Microsoft says in its sustainability report that an AI-based leak-detection solution prevented more than 8.5 million cubic metres of drinking water loss in Hungary—an amount roughly equal to the annual household water use of a city the size of Debrecen.
How the system works and why it matters
The system analyses operational data from water utility networks to identify faults that cause drinking water to leak before it reaches consumers. The technology is particularly relevant for Hungary, where poor pipe conditions mean on average about one quarter of water introduced into distribution networks is lost; in Budapest the rate is roughly 16 percent. Water lost in leaking networks represents not only a depletion of a natural resource but also a financial loss, since energy has already been expended on extraction, treatment and transport.
AI across water management
Artificial intelligence can be used across the whole water-management cycle. In collaboration with the International Water Management Institute, Microsoft developed Water Copilot, which creates a digital twin of a river basin. That integrated model provides a comprehensive view of water stocks and can be used to plan irrigation and prepare for droughts. Instead of interpreting separate measurements, decision-makers can see in one unified model where shortages may emerge and where interventions are needed.
Operational impacts and corporate perspective
Bábel Gabriella, managing director of Microsoft Hungary, said companies currently rely mostly on voluntary consumption restrictions to help mitigate the crisis. Over the longer term, advanced technologies can enable operational savings that reduce costs and emissions even after emergency measures end.
Energy and water savings in buildings and data centres
AI can also reduce energy use in buildings. IoT-based building-management systems track external temperature in real time and adjust heating, cooling and other equipment. LinkedIn uses predictive AI to optimise operations in its offices and campuses—forecasting consumption changes and adjusting systems before waste occurs. Microsoft reports that an AI-based fault-detection system at its Redmond campus saved about 4,000 megawatt hours of energy in 2025. Similar building-management solutions are spreading in Hungary and can contribute to lower energy consumption.
Efficient models and dynamic resource allocation
For some Copilot products Microsoft uses dynamic model selection: the system matches computational resources to task complexity so that simple operations do not invoke larger, more energy-intensive models unnecessarily. The company recommends that customers and developers choose models, data-management approaches and load-distribution strategies that fit the specific task when designing AI systems.
Reducing data-centre water use through improved cooling
The spread of AI increases energy and water demand in data centres, making efficient cooling critical. Since 2024 Microsoft has implemented lower-water cooling solutions in Azure facilities by circulating coolant in closed loops directly around chips, reducing evaporation losses. The company estimates this approach can save more than 125 million litres of water per data centre annually. In 2025 Microsoft also trialled microchannel cooling, which routes coolant to areas of a chip that require the most cooling depending on chip type and workload, potentially tripling cooling efficiency.
Local adaptation and water replenishment
Data-centre water supply is adapted to local weather and environmental conditions to minimise stress on municipal systems. For example, the Amsterdam facility uses almost exclusively rainwater for server cooling. Since 2022 Microsoft has improved the specific water use of its data centres by 25 percent—meaning they now require a quarter less water for the same energy output—and the company has set a target of a 40 percent improvement by 2030.
In the last year Microsoft returned 14.2 million cubic metres of water to the environment; the company says this was the first time the amount replenished exceeded the water withdrawn in the same year. Committed future water replenishment under existing contracts totals 133 million cubic metres, roughly equivalent to the annual household water use of about eighteen cities the size of Debrecen.
Conclusion
Microsoft’s reported results indicate that AI and related technologies can deliver measurable water and energy savings, an especially important contribution in regions facing high network losses and increased drought risk.
