Guangdong has the most extensive nuclear power foundation, but Zhejiang was the first to play the AI electricity card.

Yujian Energy has learned that recently, four departments including the National Development and Reform Commission, the National Energy Administration, the Ministry of Industry and Information Technology, and the National Data Administration jointly issued the "Action Plan for Promoting the Two - way Empowerment of Artificial Intelligence and Energy", for the first time clearly stating "exploring the direct connection of energy sources such as nuclear power and hydrogen energy to power computing facilities".

This document has opened a gap at the institutional level - nuclear power is no longer just a base - load power source in the power grid. It can lay a dedicated line directly to the distribution room of the data center. Globally, Microsoft, Meta, and Amazon have proven the feasibility of this direction with long - term agreements worth billions of dollars. In China, Guangdong, which has the largest increment in nuclear power installed capacity in the country, is standing at the intersection of the two growth curves of nuclear power and computing power.

01

Global tech giants

Are they all vying for the "household registration" of nuclear power plants?

The persistence of technology companies in nuclear power has gone beyond a gesture and turned into real money on the balance sheet.

In 2024, Microsoft signed a 20 - year agreement with Constellation Energy, investing $1.6 billion to restart Unit 1 of the Three Mile Island Nuclear Power Plant in Pennsylvania. The 835 - megawatt power capacity will be exclusively supplied to Microsoft's data centers. The construction progress has been accelerated, and it is expected to be put into operation in 2027.

Another tech giant, Meta, announced three nuclear power agreements in January 2026, involving Vistra, Oklo, and TerraPower. By 2035, the total locked - in nuclear power capacity will reach 6.6 gigawatts, making it the largest nuclear energy buyer among global tech companies. Just the Vistra deal alone covers three nuclear power plants with a capacity of over 2.6 gigawatts, and all the agreements have a term of 20 years. What does 20 years mean? It means that Meta believes that the demand for electricity from AI is not cyclical but structural and irreversible. Locking in the electricity cost at a fixed price of around $80 per megawatt - hour for 20 years may seem like an expensive investment today, but it could be a far - sighted move given the long - term upward trend of electricity prices.

In June 2025, Amazon signed a long - term power purchase agreement with Talen Energy until 2042, directly purchasing 1,920 megawatts of nuclear power from the Susquehanna Nuclear Power Plant in Pennsylvania to power its AWS data centers. Google has a slightly different approach. It is collaborating with Kairos Power to deploy a 500 - megawatt advanced nuclear reactor, betting on the next - generation small - scale reactor technology, and plans to supply power to its data centers starting in 2030.

These agreements have three common points: they are all long - term agreements, they all lock in a fixed electricity price, and they all point to direct supply to data centers. The collective actions of global tech giants send a signal - nuclear power is changing from "one of the options" for AI to "infrastructure".

03

The computing power hunger in the Greater Bay Area

Guangdong has the cure

Guangdong plays a significant role in China's nuclear power map.

In the first quarter of 2026, Guangdong's nuclear power generation reached 26.01 billion kilowatt - hours, ranking first in the country. The Daya Bay Nuclear Power Plant has been operating safely for 32 years, and has cumulatively transmitted over 320 billion kilowatt - hours of clean electricity to Hong Kong, accounting for a quarter of Hong Kong's total social electricity consumption. The total investment in the six "Hualong One" units of the Taipingling Nuclear Power Plant exceeds 120 billion yuan. Unit 1 was put into operation in April 2026, and Unit 4 just started construction in May. After all units are completed, the annual power generation will be nearly 55 billion kilowatt - hours.

What does 55 billion kilowatt - hours mean? The electricity consumption of data centers in Guangdong Province in 2025 was about 11 billion kilowatt - hours. The annual power generation of just the Taipingling project can cover more than four times the current electricity demand of the province's data centers. Nuclear power is already the second - largest power source in Guangdong after thermal power. More importantly, it has a complete industrial chain - from design, equipment manufacturing, engineering construction to operation, Guangdong has a whole nuclear power industrial chain. China General Nuclear Power Group has 29 operating nuclear power units with an installed capacity of 33.04 million kilowatts and 19 units under construction with an installed capacity of 23.02 million kilowatts, of which 17 units use the third - generation "Hualong One" nuclear power technology.

Meanwhile, in April 2026, the monthly electricity consumption of 307 data centers in Guangdong exceeded 1 billion kilowatt - hours, a year - on - year increase of 17.42%. The growth rate of the Shaoguan cluster is even more exaggerated. The annual electricity consumption in 2025 exceeded 170 million kilowatt - hours, a year - on - year increase of 751%. In January 2026, the electricity consumption was 26.34 million kilowatt - hours, a ten - fold increase year - on - year. The Greater Bay Area is one of the regions with the highest computing power density in China. It has requirements for both the total amount and stability of electricity - nuclear power happens to be one of the few energy forms that can meet both of these conditions.

03

Five cents cheaper per kilowatt - hour

The computing power cost war has reached the energy end

When it comes to the direct connection between nuclear power and computing power, it ultimately comes down to an economic account.

In the operating costs of data centers, electricity costs account for about 28%, and after excluding fixed - asset depreciation, it further rises to 38%. For a super - large - scale intelligent computing center with an annual electricity consumption of 1 billion kilowatt - hours, every 0.05 yuan reduction in the electricity price per kilowatt - hour can save 50 million yuan per year. Existing calculations show that the coordinated operation of computing and electricity in some provinces and cities can reduce users' electricity bills by more than 0.05 yuan per kilowatt - hour. The first large - scale coordinated green power direct - supply project for computing and electricity in China, which was put into operation in May 2026, is expected to save 100 million yuan in electricity costs annually compared with the eastern coastal regions.

The reason why direct - connected green power is cheaper is that it can avoid paying grid - access fees. If the direct connection of nuclear power follows this path, its economic advantages will be more prominent. The utilization hours of nuclear power are over 7,000 per year, far higher than those of wind and solar power. The unit power supply cost has stable predictability after being spread over the entire life cycle.

Since 2026, the electricity price mechanism of nuclear power in Guangdong has been changing rapidly. Guangdong's nuclear power units no longer implement the variable - cost compensation mechanism. In the worst - case scenario where the market - based electricity price hits the bottom at 0.372 yuan per kilowatt - hour, there is still room for an increase in the actual on - grid electricity price of nuclear power. The pricing flexibility of nuclear power in the Guangdong power market is increasing, opening up more possibilities for the direct - connection business model.

Yujian Energy has made a rough calculation: If the direct connection between Guangdong's nuclear power and data centers can achieve a cost advantage of 0.05 yuan per kilowatt - hour, based on the annual electricity consumption of 12 billion kilowatt - hours of the province's data centers, it can save about 600 million yuan in electricity bills for computing power enterprises every year. This figure will increase as the electricity consumption of computing power continues to grow. Electricity cost is evolving from an operating expense to the core competitiveness of the AI industry.

04

Fujian and Zhejiang are leading the race

When will Guangdong reveal its trump card?

Guangdong is not the only one playing the card of nuclear power plus computing power.

In terms of nuclear power generation, in the first quarter of 2026, Guangdong (26.01 billion kilowatt - hours), Fujian (22.21 billion kilowatt - hours), and Zhejiang (19.09 billion kilowatt - hours) accounted for more than 60% of the country's total nuclear power generation. In terms of nuclear power penetration rate, Fujian has a rate as high as 29.29%, the highest in the country, and the proportion of clean - energy installed capacity reaches 66.8%. With the layout of multiple bases such as Ningde, Fuqing, and Zhangzhou, Fujian has formed the densest nuclear power cluster in the country.

However, Zhejiang is half a step ahead in model exploration. In January 2026, China National Nuclear Corporation and a subsidiary of Alibaba Group jointly established China Nuclear (Xiangshan) Nuclear Energy Co., Ltd. with a registered capital of 250 million yuan, exploring the model of direct nuclear power supply to data centers relying on the Jinqimen Nuclear Power Project in Xiangshan. This is the first direct cooperation at the capital level between a tech company and a nuclear power central enterprise in China. Although the current scale is not large, its demonstration significance is clear.

Yujian Energy believes that Guangdong's differential advantage lies in its industrial depth. The complete nuclear power industrial chain, the dense computing power demand in the Greater Bay Area, and the country's largest increment in nuclear power installed capacity together constitute comprehensive conditions that are difficult for other provinces to replicate in the short term. The nuclear power equipment manufacturing, engineering construction, and operation and maintenance sectors are entering a new round of investment cycles. The total investment in the six units of the Taipingling Nuclear Power Plant exceeds 120 billion yuan, with an average investment of over 20 billion yuan per unit. China General Nuclear Power Group is promoting the demonstration project of the "Hualong One" 2.0 version and prospectively planning the R & D of the 3.0 version.

What is even more imaginative is the "hard connection" at the physical level. If intelligent computing center clusters can be laid out around nuclear power bases first and power can be supplied nearby through direct - connection dedicated lines, the resulting comprehensive advantages will be quite significant: on one side is a stable zero - carbon base - load power source, on the other side is the continuously expanding AI computing power demand, and the two are connected by a dedicated line. Once this path is successful, Guangdong's position in the AI industry map will not only be determined by its computing power scale but also be enhanced by an energy - cost moat.

This article is from the WeChat official account "Yujian Energy", author: Zhao Jianan, published by 36Kr with authorization.