1. Ten kilometers off Shanghai’s coast, a 32-meter steel platform rises from the East China Sea, beneath which a submerged cylindrical cabin houses a high-density computing hub [para. 1][para. 2]. Built by Shenzhen HiCloud Data Center Technology Co., this facility is claimed to be the world’s first underwater data center directly powered by offshore wind, naturally cooled by the ocean with an ultra-efficient PUE rating of 1.15, and launched commercially in February [para. 3].
2. The facility serves as a testing ground for China’s “computing-power synergy” strategy, which mandates that ultra-large AI clusters be developed in lockstep with power infrastructure [para. 4][para. 5]. In March, China’s cabinet enshrined this concept in its annual government work report, calling for a new “intelligent economy” [para. 5].
3. This strategy is most evident in Shanghai’s Lingang Special Area, where power consumption surged nearly 30% in the first five months of 2026, driven overwhelmingly by data centers [para. 6]. Lingang’s intelligent computing capacity is expected to double by year-end, and local grids have doubled high-voltage substations, with the government disbursing millions in subsidies [para. 7][para. 8].
4. Operating data centers underwater slashes energy use by natural cooling and saves land. The Lingang design overcomes a major hurdle from a 2020 pilot by using a surface platform for power equipment and a submerged 1,950-ton cabin for servers, drastically cutting maintenance costs and drawing power from three sources: an offshore wind farm, the onshore grid, and backup diesel generators [para. 10][para. 11][para. 12].
5. Despite being wired to the offshore wind farm, HiCloud purchases green electricity through the state grid at standard commercial rates (~0.75 yuan/kWh), more than double the direct cost of the wind farm’s supply [para. 13]. Industry insiders say implementing true point-to-point direct-purchase policies could drastically cut costs for the project’s second phase, set to break ground later this year [para. 14].
6. Under the national “East Data, West Computing” strategy, China Telecom dispatched computing tasks from Lingang to a green center over 4,000 km away in Karamay, Xinjiang, in a May test [para. 16]. This shifted demand westward, dropping the Shanghai facility’s local power load by 75%, and seamlessly transferred tasks across heterogeneous hardware [para. 17].
7. China Telecom’s Lingang chief Yuan Xiaoyang noted similar tests to Hubei and Fujian in 2025, but these cross-country dispatches remain largely experimental [para. 18]. Making them routine requires a market-driven pricing mechanism; currently, the structural shortage of high-end computing chips and varying energy demands of AI models pose challenges [para. 19].
8. Back on land, data centers deploy large-scale battery storage and AI models to cut costs. SenseTime Group has integrated an 18-megawatt energy storage station (backed by CATL) into its Lingang hub [para. 21][para. 22]. Shanghai’s peak-to-valley electricity price gap can be 4.6 times; SenseTime uses a proprietary AI energy model to predict loads and release stored electricity, avoiding peak-capacity surcharges [para. 23].
9. SenseTime’s Lingang head Zhang Xu says this hardware-software synergy shaved annualized electricity costs by 7%, below the regional industry average [para. 24]. As Shanghai transitions to a continuous spot electricity market, Zhang anticipates this AI-driven approach will become essential, with the ultimate goal of precisely calculating energy consumed per AI-generated token [para. 25][para. 26].
AI generated, for reference only
Sign in with Google
Sign in with Facebook
Sign in with 财新