AI's 448 TWh hangover: Underwater data centers test a 23% cooler future

As the AI industry races toward a $5 trillion market by 2033, its environmental appetite looms larger than ever—data centers already consume more electricity than Saudi Arabia. Now, a pioneering underwater data center off Shanghai promises to cut power consumption by 23% and eliminate water use entirely, offering a glimpse of how subsea cooling could unchain AI's growth from its carbon and water chains. For machine learning engineers and infrastructure architects, this deployment translates theoretical efficiency gains into a live, 24-megawatt proof-of-concept that could reshape where and how training clusters are sited.

China's operational deployment of an underwater data center off the coast of Shanghai marks a significant real-world test of subsea infrastructure as a means to mitigate the escalating environmental toll of artificial intelligence. The $226 million project, built by HiCloud Technology in partnership with the Chinese government, submerges a 24-megawatt data center six miles offshore, powered by coastal wind turbines. The facility's design reportedly slashes electricity consumption by close to 23% compared to land-based equivalents, while eliminating freshwater usage entirely and reducing land footprint by over 90%. This tangible step follows years of exploratory research, most notably Microsoft's Project Natick, which from 2018 to 2020 operated a submerged shipping-container-sized data center 117 feet deep off Scotland, proving the technical viability of underwater server operations. Microsoft has since relegated Natick to a research platform, emphasizing ongoing investigation into reliability and sustainability rather than immediate commercial rollout.

“

The United Nations University Institute for Water, Environment and Health (UNU-INWEH) reported in June 2026 that the global AI market is projected to balloon from $189 billion in 2023 to nearly $5 trillion by 2033, a nearly 27-fold increase.

The backdrop to this development is an AI industry whose computational appetite is growing at a breathtaking pace. The United Nations University Institute for Water, Environment and Health (UNU-INWEH) reported in June 2026 that the global AI market is projected to balloon from $189 billion in 2023 to nearly $5 trillion by 2033, a nearly 27-fold increase. This expansion drives an unprecedented surge in data center construction, with UNU-INWEH estimating that global data centers already consumed 448 terawatt-hours of electricity in the most recent annual figures—enough to rank as the world's 11th largest electricity consumer if treated as a sovereign nation, outpacing Saudi Arabia. The report also warned that the water required to cool AI data centers will soon rival the consumption of entire nations, intensifying pressure on increasingly scarce freshwater resources in many regions.