computing power, and communication infrastructure supports data flow. Huawei and Yuntianhua cooperate on gasification business to optimize production through automation and intelligent means; Yunnan Communications Investment & Construction Group explores the application of large models in transportation business to improve the processing capability of specific tasks. Yunnan Baiyao and aluminum industry enterprises are also promoting the application of cloud platforms and industry models.
The common point of these projects is the combination of external technical capabilities and local industrial scenarios to solve specific problems. Xiong Yunbiao, Dean of the School of Economics and Management of Yunnan Minzu University, defines this model as exchanging resources for industries, exchanging resources for markets, and exchanging resources for technologies.
This is a path with practical foundation. Resources can be used as negotiation chips, markets can attract enterprises, and industrial scenarios can also provide landing space for new technologies. For regions with relatively weak industrial foundation, introducing mature enterprises and technologies can shorten the cultivation cycle.
However, project landing is only a node in industrial development.
A production system has been built, who will maintain it? After the external technical team leaves, can local employees continue to optimize it? A technology has been verified in one scenario, can it be migrated to other businesses? These problems cannot be directly answered by project investment amount and output value.
Technology can be procured, equipment can be installed, and systems can be delivered. However, the business experience, process understanding and continuous improvement capability within the enterprise will not automatically land along with the project.
AI is bringing this problem to more enterprises.
On the production line, intelligence usually has clear objects, such as temperature control, equipment monitoring, and energy consumption optimization, and the results can be measured by output, yield and failure rate. When it comes to procurement, finance, contracts and daily management links, many jobs do not have such clear boundaries. Data is scattered in different systems, processes rely on manual connection, and judgment standards are hidden in employees' experience.
This type of work also consumes a lot of time of enterprises, but rarely appears on the investment list of major projects.
Chen Peng, Director of the Procurement Center of Yunnan Hengtong Coal Industry Co., Ltd. under Yunnan Energy Investment Group, discovered another possibility of AI in such work.
III. The Man Holding the Calculator
In 2017, Chen Peng saw an employee holding a calculator to check inventory data.
The data had been entered into the computer, but the employee still needed to calculate item by item and compare repeatedly to confirm whether the results were accurate. Chen Peng realized that the problem was not the lack of spreadsheets, but that employees had not found a more effective working method.
Many years later, similar situations still exist.
Hengtong Coal Industry uses an ERP system to manage its business. Due to security requirements, system interfaces cannot be opened arbitrarily, so employees need to export data and then check and sort it manually. A daily task may take an employee four hours. When encountering inspections or centralized processing tasks, work often drags on late into the night.
Data has been digitized, but work has not completely got rid of manual labor.
In July 2026, Chen Peng discussed AI applications with the Kingsoft Office team. In August, WPS Comate began to be implemented at Hengtong Coal Industry. Also in this month, Kingsoft Office launched the "WPS Comate Hundred Cities AI Agent Empowerment Program". At that time, Wu Qingyun, Vice President of Kingsoft Office, put forward a view: the industry has been talking about computing power, algorithms and data in the past few years, and this year we can finally bring applications to the forefront.
Behind this is to solve the deep-seated problem of AI technology from demonstration to large-scale implementation.
As an early pilot enterprise, Hengtong Coal Industry organized business personnel to work with Kingsoft Office pre-sales consultants to sort out processes, and tried to assign the repetitive work originally done by humans to digital employees.
This work did not start with replacing the ERP system, and of course the original infrastructure of the enterprise was not rebuilt. The team chose to find tasks that can be disassembled and verified from the existing processes.
This is also the idea that Kingsoft Office designed this product. As the core AI component of WPS 365, WPS Comate does not replace the enterprise's original ERP or OA system. It adds a layer of connection capability on top of the existing business systems, opens up data interfaces, unifies data calibers, provides an agent platform, and allows business personnel to define tasks in natural language. The multidimensional spreadsheets of Hengtong Coal Industry first solved the problem of data standardization, and Comate then built automatically executable skills on this basis.
In the actual application process, digital employees need to extract data from different sources, verify it according to rules, and then write the results back to the specified location. At two o'clock every morning, it starts processing data, and at nine o'clock in the morning, the relevant reports can be delivered to employees.
The work that originally took more than two hours now only takes a few minutes.
Changes have taken place within the existing business processes. Employees no longer need to spend time on repeatedly moving and checking data, and can devote their energy to links that require judgment.
Chen Peng also compiled user manuals and operation manuals for digital employees. For the first time, he found that writing job responsibilities for a non-human "colleague" is similar to arranging work for employees: tasks must be made clear, execution rules must be clear, and situations that require a pause must be agreed in advance.
He even arranged weekends off for the digital employee.
However, the digital employee did not take over the entire procurement process.
The procurement business of Hengtong Coal Industry involves tasks such as demand data import, calibration data import, performance data import, and external market information collection. The first batch of skills mainly focus on data processing and links with clear rules. Procurement intention, target judgment, procurement disputes and abnormal situations still need to be handled by humans.
This boundary is very important. Repetitive work can be handed over to AI, but matters involving interest trade-offs, responsibility taking and complex judgment still need to be checked by business personnel.
A more interesting test soon emerged.
During one operation, the digital employee encountered the 1 million Token context limit for a single dialogue, the task was interrupted, and the business could not proceed normally for nearly ten days. The problem cannot be solved just by adding more computing power. The original task concentrated in one dialogue was too complex, and the processing process also exceeded the scope suitable for a single work unit.
Chen Peng and the team re-disassembled the process, divided the original work into four independent skills, and added a coordinating dialogue, with a total of five dialogues working together to complete the task.
The task ran through again.
The inspiration brought by this failure has gone beyond the usage of the tool itself.
Business personnel need to understand which links in the process are relatively independent, which tasks can be split, how data is transmitted between different links, and who will take over when an exception occurs. In the past, these things might be handled by product managers, process designers or technical teams; in the process of AI implementation, business personnel also began to participate.
Chen Peng's work has therefore added a new dimension. He needs to understand the procurement business, convert the business into tasks that AI can execute, and make continuous adjustments according to the actual operation situation. Whether digital employees can work stably increasingly depends on whether the enterprise can clearly describe its own work.
This also explains why the same AI tool may produce completely different results in different enterprises. The model provides general capabilities, while enterprises need to provide specific data, rules, processes and judgment standards.
The practice of Hengtong Coal Industry is small in scale, but it has touched a key threshold for enterprise AI applications: moving from letting employees use AI to letting AI participate in the work process.
IV. Yunnan's Next Step
The case of Hengtong Coal Industry is not unique.
Yunnan Normal University, facing the evaluation demand of thousands of instructional design assignments for public courses every semester, also tried to use WPS Comate to undertake part of the work. The school completes the preliminary evaluation through the instructional design evaluation agent, and then teachers conduct sampling review. It runs twice regularly every day, so that students submit assignments on the same day and get feedback on the same day.
The sampling inspection results of teachers will also be continuously fed back to the evaluation model to help the system improve its judgment.
Coal enterprises and universities have very different businesses, but their working methods have something in common: first find a task with relatively clear boundaries and high repetition frequency, then clarify the evaluation rules and manual review mechanism, so that AI can be gradually adjusted in actual operation.
This type of application can hardly be completed just by purchasing software. Just as Wu Qingyun, Vice President of Kingsoft Office, said, the large model itself has general capabilities, but it does not understand the data, experience and rules of each enterprise. To make AI truly participate in the business, enterprises need to convert these knowledge originally scattered in employees' daily work into content that the system can use.
This means that AI applications are no longer just technical projects of the IT department.
Financial personnel need to know which calibers the budget analysis depends on, procurement personnel need to know which data can be automatically verified, and teachers need to know how the evaluation standards are implemented. The technical team is responsible for building tools, while the business team needs to participate in defining tasks, testing results, handling exceptions, and correcting processes in repeated use.
This is a more trivial kind of work, but it determines whether AI can move from pilot to daily operation.
For Yunnan, this problem also has a larger industrial background.
Yunnan proposed that by 2027, the proportion of the added value of non-tobacco and non-energy industries in the added value of industries above designated size will increase by about 3 percentage points. Industries such as green aluminum and new energy batteries are undertaking the task of adjusting the industrial structure.
At the same time, Yunnan's development and reform department also pointed out in the reply to relevant proposals that the energy value of green power has not been fully converted into higher industrial added value.
The advantage of low power cost can attract high-energy-consuming industries, and data centers can expand the supply of computing power, but the economic value of resources is ultimately realized through the production, management and innovation activities of enterprises.
Under the same power conditions, can enterprises make products with higher added value? With computing power infrastructure, can local enterprises use AI to improve R&D, procurement, finance and business decisions? These capabilities will affect how much industrial value the resource advantages can eventually be converted into.
Relevant assessment data from the China Academy of Information and Communications Technology also shows the implementation gap of AI applications: about 40% of application software has integrated professional agents, but the actual deployment rate of agents within organizations is about 17%.
The statistical objects and calibers of the two data are different, and they cannot be subtracted directly, but they illustrate a reality: there is still a gap between the software having relevant capabilities and enterprises applying the capabilities to daily work.