1. The article begins with a case study of Datang International Power Generation, whose stock surged over 130% in a month after launching a solar plant in a cloud computing hub, triggering multiple risk advisories. The market overreacted to the concept of "computing and power synergy," even though Datang only invested in the power side and had no operational synergy project. [para. 1][para. 2][para. 3][para. 4][para. 5][para. 6]
2. In March 2026, the Chinese government officially endorsed computing-power coordination in its work report, and on May 8, four top bodies issued a joint action plan targeting a secure green energy system for AI by 2027 and full two-way empowerment by 2030. [para. 7][para. 8]
3. A key technical challenge is not total electricity volume but the grid's ability to deliver stable instantaneous power to specific locations at specific times, as explained by Wang Zesen of State Grid Jibei. Energy volume and instantaneous power are distinct constraints. [para. 9][para. 10][para. 11]
4. AI token usage in China surpassed 140 trillion daily by March 2026, a 1,000-fold increase from early 2024. Computing electricity consumption rose from 82.4 billion kWh in 2019 to 196 billion kWh in 2025 and could reach 500–700 billion kWh by 2030, while renewables now account for 48% of installed capacity, creating coordination difficulties. [para. 12][para. 13][para. 14][para. 15]
5. China's "Eastern Data, Western Computing" strategy, launched in 2022, directs data centers to resource-rich western hubs that must source over 80% of power from green energy. Wang Yongzhen notes that abundant cheap green electricity can compensate for chip manufacturing gaps. [para. 16][para. 17]
6. Gao Xing of China Securities dismisses fears of nationwide shortages, noting AI load is only ~1% of total consumption, unlike the U.S. where flat demand led to underinvestment. The real threat is a fundamental shift in grid operation. [para. 18][para. 19][para. 20]
7. The "power dictates computing" model is exemplified in Zhongwei, where a Datang solar farm and planned wind farm cut electricity costs to 0.36 yuan/kWh, 20% below local average. Ulanqab in Inner Mongolia uses a "source-grid-load-storage" model with 67% green power and low latency to Beijing, while Tencent experiments with a zero-carbon park in Chifeng. [para. 21][para. 22][para. 23][para. 24][para. 25][para. 26][para. 27]
8. Local grid strains are evident: Datong's computing sector consumed over 6 billion kWh in 2025 (40% jump, surpassing coal), accounting for 26.2% of grid load; similar surges occur in Zhangjiakou and Gui'an. [para. 28][para. 29][para. 30]
9. Inference workloads create a "peak upon a peak" effect, and AI servers cause instantaneous fluctuations up to 10,000 kW. Regions with high renewable inverter-based generation and large data center loads face a "dual-high" stability risk, as illustrated by U.S. incidents in Virginia where 1,500 MW of load vanished in seconds. [para. 31][para. 32][para. 33][para. 34]
10. A severe timeline mismatch exists: data centers can be built in 8–24 months, while substations take 3–5 years. Opaque utilization rates (10–80%) make grid planning difficult, risking waste or stranded assets. [para. 35]
11. China is testing virtual power plants (VPPs): Shanghai migrated inference tasks to Fujian in three minutes to shed 50 kW, and Guangdong integrated three telecom data centers into its spot market for flexible load management. However, true synergy remains in its infancy due to data transmission latency, physical hard-drive transport, and high bandwidth costs. [para. 36][para. 37][para. 38][para. 39][para. 40][para. 41]
12. The path forward requires unifying fragmented power, computing, telecom, and carbon markets, with tech giants like Alibaba and Tencent investing in next-generation nuclear technology. Wang Yongzhen expects the experimental phase to last another three years, with the ultimate goal of integrating security, green energy, and economic efficiency. [para. 41][para. 42]
AI generated, for reference only
Sign in with Google
Sign in with Facebook
Sign in with 财新