In the first nine months of 2026, the stock prices of Chinese power companies barely moved. Some fell. The utilities that supply electricity to the data centers powering China's AI boom traded at valuations that would puzzle a Western portfolio manager.
The data centers themselves were thriving. Inner Mongolia's computing capacity reached 345,000 PFlops by July 2026, up more than 80 percent year over year. Alibaba, ByteDance, and DeepSeek were building facilities in Ulanqab, a city on the Mongolian plateau whose average annual temperature is 4.3°C. The electricity that powered those facilities cost roughly 0.33 yuan per kilowatt-hour — about $0.046. In the eastern provinces where most of the users lived, industrial electricity cost 0.76 to 0.80 yuan per kilowatt-hour. In Ohio, data center operators were paying the equivalent of roughly 0.60 yuan.
The power was cheap. The AI was growing. The utility stocks were not.
That divergence reflects how China prices electricity — and it explains why the country's AI industry has access to some of the cheapest power in the world, even as its grid operators carry nearly 4 trillion yuan in debt.
The Cross-Subsidy Mechanism
The mechanism that keeps Chinese electricity cheap for data centers is called cross-subsidization. Residential and agricultural users pay regulated tariffs set by provincial price authorities. Industrial and commercial users pay market-based rates that include transmission and distribution charges, system operation fees, government funds, and a policy-mandated cross-subsidy component.
The cross-subsidy flows from commercial and industrial customers to residential and agricultural users. It is designed to keep household electricity affordable — a policy objective described in Chinese regulatory documents as "protecting people's livelihood."
The effect on data centers is indirect but significant. Data centers are classified as industrial users. They do not receive the residential subsidy. But they benefit from a system in which the grid operator — State Grid or China Southern Power Grid — is not required to earn a commercial return on its transmission assets. The grid is a state-owned utility with a policy mandate.
In practice, this means the grid can route power from where it is cheapest to where it is needed, without the cost of that routing being determined by a market-clearing price. When a data center in Ulanqab draws power from a wind farm 41 kilometers away via a dedicated 220 kV line, the transmission cost is set by a regulated tariff designed to make the project viable.
The result is a price for AI compute that reflects policy choices rather than the marginal cost of generation and transmission.
The Green Power Direct Connection
The policy has a name: computing-power coordination — a framework that aligns data center construction with grid planning. In March 2026, it was written into the government work report for the first time. The report called for "implementing ultra-large-scale intelligent computing clusters and computing-power coordination as new infrastructure projects."
The operational mechanism is called green power direct connection. Under guidelines issued by the National Development and Reform Commission and the National Energy Administration in May 2026, computing facilities are prioritized for direct connection to renewable generation. The power does not pass through the public grid. It flows from a wind farm or solar plant directly to the data center via a dedicated line.
The flagship project is in Ulanqab. A 360 MW phase-one facility — 300 MW wind, 60 MW solar, and 65 MW of storage — delivers more than 700 million kilowatt-hours of green power annually to data centers in the Helingeer New Area. The second and third phases, totaling 500 MW, were approved in 2026 and are scheduled to be operational by year-end.
The pricing is negotiated between the generator and the data center operator, with the grid company providing the transmission line at a regulated rate. The resulting cost — 0.33 yuan per kilowatt-hour in Inner Mongolia, 0.36 yuan in Ningxia's Zhongwei — is roughly 45 percent of the eastern industrial rate.
For a 1-gigawatt data center consuming 8 terawatt-hours annually, that price difference translates to 3.4 billion yuan ($470 million) per year, based on the rate differential and typical utilization. Over a 15-year facility life, the savings exceed $7 billion.
Why the Stocks Don't Move
The cheap power is a subsidy to the AI industry. It is not a subsidy to the utility.
Chinese power generators — the companies that own the coal plants, hydroelectric dams, and wind farms — are not capturing the value of the AI boom. Their revenues are determined by regulated tariffs and market prices that have been falling, not rising.
In 2024, China established a capacity pricing mechanism for coal-fired power. But it recovers only 30 to 50 percent of fixed costs, compared with the PJM capacity market in the United States, where capacity prices spiked in 2024 and again in 2025. Huaneng International, China's largest listed coal power generator, earned a profit of just 1.9 fen per kilowatt-hour on its domestic coal power in 2024 — up from 0.1 fen in 2023 but far below the levels of the 2022 supply crunch.
The stock market has noticed. The price-to-book ratio for Chinese power utilities fell from above 2x in late 2021 to below 2x by 2025. Over the same period, the price-to-book ratio for U.S. power companies rose from roughly 2x to more than 7x. The U.S. utilities were re-rated as AI infrastructure assets. The Chinese utilities were not.
A July 2026 investor question to China Resources New Energy captured the frustration. The question noted that "European and American institutions view power as the hard currency of the AI era, a scarce infrastructure asset deserving a growth premium," while "the A-share market has long treated power as a low-growth, policy-constrained, cyclical utility." The company's response was measured: "Market pricing is affected by multiple external factors. Asset value discovery takes time."
The grid companies that deliver the power are also not benefiting. State Grid and China Southern Power Grid together carry nearly 4 trillion yuan in debt. Their capital expenditure plans — 5 trillion yuan for grid upgrades during the 15th Five-Year Plan — are funded by state investment and debt, not by the returns generated from data center customers.
The AI industry gets cheap power. The utility that delivers it gets a regulated return. The investor who owns the utility gets a stock that does not move.
The American Comparison
The United States has taken the opposite approach, and the results are visible on household electricity bills.
In May 2026, the average U.S. residential electricity price reached 18.44 cents per kilowatt-hour, up 6.2 percent year over year and 40.8 percent since May 2020, according to EIA data. In PJM, the largest U.S. grid operator, the capacity auction for 2028-29 cleared at $325 per megawatt-day — the maximum allowed — with data centers contributing roughly $6.3 billion of the $16.4 billion total cost, socialized across every ratepayer on the grid.
The two systems distribute the burden differently. In the U.S., the cost of serving a data center is recovered through rates that all customers share. In China, the cost of serving a data center is partially absorbed by the policy objective of keeping the grid cheap — and the grid operator is not required to earn a market return.
One system socializes the cost of AI infrastructure across households. The other subsidizes it through state-owned enterprises and regulated tariffs. Neither is free.
What the Model Reveals
The cheap electricity that powers China's AI industry reflects policy choices rather than market outcomes.
The grid is not priced to clear a market. It is priced to serve a set of objectives: keep household electricity affordable, keep industrial costs competitive, route computation to where the resources are, and absorb the cost of those objectives across a balance sheet that includes 4 trillion yuan in debt.
For the AI industry, that model delivers a structural cost advantage that is difficult to replicate. It is not a technology gap that can be closed with a better chip. It is a pricing gap that reflects a different set of institutional choices about who pays for infrastructure and who captures the value.
The stocks don't move because the model was not designed to make them move. The power is cheap because the cost is distributed elsewhere.
Sources: Sina Finance (June 21, 2026); China Youth Net (September 24, 2026); STHeadline (August 18, 2026); OFweek (March 9, 2026); Fortune (May 20, 2026); Fox Business (February 26, 2026); EIA (May 2026); National Development and Reform Commission (May 2026); State Council (March 2026); 10jqka Investor Relations Platform (July 15, 2026); China Fortune Network (September 23, 2026); Sohu (June 30, 2026); UBS China Power Equipment Report (September 24, 2026); Huatai Securities (2026).
Disclaimer
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations
This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.
Forecasts from third-party analysts can change with market conditions.
Cost and pricing examples are point-in-time estimates; actual rates vary.
Country and company comparisons rely on public reporting, not operational data.
This sector moves fast; timelines and deal terms may be updated later.
Company deals and regulatory rulings may evolve; verify current status.
AI infrastructure is changing quickly; claims can become outdated soon.
Sources
- Sina Finance (June 21, 2026)
- China Youth Net (September 24, 2026)
- STHeadline (August 18, 2026)
- OFweek (March 9, 2026)
- Fortune (May 20, 2026)
- Fox Business (February 26, 2026)
- EIA (May 2026)
- National Development and Reform Commission (May 2026)
- State Council (March 2026)
- 10jqka Investor Relations Platform (July 15, 2026)
- China Fortune Network (September 23, 2026)
- Sohu (June 30, 2026)
- UBS China Power Equipment Report (September 24, 2026)
- Huatai Securities (2026).
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations: This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.; Forecasts from third-party analysts can change with market conditions.; Cost and pricing examples are point-in-time estimates; actual rates vary.; Country and company comparisons rely on public reporting, not operational data.; This sector moves fast; timelines and deal terms may be updated later.; Company deals and regulatory rulings may evolve; verify current status.; AI infrastructure is changing quickly; claims can become outdated soon.