By Nilhari Neupane - 14 August 2026

The servers may be national. The rivers are not.

The next big argument over South Asia's rivers may not begin with a dam or an irrigation canal. It may begin with a data centre.

AI looks weightless on a screen, but data centres need firm electricity, cooling, land and fibre. The International Energy Agency expects data-centre electricity use to rise sharply this decade. In South Asia, that demand will arrive where power is uneven and water already supports cities, farms, fisheries and ecosystems.

The build-out is already moving. India's IndiaAI Mission is expanding compute capacity. Pakistan has approved an Artificial Intelligence Policy 2025; Bangladesh has a draft National AI Policy for 2026-2030; Nepal a National Artificial Intelligence Policy; Bhutan a National AI Strategy; and Sri Lanka an AI Strategy. Different labels, same direction: more compute and more demand on physical infrastructure.

Planning this country by country would ignore the obvious: South Asia's electricity trade and great rivers already cross borders.

Nepal and Bhutan have large hydropower potential relative to their domestic systems. India and Bangladesh are much larger power markets. The World Bank notes that wet-season hydropower in Nepal and Bhutan complements peak demand in India and Bangladesh.

Regional trade can turn that difference into value. A more integrated electricity market among Bangladesh, Bhutan, India and Nepal - the BBIN countries - could save an estimated US$17 billion in capital costs, according to the World Bank, because countries can share generation and reserve capacity instead of each building separately for its own peak.

If a large data centre in India or Bangladesh relies on imported Himalayan hydropower, part of its footprint moves upstream. New demand can encourage dams, storage and transmission in Nepal or Bhutan while much of the digital value remains downstream.

If some compute is built upstream, Nepal or Bhutan could turn part of that electricity into higher-value digital services and local skills. But it would also reduce exportable power and add a year-round load to seasonal systems. Upstream siting changes the trade-off; it does not remove it.

That choice also affects local people. A hydropower project or new transmission corridor can take land, change access to rivers and alter livelihoods long before a data centre creates its first digital job. If upstream countries are asked to supply the clean power behind a regional AI economy, affected communities should share in the benefits through reliable electricity, local procurement, skills, revenue-sharing or other locally negotiated gains.

There is an important reason to consider upstream compute. Hydropower is generally an instream or largely non-consumptive use: water passes through turbines and continues downstream. The US Department of Energy describes the same basic feature.

The same flow can therefore generate electricity in Nepal or Bhutan and later irrigate crops, support fisheries, supply towns and sustain downstream ecosystems. Water can create several layers of value as it moves through a basin.

But hydropower water and data-centre cooling water are different. Evaporative cooling can consume water. Upstream facilities make more sense where cooling is low-water, dry or closed-loop and seasonal withdrawals are transparent.

This multiple-use logic is especially important in South Asia, where a river is rarely serving only one purpose. A flow that produces power in the mountains may later water rice or wheat, support a fishing household, maintain a wetland or supply a town. The policy goal should be to preserve that chain of benefits rather than maximise only the first use.

And non-consumptive does not mean harmless. Reservoirs and diversions can change flow timing. For farmers, fishers and river communities, a disrupted fishing season or irrigation schedule can matter as much as annual water volume.

A downstream data centre can still reshape investment upstream. More demand for imported hydropower can bring export revenue, but it can also mean new dams, transmission corridors, land acquisition and pressure on river-dependent communities.

India and Bangladesh also face direct competition for water among cities, industry, irrigated agriculture, fisheries, wetlands and households. In stressed locations, large cooling demand can make that competition harder. Dialogue Earth's reporting from India already shows why siting matters.

For downstream farmers and fishers, the issue is not abstract. Irrigation depends on water arriving when crops need it. Fisheries depend on healthy flows and river habitat. Poorly coordinated energy development can therefore shift costs onto people who may never use the digital services that created the new electricity demand.

So the regional question is not simply, 'Which country has cheap renewable electricity?' It is: where can a given workload be hosted with the least pressure on water, the most reliable low-carbon power and the fairest distribution of benefits and costs?

The Indus, Ganges and Brahmaputra are shared river systems. Their water produces hydropower in the mountains, food in the plains, fisheries and wetlands along the way, and drinking water across the region.

Water should therefore be treated as a regional good in development planning - not as a claim that erases sovereignty, but as recognition that one country's decision changes another country's options. National ownership does not make regional consequences disappear.

Thinking regionally also changes how value is measured. The best project is not necessarily the one that earns the highest power tariff or attracts the largest data-centre investment. It is the one that creates the greatest combined value from electricity, food production, domestic water, fisheries, ecosystems and digital services while keeping risks manageable across the basin.

This logic is well established. Research on water-energy-food policy coordination in South Asia shows how decisions in one sector can shift costs into another. Work on moving 'beyond hydropower' and renewable energy in South Asia makes the same case for capturing multiple upstream and downstream benefits.

The practical response is not a new South Asian AI bureaucracy. It is a regional-impact test for major compute investments.

Before approving a large data centre, governments should know its dry-season power source, effects on electricity trade, any new dams or transmission it would require, how much cooling water it would consume, and what happens to farmers, fishers and communities sharing the river.

In practice, that could mean a simple regional-impact statement for very large facilities: disclose the source of firm electricity, expected cross-border power flows, any new generation or transmission required, seasonal cooling-water demand and likely effects on river-dependent livelihoods. Neighbouring countries would not receive a veto over national investment, but they would have the information needed to plan around consequences that cross the border.

Location should follow the workload. Low-latency services may belong near major users in India or Bangladesh. Flexible workloads such as some model training or scientific simulation could sit closer to clean upstream generation if fibre is reliable, cooling is water-efficient and local communities receive real benefits.

That is a better regional bargain: downstream markets gain digital services, upstream countries capture more value from clean power, and river water can continue to support agriculture, fisheries, households and ecosystems.

South Asia should not stop building AI infrastructure. It should stop pretending the infrastructure is only national. Put servers where they create the most value with the least regional cost, and make river communities part of that calculation from the start. Otherwise, the servers may be built nationally, but farmers, fishers and river communities will pay regionally.

Dr Nilhari Neupane is an applied economist whose work focuses on water, energy, climate resilience and development policy. His research includes regional co-operation across water-energy-food sectors , hydropower, renewable energy and policy coordination in South Asia. He has more than 15 years of research and development experience in South Asia. Currently,  he is Visiting Faculty at Nepal Open University.

Photo by Orhan Akbaba from Pexels