Recently, the 2026 Green Computing Power (Artificial Intelligence) Conference was held in Hohhot, Inner Mongolia.

At the conference, Hohhot and Ulanqab signed 13 projects with leading companies such as Volcano Engine and Cambricon, with a total investment of 136.1 billion yuan, about 6.8 times that of 2025, covering multiple fields such as chip testing, equipment manufacturing, computing power centers, and keyword factories.

This trillion-yuan investment is not simply about "building data centers and expanding capacity," but covers the entire upstream, midstream, and downstream of the computing power industry chain: upstream efforts to address hardware shortcomings such as chip testing and computing equipment manufacturing; Strengthening core supporting facilities such as the midstream computing power center and computing-power collaboration; Downstream layout includes word element trading and AI scenario applications. This marks the official departure of Inner Mongolia's computing power industry from the previous single "data center + power" basic service model to a complete, closed-loop full industrial chain.

Signal One: Computational-Electric Synergy—The "Power Revolution" in the AI Era

Electricity costs account for about 60% to 70% of data center operating costs; whoever controls cheap and stable green electricity controls the "oil field" of the AI era.

Inner Mongolia's confidence comes from three sources:

First, the region enjoys unique advantages in wind and solar green power resources, with a total developable capacity of 1.88 billion kilowatts of new energy technology across the region, and its installed capacity and power generation have long led the nation.

Second, the electricity price advantage is obvious. The green electricity usage rate in the Helingeer cluster has stabilized above 86%, with household electricity prices of about 0.32–0.4 yuan per kWh, consistently at the lowest level nationwide.

Third, power supply is solid. The "four horizontal and six vertical" grids in western Mongolia and the "seven horizontal and one vertical" power backbone grids in eastern Mongolia have been fully completed, and the power supply reliability rates in Hohhot, Ulanqab, and other areas have reached 99.999%.

Computing-electricity collaboration has been included in the 2026 government work report, moving from a concept to a practical application. At this conference, GCL Energy Technology's "Integrated Computing-Power Integration Project" is a typical example—using Inner Mongolia's abundant wind and solar energy to supply green electricity to high-energy-consuming computing power centers, and then using the computing power load to regulate grid fluctuations, forming a virtuous cycle of "strengthening computing with electricity, promoting power through computing."

Signal Two: Word Element Economy—A Paradigm Leap from "Selling Computing Power" to "Selling Smart."

If computational-electric collaboration solves the problem of "where computing power comes from," then the word economy solves the problem of "how to use computing power."

Tokens are the basic units of measurement used by large models when processing input and generating output. By pricing and settling based on word values, each model call can be turned into a tradable service. As of March 2026, the average daily number of word calls in China has surged from about 100 billion at the beginning of 2024 to about 140 trillion, an increase of more than a thousand times.

This conference launched the comprehensive service platform for digital service going global (keywords going global), released the nation's first green computing AI full-stack keyword trading platform, and Hohhot set a goal to achieve revenue of over 100 billion yuan in keyword digital industry revenue by 2028. This means the computing power industry is upgrading from a resource rental model of "selling data centers and bandwidth" to a value operation model of "selling keywords and selling intelligence"—the commercial value of computing power has been thoroughly expanded.

Signal 3: Domestic substitution—the entire chain from chips to equipment is independently controllable

In this signed project, Cambricon will build a chip laboratory in Hohhot, and Haiwu will settle in the computing power equipment industrial park. This means that domestic AI chips are completing localized support from R&D and testing to deployment, and the domestic computing power hardware and software ecosystem continues to improve.

Unlike the past when it simply introduced external companies to build data centers, this batch of projects is building a complete local computing power industry chain in Inner Mongolia—in the future, servers, temperature control equipment, and energy storage devices will all be produced locally, and the industrial cluster effect will become increasingly strong.

For the overall autonomy and controllability of the AI industry, this is a long-term positive development.

Kingtech's Perspective | Understanding the Three Investment Themes Behind 136.1 Billion

Of the 136.1 billion yuan investment, a large portion will flow toward the collaborative construction of computing and electricity. Companies with integrated generation-grid-load-storage capabilities will continue to benefit, with a focus on sub-sectors such as green power operations, energy storage systems, liquid cooling, and smart grid dispatch.

Segments such as word exchange platforms, data annotation, model training and inference, and AI application implementation are forming a brand-new commercial closed loop. Hohhot has set a goal for the ciyuan industry revenue to exceed 100 billion yuan by 2028, signaling that this sector will enter a boom period over the next two years.

The implementation of projects such as Cambricon Chip Lab and Haiwu computing power equipment marks the transition of domestic computing power from being "usable" to "user-friendly." Against the backdrop of sustained explosive demand for AI computing power, the pace of domestic substitution in upstream segments such as domestic chips, computing equipment, and high-speed optical modules is expected to accelerate further.

In the short term, the contract amount does not equal immediate profit. The total investment of 136.1 billion yuan is the project, and implementation takes 3–5 years. Investors should be wary of short-term sentiment speculation; In the long run, the logic of AI computing power competition has been completely rewritten—from "competing on chip quantity" to "competing on energy costs," and from "selling computing resources" to "selling keyword services."