From "Eastern and Western Wings" to "Largest in the Central Region": An industry experiment related to the inclusive distribution of computing power is unfolding simultaneously in China's eastern and western regions.

Written by: Web4 Research Center

The Ministry of Industry and Information Technology has recently issued a notice to organize a special campaign to empower small and medium-sized enterprises (SMEs) with inclusive computing power. For the first time, it explicitly proposed exploring innovative businesses such as “Computing Power Banks” and “Computing Power Supermarkets.” It will support SMEs in depositing idle computing power resources and achieve flexible access and retrieval through cross-region and cross-cycle scheduling, thereby reducing computing costs. The notice states that by the end of 2028, a generally established inclusive computing power services system will be basically in place, with broad coverage, low costs, and high-quality services, covering no fewer than 10 industry categories among the main sector classifications of SMEs.

The policy signals are clear and strong. And on the front line of policy implementation, the practices of early movers have already begun—Shanghai Telecom’s “Computing Power Supermarket” has gone live, and the Henan Airport Intelligent Computing Center has completed Token API interface deployment, supporting SMEs to call domestic intelligent computing power as needed.

At 3:00 a.m., on a row of server racks at the Qingpu Intelligent Computing Center in Shanghai, thousands of server indicator lights flicker at high frequency like a sea of stars. But at this moment, what is truly being used for these powerful computing capabilities is not the training of a trillion-parameter AI model from some AI unicorn, but an industrial simulation task from a small and medium-sized manufacturing enterprise in Suzhou. It ordered 4 hours of intelligent single-card computing power online through Shanghai Telecom’s “Computing Power Supermarket,” and the entire process took no more than two minutes.

At the same time, in Zhengzhou Airport Economic Zone in Henan, the server room of the Henan Airport Intelligent Computing Center is also abuzz. With computing capacity of 10,000 P, this largest intelligent computing center in central China could complete its entire computing workload in 1 second if all 7 billion people on Earth were to calculate non-stop for 40 years. Just a few weeks ago, this massive system completed the deployment of the Token API interface, enabling SMEs to call domestic intelligent computing power on a per-Token basis without having to make a one-time large capital investment.

This is China in the spring of 2026. Computing power—the core productive factor of the AI era—is gradually transforming from the “heavy weaponry” of a handful of giants into “everyday goods” that SMEs can access.

I. A “Silent Revolution”: When Computing Power Becomes Water, Electricity, and Coal

Computing power is the core productive factor for the digital transformation of SMEs in the AI era, akin to water, electricity, and coal in the industrial era—directly determining a company’s innovation capacity and market competitiveness.

This line is not a slogan from a technology company, but a precise summary by Zhang Linshan, a researcher at the Macroeconomic Research Institute of the National Development and Reform Commission, on the positioning of the computing power industry. Indeed, large-model applications, industrial simulations, intelligent R&D, and data analysis—these intelligent scenarios that once belonged only to large enterprises—have now become key tools for SMEs to reduce costs and increase efficiency and to seize niche market opportunities.

But reality does not allow optimism. The problem of mismatched supply and demand for computing power still exists. Although the total amount of computing power in China is large, there is a misalignment in supply and demand between eastern and western regions, and overall utilization rates remain low. More importantly, SMEs’ demand for computing power exhibits “small batches, fragmentation, and temporariness,” while traditional computing power services follow a capital-intensive model of “long-term binding and large up-front payments.” The two logics are completely misaligned.

How serious is this contradiction? A set of data is enough to explain the situation. According to statistics from the National Data Bureau, as of March this year, China’s daily Token call volume has already exceeded 140 trillion. Compared with 100 billion at the beginning of 2024, this is an increase of over 1,000 times. Compared with 100 trillion at the end of 2025, the increase in just three months exceeds 40%. The explosive growth in Token call volume means AI applications are penetrating every sector at an unprecedented speed. At the same time, however, the annual growth rate of China’s total computing capacity is about 30%, far behind the surge in demand on the supply side.

This is not simply a matter of “not enough computing power.” The deeper root of the problem is that a large amount of computing resources are idle or used inefficiently, while SMEs that urgently need computing power are deterred by barriers that are too high. The computing power divide is becoming a key bottleneck restricting the innovation vitality of SMEs.

II. “Computing Power Banks”: Making Computing Power Flow Like Money

On April 2, 2026, the Ministry of Industry and Information Technology issued a notice to organize a special campaign to empower SMEs with inclusive computing power. For the first time, it explicitly proposed exploring two major innovative businesses: “Computing Power Banks” and “Computing Power Supermarkets.” The moment the policy was released, it immediately drew strong attention from the industry.

Why are these two concepts so eye-catching? Because, at root, they reconstruct the logic of computing power supply.

The core logic of a “Computing Power Bank” is not complicated. It draws on the traditional banking “deposit and loan” model: SMEs can “deposit” idle computing power resources into a unified resource pool. Through intelligent scheduling across regions and across cycles, the platform enables flexible storage and retrieval, peak-shifting allocation, and value realization of computing power. In other words, in the past computing power was a set of silent servers—once bought, it could only be used for the buyer’s own needs, and if it was not used up, it could only remain idle. Now, computing power becomes an asset that can be “deposited” and “withdrawn.” When it is idle, it can be stored in the resource pool to obtain returns; when it is busy, it can be extracted at any time to supplement demand.

The essence of a computing power bank is to transform fragmented idle computing power into a circulatable inclusive resource.

Some may ask: SMEs themselves lack computing power—where does the idle computing power to “deposit” come from? This is precisely a cognitive misconception. A small and medium-sized enterprise engaged in industrial design may need large amounts of computing power for rendering and simulation during peak project seasons, but during off seasons those servers are almost in sleep mode; an AI startup may have tight demand during the day but a great deal of idle computing power at night. Under the traditional self-built and self-used model, such cyclical idleness cannot be monetized and is simply wasted. Computing power banks precisely provide a path to activate these “sleeping computing powers.”

The policy design by the Ministry of Industry and Information Technology goes beyond that. The notice clearly requires improving China’s computing power platforms’ SME-focused areas, the resource docking zones within the China SME Service Network, and the dedicated zones for national computing internet service nodes—so as to promote precise matching between SMEs’ demand and the supply of computing power resources. At the same time, it will promote flexible payment models such as paying by “card hours,” “compute cycles,” and Token-based billing. It also encourages local departments to subsidize SMEs that align with industrial orientations through SME platform special zones using “computing power vouchers,” “storage power vouchers,” and “transportation power vouchers.”

The intention of this combination is very clear: computing power banks solve the problems of “deposit” and “scheduling.” Computing power vouchers solve the problem of “being able to afford it.” Platform special zones solve the problem of “being able to find it.” With coordination among all three, they form a complete closed-loop system for the supply of inclusive computing power.

III. “Computing Power Supermarkets”: Buying Computing Power Like Online Shopping

If computing power banks solve the problem of resource circulation, then the computing power supermarket targets the problem of consumer experience.

“Computing Power Supermarkets” are positioned as public portals that aggregate various computing power services and support online negotiation and transactions for computing power services. Put more plainly, it is like an e-commerce platform in the computing power sector—intelligent single-card and multi-card clusters, bare-metal servers, GPU cloud hosts. Various computing power products are clearly priced. After logging into the platform, enterprises can place orders on demand and use them immediately, just like selecting everyday goods.

This logic may sound simple, but in the computing power field it is revolutionary. In the traditional model, for SMEs to obtain computing power, they either build data centers themselves (with investments starting at the scale of hundreds of thousands and beyond), or rent long term (with contract lock-in and large prepayments), or simply give up on using it. After computing power products are turned into standardized, retail-style offerings, enterprises can pay only for the portion they actually use. With billing by card hours, compute cycles, or even by Token, it truly realizes “use as much as you need, pay as much as you use.”

The practice of Shanghai Telecom’s computing power supermarket has already proved the feasibility of this model. The platform connects the robust computing power resources of the Qingpu and Lingang “two-wing” intelligent computing centers, and opens them up to computing power suppliers, SMEs, and public users. It supports all kinds of computing power service providers to settle in and operate. At present, it has built district-level computing power operation special zones for relevant areas, offering online ordering services such as intelligent single-card/multi-card systems, bare metal, and GPU cloud hosts, and it has core functions including multi-level account management and precise metering and billing. It provides one-stop service support for customers across multiple industries such as finance, government, and education, supporting “computing + models + applications.”

The Henan Airport Intelligent Computing Center is also verifying the feasibility of inclusive computing power in another way. As the first intelligent computing center in central China to fully deploy the DeepSeek model, it relies on an approach of “one-point access, immediate retrieval and use” to help SMEs use domestic leading AI large models more conveniently. A domestic AI chip company, Taichu Yuanqi, builds the intelligent computing foundation for it and completes the deployment of the Token API interface. This means SMEs can call domestic intelligent computing power just like calling an API, without needing to invest a large amount of capital at one time. To reduce trial-and-error costs, the intelligent computing center also provides Token trial services to SMEs and university research institutions.

From Shanghai to Henan, both paths point to the same direction: making computing power change from a “luxury for a few” into “everyday goods for most.”

IV. Target 2028: A Roadmap for Inclusive Computing Power

If the practices in Shanghai and Henan are “happening now,” then the timetable set by the Ministry of Industry and Information Technology for inclusive computing power is the “future roadmap” for the entire industry.

The notice explicitly proposes that by the end of 2028, a broadly covered, low-cost, high-quality, eco-strong, and talent-strong inclusive computing power services system will be basically established. It will cover no fewer than 10 categories among 15 industry categories that apply to the division standards of SMEs. It will further strengthen public-service support for SMEs’ computing power applications, and significantly reduce the thresholds for SMEs to obtain and use computing power.

This goal can be interpreted from three dimensions.

First, breadth of coverage. Covering at least 10 out of 15 industry categories means inclusive computing power will penetrate core national economic sectors such as manufacturing, education, agriculture, and finance. This is not an “optional item,” but a hard requirement from the policy level. The notice also specifies that it will focus on leveraging pilot cities for SMEs’ digital transformation, SMEs’ featured industrial clusters, and innovative industrial clusters. It will concentrate on key links for manufacturing SMEs—such as R&D and design, production and manufacturing, equipment operations and maintenance, and supply chain management—and accelerate the promotion of low-cost, lightweight, and easy-to-deploy inclusive computing power solutions.

Second, depth of services. Broad coverage is just the first step; the truly critical aspects are high-quality services, a vibrant ecology, and strong talent. To this end, the notice has deployed five major key tasks—actions to enhance computing resource allocation, actions to promote inclusive supply of computing power services, actions to empower enterprises in key industries, actions to co-build industrial ecosystem collaboration, and actions to build talent cultivation capacity. From resource deployment to talent cultivation, this is a systematic project covering the entire chain of computing power services.

Third, mechanism innovation. Promoting a “1+N” pairing and assistance mechanism between large, medium, and small enterprises is a highlight of this special campaign. The notice encourages large enterprises to establish normalized and institutionalized pairing and assistance relationships with SMEs, and to promote sharing the experience of large enterprises in computing power technologies, operations management, cost control, and scenario implementation. It will organize integration and matchmaking activities among large, medium, and small enterprises, and encourage the “chain leader” enterprises to open up their idle internal computing power resources to SMEs when releasing demand scenarios. This “large helps small” model provides institutional support for a virtuous cycle in the computing power ecosystem.

V. From “Building” to “Using”:

A Profound Shift in Industrial Logic

Looking across the overall design of the entire special campaign, it is clear to see one core signal: the focus of China’s computing power industry is shifting from “building” to “using.”

In the past few years, China has invested heavily in building computing power infrastructure. Intelligent computing centers have sprung up everywhere, and the total scale of computing power has continued to rise. But once the infrastructure is built, who will use it, how will it be used, and can they afford it? These questions have not received equal attention. As a result, there have been many instances of idle computing power resources and low utilization rates, creating a stark contrast with the strong computing power demand from SMEs.

The significance of the special campaign lies in the fact that it no longer pursues “a larger scale,” but pursues “higher efficiency.” It no longer focuses on “who builds,” but on “who uses.” It no longer emphasizes “technological advancement,” but emphasizes “service inclusiveness.” This is an industrial logic shift from supply-driven to demand-driven.

Of course, the development of any new model is not accomplished overnight. For computing power banks and computing power supermarkets to truly take root across the country, they still face a series of real challenges. Latency issues in cross-region scheduling, interface standardization problems among computing power products from different vendors, data security and privacy protection issues in computing power transactions, and the market mechanism design problems for pricing idle computing power—all of these need to be explored and resolved step by step in practice. As Heidegger revealed when questioning the essence of technology, the true value of technology is not in its instrumentality, but in how it changes the relationship between human beings and the world. The ultimate significance of making computing power inclusive may not be simply enabling more enterprises to use AI, but ensuring that AI truly becomes a tool in everyone’s hands—not a privilege for a few.

When computing power flows like water and electricity.

Innovation will no longer have barriers.