Will AI drink China’s farms dry?

From the drought-stricken valleys of the American West to the arid plains of northern China and parts of Latin America, a fundamental clash of resource priorities is emerging: the insatiable thirst of the artificial intelligence (AI) economy versus the basic water needs of rural agriculture and global food security.

As generative AI models expand and hyperscale computing clusters proliferate, physical data centres hum with unprecedented energy demands. To keep millions of densely packed microprocessors from overheating, these facilities rely on cooling systems that consume freshwater on a staggering scale, evaporating millions of gallons directly into the atmosphere every day.

This clash is not confined to any single nation or political model. In desert counties across Arizona and the dry agricultural basins of Oregon, local farming communities have staged vocal protests against tech giants, arguing that municipal authorities prioritise server cooling over farm irrigation during multi-year droughts. In southern Europe and Latin America, most notably in central Chile and the dry highlands of Mexico, local regulators face growing pushback from agricultural unions as power grids and watersheds are reallocated to accommodate massive data campuses. The digital economy promises borderless and intangible value, but its physical engine extracts finite and highly localised natural resources.

Zooming in: China’s structural water crisis

While this tension spans the globe, the problem reaches a unique operational intensity in China. China faces a harsh environmental reality: its per capita freshwater resources stand at barely 2,100 cubic metres — roughly one-quarter of the global average. Worse still, China’s water distribution suffers from a profound geographic mismatch. The lush southern provinces hold roughly 80% of the country’s water supply, while the vast northern and western regions are chronically dry.

Against this ecological backdrop, China’s national policy landscape is driving an unprecedented surge in digital infrastructure. Under the 15th Five-Year Plan (2026–2030), the central government has prioritised the rapid scaling of AI construction and the realisation of a national integrated computing network. Central to this strategy is the concept of “算电协同” (suandian xietong, or computing-electricity synergy), a policy framework designed to dynamically pair energy-intensive computing tasks directly with regional power supplies.

To achieve suandian xietong, China has accelerated its landmark “East Data, West Computing” (东数西算 dongshu xisuan) project. The policy routes massive data processing workloads from the land-constrained, high-cost coastal megacities of the east (like Shanghai, Hangzhou and Shenzhen) to the vast inland provinces of the west including Inner Mongolia, Gansu, Ningxia and Xinjiang.

On paper, the synergy appears brilliant. Western China possesses massive reserves of coal, alongside world-leading wind, solar and hydroelectric infrastructure. Cheap electricity, vast tracts of inexpensive land and cold ambient winter air make western hubs like Ulanqab in Inner Mongolia ideal destinations for “data capitals”. 

However, this policy framework contains a glaring geographical paradox: northwestern China is energy-rich, but severely water-scarce. Unluckily for the digital roadmap, the very regions best endowed with coal, hydro, wind and solar power are precisely those suffering from acute hydrological deficits. Over 70% of China’s data centre capacity is built in areas classified as suffering from moderate to extreme water stress. In dry northern and western provinces where annual rainfall is sparse and subterranean aquifers recharge at painfully slow rates, relying on traditional evaporative cooling systems to keep AI servers operational threatens to push regional watersheds past their breaking point.

Global warming exacerbates the crisis

This existing resource bottleneck is being dramatically worsened by the accelerating impacts of global warming. Climate change acts as a force multiplier, creating a compounding feedback loop that squeezes both computing facilities and agricultural watersheds simultaneously.

As global temperatures rise and extreme heatwaves become more frequent and prolonged, ambient air cooling systems in data centres lose operational efficiency. When outdoor air temperatures cross critical thresholds, dry cooling can no longer safely dissipate server heat. Facility operators are forced to switch to or ramp up evaporative water cooling to prevent thermal hardware failure. Consequently, data centres consume their maximum volume of water at the exact moments when ambient weather is hottest and regional water stress is most acute.

In northwestern China, including the dry corridors of Gansu, Xinjiang and Inner Mongolia, agricultural irrigation and river systems depend heavily on seasonal snowpack and glacier melt from mountain ranges such as the Qilian and Tianshan mountains. Global warming is accelerating glacier retreat across the Qinghai-Tibet Plateau. While rapid melting initially causes unpredictable seasonal flooding, it ultimately leads to the long-term depletion of headwaters, shrinking river runoff during the crucial summer farming months.

Simultaneously, warmer summer temperatures increase crop evapotranspiration rates. Corn, wheat and forage crops require significantly higher volumes of irrigation water per acre just to survive the summer heat. As a result, global warming drives agricultural water demand to its peak at the precise time that data centre cooling needs reach their maximum, triggering severe seasonal water deficits across rural basins.

Agriculture and food security under strain

Unlike standard municipal indoor water use, where water passes through drains, enters treatment facilities, and flows back into local river basins, evaporative cooling in data centres converts liquid freshwater into atmospheric vapour. This water is permanently lost to the local watershed, directly reducing the total volume available for farm canal irrigation.

In water-stressed zones across Inner Mongolia and Gansu, industrial-scale groundwater extraction by tech campuses lowers localised water tables. As shallow wells dry up, local farmers are forced to expend scarce capital drilling deeper wells and purchasing higher-capacity pumps. For family farms operating on razor-thin profit margins, these rising capital expenditures can mean the difference between economic viability and bankruptcy.

Technology enterprises, backed by state initiatives or massive corporate capital, operate with profit margins that dwarf those of traditional farming. When local municipal authorities manage water extraction quotas or set industrial utility rates, high-tech facilities can easily outbid agricultural cooperatives. This economic imbalance risks forcing arable farmland out of production, converting fertile agricultural soil into dry, fallow land and threatening regional food security.

Striking a balance: technical and policy measures

Resolving China’s East Data, West Computing water dilemma — and navigating similar conflicts worldwide — does not require dismantling digital infrastructure. Instead, it mandates enforcing strict ecological boundaries, adopting advanced cooling engineering, and implementing water-aware spatial planning.

In water-deficient provinces like Inner Mongolia and Ningxia, regulatory bodies must ban traditional open-loop evaporative cooling for new construction. Data centres should be mandated to adopt air-cooled and closed-loop systems which continuously recycle heat-transfer fluids with near-zero evaporative loss.

Direct liquid cooling isolates microprocessors in non-conductive fluids, drastically improving heat dissipation efficiency without consuming freshwater. Governments need to enforce rigorous water usage effectiveness limits — capping water consumption below 1 litre per kilowatt-hour.

Data centre operators should be legally barred from drawing potable groundwater or tapping pristine agricultural aquifers. Facilities must be required to build dedicated supply pipelines to utilise treated municipal wastewater or industrial effluent for cooling needs, leaving clean freshwater reserved exclusively for human consumption and crop irrigation.

Under the 15th Five-Year Plan’s vision for suandian xietong, computing networks should incorporate hydrological data into workload dispatching. Non-latency-sensitive tasks — such as training large language AI models or processing background data batches — can be dynamically routed to data centres located in water-abundant southwestern regions during periods of peak local heat and drought.

Tech conglomerates operating in rural western basins should be legally required to invest in local agricultural water resilience. By funding drip-irrigation retrofits, lining unpaved irrigation canals, and installing smart soil-moisture sensors for nearby farms, data centre operators can save vastly more water for the local watershed than their server farms consume, generating a net-positive ecological return for rural farmers.

China’s East Data, West Computing strategy and its 15th Five-Year Plan reflect a bold vision for an AI-driven, digital economy anchored by suandian xietong. Yet its ultimate sustainability hinges on physical foundations anchored in fragile natural ecosystems, drawing from the same finite aquifers that nourish crops and sustain rural communities.

Whether in the grasslands of Inner Mongolia or the arid valleys of the American West, powering the digital future cannot come at the expense of draining the fields that feed humanity. By pairing technological innovation in server cooling with firm regulatory protections for agricultural watersheds, China and the global community can build a digital economy that thrives in harmony with the natural world.

Related: China’s AI race: Can it make energy and computing work as one? | The true price of AI: Water and energy demands behind Asia’s data centre boom