In Gansu Provincial Hospital, the "Longshang Little Doctor" traditional Chinese medicine AI-assisted diagnosis and treatment system is being called upon frequently. This artificial intelligence model recently completed the province's first generative AI service registration, becoming Gansu's first licensed medical large model.
Meanwhile, on the Lanzhou municipal government portal, when residents ask about social security or administrative procedures, the "Xiaolan AI Answer" built by China Electronics Wanwei provides responses within seconds. Behind these visible and practical changes lies a "new track " that Lanzhou is accelerating its layout of — the token economy.
Lanzhou is speeding up the construction of a complete intelligent industry chain from computing power infrastructure to scenario implementation, striving to build an important "smart factory" in the northwest.
Six Major Initiatives Systematically Advance Token Industry
What is the token economy? "The token economy is about converting computing power, data, and models into callable artificial intelligence services, using the input and output tokens processed by models as usage units, priced, traded, and settled according to different models and service specifications," said Yao Zhijun, professor and doctoral supervisor at the School of Artificial Intelligence, Lanzhou University, in one sentence.
"Simply put, whoever can continuously 'produce intelligence and sell intelligence' will seize the initiative in the new round of industrial competition."
Since the beginning of this year, Lanzhou has focused on the two major directions of "basic support and integrated application" defined in the "Lanzhou Artificial Intelligence Empowering High-Quality Development Implementation Plan," vigorously implementing six major initiatives — solidifying computing power hardware support, activating new data elements, tackling key model cores, deepening scenario main positions, cultivating industrial growth poles, and safeguarding security lifelines — to accelerate the construction of a complete industry chain from computing power infrastructure, data supply, and model cultivation to scenario implementation and circulation trading.
Green Power and Computing Power Dual Empowerment Highlights Western Inference Advantages
For tokens to be produced at scale, the prerequisite is stable, low-cost, and schedulable computing power. For universities and small and medium-sized enterprises, the computing power problem is often not "whether it exists" but "whether it is affordable."
Training a large model is a one-time investment, but after the model goes live, every question-and-answer session and every analysis continuously consumes computing power. The more calls are made, the more energy consumption, equipment utilization, and network costs deserve careful calculation. For the same task, how much electricity is used, how many cards are run, and how long it takes directly determines whether a token can be sold.
"Based on our province's abundant wind and solar resources, as well as the computing power and network foundation brought by the 'East Data West Computing' initiative, we can combine green power direct connection or green power integration with computing power scheduling to reduce the cost of completing a single task while meeting response speed and service quality requirements," Yao Zhijun said.
In his view, localized applications are another experimental field. For tasks like scenic area inquiries and enterprise knowledge Q&A, many have clearly defined scopes and can be handled using lightweight models combined with industry knowledge bases. For scenarios with requirements on response speed or data retention, local deployment can be explored; for large-batch tasks that are less sensitive to latency, they can be centralized at computing power centers.
"Gansu has both the conditions to make inference more economical and the opportunity to refine models for better usability in local scenarios."
Leveraging the distinctive characteristics of the northwest region, Lanzhou University has produced phased results in western vertical large models and specialized corpus directions. The "Smart Gansu AI Cultural Tourism Large Model" is a typical example — this is a result jointly released by the Gansu Provincial Department of Culture and Tourism, Gansu Mobile, and Lanzhou University, targeting cultural tourism management departments, cultural tourism enterprises, and tourists. It organizes information on Gansu's cultural resources, scenic areas, transportation, meteorology, and more into a fine-tuning dataset containing hundreds of thousands of data entries, supporting analysis of tourist volume, source markets,停留 duration, and spatial distribution, helping management personnel understand the province's cultural tourism operating trends and key scenic area visitor flow changes.
University-Industry Joint Research Institute Accelerates Research Commercialization
How to turn laboratory models into industry-usable services? "From 'research tokens' to 'industry-usable tokens' means taking models that work in the laboratory and turning them into services that can connect to business systems, operate stably in daily use, and have clear usage accounting," Yao Zhijun explained.
For example, clinical medical intelligent agents integrate medical guidelines, examination and testing standards, and evidence-based medicine knowledge bases to assist with clinical consultation information organization, preliminary risk screening, differential diagnosis references, medication compatibility checks, medical record document generation, and follow-up management, providing evidence-based辅助 decision support for clinicians. Government intelligent agents, relying on government data and administrative rules, automatically complete policy interpretation, material pre-review, matter consultation, form filling, process reminders, and demand collection, simplifying administrative procedures.
The important platform for this transformation is the Western Computing Power Technology Research Institute jointly built by Lanzhou University and Gansu Mobile. The institute has a board of directors that deliberates on research project approval, bylaws, and project management. Its research directions mainly include computing power networks, artificial intelligence, and information security.
"This platform needs to study how computing power is scheduled and how models are deployed on one end, and on the other end, it needs to test whether the technology can be used in real business scenarios," Yao Zhijun said.
From laboratory to market, talent is a key link. Yao Zhijun acknowledged that the current technical shortcoming is turning demonstrable models into systems that can operate stably over the long term. Laboratory data is neat and tasks are clear, but in enterprises, one must face data changes, call volume fluctuations, and security maintenance.
Today, composite talents who understand technology, can comprehend industry needs, and can organize collaborative advancement across parties are precisely what the industry urgently needs. To this end, the research team is promoting earlier involvement of students in real projects, jointly guided by university faculty and enterprise engineers, to fill the critical gap from scientific research to industry and accumulate human resource support for the long-term development of the token economy.
Comments