USG Amandeep Gill – Keynote Speech at Dushanbe Water Process Conference

Water and AI: A Two-Way Nexus

Remarks by Under-Secretary-General Amandeep S Gill
Dushanbe Water Process Conference
Dushanbe, Republic of Tajikistan  ·  25 May 2026

Excellencies, distinguished delegates, friends,

Let me begin by thanking the Government of the Republic of Tajikistan, and the people of Dushanbe, for convening us once again around one of the most fundamental question there is: how we live with water.

We meet at a moment of two accelerations. One is visible from the mountains that ring this city: the accelerating retreat of the glaciers the world paused to honour during the International Year of Glaciers’ Preservation in 2025. The other is harder to see, though no less consequential. It is the acceleration of artificial intelligence, a technology now reshaping how societies think, plan, and decide.

The instinct is to treat these as two separate stories. They are not. They are one story, with a nexus that runs in two directions. AI can help us manage water. And water, along with energy, is increasingly what AI consumes. We will not get the first right if we ignore the second.

Let me speak to both, and then ask us to do something together.

First, the promise. Around the world, AI is quietly becoming an instrument of water security. And the most striking examples come not from the wealthiest places, but from the most exposed.

In the Ganges and Brahmaputra basins, an AI flood-forecasting system that began as a pilot in India and Bangladesh now issues river forecasts up to seven days in advance, across more than eighty countries, reaching on the order of 460 million people. Many of them are in places that until recently had almost no warning at all. A week’s notice, for a family on a floodplain, is the difference between losing a harvest and losing a life.

Closer to where we sit, researchers are turning deep-learning models onto the glaciers of Central Asia, modelling the mass balance of tens of thousands of glaciers across these ranges. These are the frozen reservoirs that feed the rivers of some 64 million people. In the Tien Shan, machine-learning models of river flow have begun to outperform the traditional hydrological models we relied on for decades, and they do so precisely in the high, data-scarce places where a measuring gauge has never stood.

And in the Jordan Valley, AI-guided irrigation, with sensors and learning algorithms advising farmers when and how much to water, has saved on the order of 30 to 50 percent of water while holding or improving yields. In a water-scarce region, that is not an efficiency statistic. It is sovereignty over a scarce resource.

Three regions and three very different problems: floods, glaciers, and fields. Yet there is one common thread. AI is most valuable exactly where the data is thinnest and the stakes are highest.

But I want to resist the temptation to present this as solved. The honest posture before a technology this young is to ask questions openly, and to sit with them. So let me leave a few questions in this room, rather than answers.

If an algorithm can warn a village of a flood, who is accountable when the warning is wrong? And who ensures it reaches the last person on the riverbank, and not only the first with a smartphone?

If AI can make a glacier legible and a utility efficient, whose data trained it, in whose language, reflecting whose landscape? A model tuned to a temperate river can be confidently wrong about a glacial one.

And if these tools become essential to managing water, how do we ensure they strengthen the hand of the hydrologist here in Dushanbe, and of Dushanbe’s downstream neighbours, rather than deepening a dependence on systems they cannot see inside, question, or own?

These are not reasons to wait. They are the design specification for doing this well.

Which brings me to the second direction of the nexus, the one we discuss far less.

AI has a physical body. It lives in data centres, and those data centres drink. In one US state alone, data centres are projected to consume tens of billions of gallons of water this year, and by one study several hundred billion by 2030. Much of that is for cooling, and much of it evaporates and is never returned. A striking share of new facilities are being built precisely in water-stressed places. By one analysis, around two-thirds of those built or planned since 2022. And today we cannot even say with confidence how much water this industry uses, because too little of it is disclosed.

For a gathering devoted to water, this cannot be a footnote. A world that uses AI to save water in the field while quietly straining the aquifer beneath its server halls has merely moved the problem. It has not solved it.

So let me offer the infrastructure builders in this room, and those not yet in it, two concrete ways to earn the trust of the communities whose water they will share.

First, disclose. Publish, basin by basin, how much water a facility will withdraw and how much it will return, before it is built and not after. Trust begins with a number a community can verify.

Second, give back to the watershed you join. Commit to closed-loop and reuse cooling that can sharply cut freshwater draw, and pair every facility with a local water-stewardship agreement, one co-designed with the community rather than presented to it, so that the arrival of a data centre leaves the basin more resilient, not less.

This is where Tajikistan’s leadership matters beyond its borders. It was this country that brought to the United Nations General Assembly, last July, the first resolution on the role of artificial intelligence in sustainable development for Central Asia. It was adopted by consensus, and it proposed a Regional AI Centre here in Dushanbe.

A region defined by shared rivers and shared mountains is, by its nature, one that must cooperate or fail together on water. That same instinct, of pooling rather than hoarding, is exactly what AI governance now needs. At the United Nations, that governance is taking shape across three pillars. On science, the General Assembly last August established an Independent International Scientific Panel on AI, to give governments a shared, evidence-based reading of the technology. On policy, an annual Global Dialogue on AI Governance now seats every Member State at one table. And on capacity, the effort to finance and close the AI divide aims to ensure that developing countries help shape these systems rather than simply inherit them. A challenge as global as water and AI calls for exactly that: common science, common rules, and shared capability.

Central Asia is well placed to show the world that countries can share the benefits of this technology the way they must share a river: as a commons.

So let me close with that call.

The water and AI nexus will not be solved by any one nation buying any one system. It will be solved by cooperation, and especially by South-South and triangular cooperation, among those who face the same monsoons, the same droughts, the same melting peaks.

Let me suggest four practical tracks.

Share the use cases. Let a flood model proven on one river be the starting point for the next, so that no country pays twice to learn the same lesson.

Collaborate on the data. Pool the hydrological and satellite records across borders, because a glacier and a river basin do not respect those borders, and neither should the dataset that models them.

Adapt open models to local context. Take open-source AI and retrain it on local rivers, local crops, and local languages, so that the intelligence is genuinely ours, and not merely rented.

And invest in people, across domains. The rarest resource in this room is neither the model nor the data. It is the person fluent in both hydrology and machine learning. Train the hydrologist who can speak to the algorithm, and the engineer who understands a watershed. That talent, more than any tool, is what will carry us.

Technology is not destiny. Our choices are. Let us make them here, with our feet beside these rivers and our eyes on those glaciers, wisely, and together.

Thank you.