AI’s Energy Footprint Turns Green Power into a Digital-Infrastructure Constraint

On June 4, Sina Finance reported that the UN University Institute for Water, Environment and Health had released a report warning that global data-center electricity and water consumption could double by 2030 because of surging AI demand. The article stated that 2025 global data-center electricity consumption reached 448 TWh, exceeding Saudi Arabia’s national electricity use, with AI compute accounting for one-fifth of the total. By 2030, data-center electricity use is expected to reach 945 TWh, roughly equivalent to Japan’s national electricity consumption, and AI’s share could rise to 40%.

The water numbers are equally important. The same report said 2025 data-center water consumption reached 4.5 trillion liters, enough to meet the annual water needs of more than 600 million people in sub-Saharan Africa, and warned that water demand could also double. It also said data-center land area could expand from 6,900 square kilometers to more than 14,500 square kilometers. For ESG analysis, AI is therefore not only a technology theme. It is an electricity, water and land-use theme.

This matters for China because the country is pushing both AI development and green-power expansion. Earlier policy language around computing and energy coordination has treated green-power use as part of the data-center agenda. The UN warning makes the constraint sharper. If compute demand grows faster than clean electricity, data centers can increase fossil load. If facilities cluster in water-stressed regions, cooling becomes a social and environmental risk. If land and grid connections are poorly planned, AI infrastructure can crowd out other industrial loads.

Investors should be careful with the simple conclusion that AI is automatically bullish for green power. More load can support renewable demand, but only if procurement, grid capacity, storage and demand response are real. A data center that signs a green-power contract but strains local evening peak demand may still create system stress. The valuable companies will be those that can pair compute growth with credible power sourcing, flexible operation, storage integration and water-efficient cooling.

The article also cited estimates of more than 15,000 data-center projects and more than 600 GW of project reserves globally, with major distribution across the United States, Europe and China. That pipeline suggests that the ESG battle will move from abstract AI ethics to physical-resource governance. Power purchase agreements, certificate cancellation, hourly matching, water disclosure and site selection will become board-level issues for digital infrastructure.

For China ESG, the positive side is that AI demand can accelerate investment in clean power, grids and storage. The negative side is that it can expose gaps in accounting and local resource planning. A data center is not low-carbon because the model it runs is advanced. It is low-carbon only if the power, water and land behind it are managed transparently.

The takeaway is that AI’s environmental footprint will test the credibility of green-power systems. China has the renewable scale to serve part of the load, but scale is not the same as clean, time-matched delivery. As AI becomes a strategic industry, its ESG profile will depend on whether digital growth is coordinated with energy reality. Compute is virtual. Its resource footprint is not.

For data-center operators, disclosure should become more granular. Annual renewable-power percentages are useful, but they do not show whether clean electricity is available when servers are consuming power. Water-use effectiveness, cooling technology, heat reuse, grid-interconnection status and location-level stress should all matter. In regions with coal-heavy marginal power or limited water availability, a data center can look efficient on paper while imposing real local costs. AI companies will face growing pressure to explain that physical footprint.

This also creates an opening for China’s power and storage firms. Data centers need reliability, low-carbon supply and increasingly flexible load management. Green-power direct supply, storage-backed procurement, virtual power plants and demand-response services can become part of the AI infrastructure stack. But the commercial opportunity should not obscure the governance challenge. If every AI campus claims priority access to clean electricity, other industries and households still need power. Allocation, pricing and transparency will decide whether AI becomes a catalyst for green grids or a new source of transition tension.