Asia's AI Boom Draws on Coal as Communities Face Rising Power and Water Costs

Cheap coal is helping fuel a costly expansion

Malaysia's data centres could see their electricity demand rise from 8.5 terawatt hours in 2024 to 68 terawatt hours by 2030, potentially taking as much as 30 per cent of the country's electricity supply. In Johor, developers have been told to defer certain expansion projects using water cooling until 2027. Those pressures show how Asia's artificial intelligence race is becoming a contest for electricity, water and land.

Across the region, coal is helping fill the gap between rapidly growing computing demand and the power systems needed to support it. Asia holds close to three fifths of the world's known coal reserves by some estimates. That makes the fuel relatively cheap, widely available and less exposed to overseas shipping disruptions than oil or gas.

Data centres house the servers that train and run artificial intelligence systems. They need electricity continuously, often in large quantities, and their equipment produces heat that must be removed. Operators promising service at almost any hour need power even when solar panels stop producing at sunset or wind output falls.

Alexander Kheder, a market research analyst at BMI who follows artificial intelligence infrastructure and data centre expansion, described the mismatch between computing investment and new electricity supplies.

AI demand is materialising faster than clean energy generation can be commissioned.

Solar and wind generation are expanding quickly across Asia. Their output varies, however, and batteries, other storage systems, transmission lines and dependable generation with low carbon emissions are needed to maintain a continuous supply. Until enough of that infrastructure is available, coal and gas are likely to cover the shortfall.

Technology companies gain access to reliable electricity, while nearby communities can face pollution, pressure on utilities and the costs of new infrastructure. The decisions being made now could tie computing facilities to fossil fuels for decades.

Why AI creates concentrated pressure on electricity grids

Artificial intelligence facilities place a heavier burden on electricity systems than conventional data centres because they use large numbers of specialised graphics processors and other chips at high levels of utilisation. Training a model can require thousands of chips working together for days or weeks. Popular services then need computing power to answer users at any hour.

Installed capacity and actual electricity use describe different things. Capacity is the amount of power a facility is designed to draw at a given moment; electricity use measures how much it draws over time. An announced project may include substantial future capacity, but grid planners must prepare for the possibility that much of it will operate simultaneously.

In several markets, requests for connections of 20 megawatts or more can trigger lengthy technical studies. Some operators face waits of a decade or longer for a traditional grid connection. Those delays are encouraging companies to seek electricity directly instead of waiting for public networks to expand.

The International Energy Agency expects electricity use by data centres worldwide to roughly double by 2030. They accounted for about 1.5 per cent of global electricity demand in 2024, and their share could remain below 3 per cent by 2030. Local networks face a more concentrated challenge: one large facility can consume as much electricity as an industrial plant, while a cluster can become one of a regional grid's largest loads.

Operators are considering batteries, microgrids, renewable energy parks, fuel cells and nuclear generation. A microgrid is a local electricity system serving a facility or group of users. Some companies are building gas plants beside their facilities. Where alternatives are unavailable or unaffordable, existing coal plants can receive new contracts and operate longer than planned.

A data centre designed to run for 20 or 25 years can support a power station for the same period. This is fossil fuel lock in: an immediate supply decision becomes a lasting financial commitment, even when national climate policies call for coal plants to close.

Malaysia's investment boom meets utility limits

Malaysia has become one of Asia's fastest growing data centre locations, helped by its proximity to Singapore, available land and comparatively lower costs. Cyberjaya, a technology district south of Kuala Lumpur, already contains dozens of facilities, with more planned. Google, Amazon, Alibaba, Microsoft and ByteDance have shown interest in the country's digital infrastructure.

Malaysia announced about 850 megawatts of potential data centre electricity demand in the first half of 2024 alone. A regional assessment projects annual demand rising from 8.5 terawatt hours in 2024 to 68 terawatt hours in 2030. That is an eightfold increase, potentially reaching as much as 30 per cent of the country's electricity supply.

Malaysia plans to add between six and eight gigawatts of generation powered by gas by 2030 to meet rising consumption. The additional capacity may help prevent shortages, but it sits uneasily beside the country's emissions targets. Coal and gas already supply much of the existing power system.

Adit Rahim, a communications executive aged 49 who lives near Cyberjaya, said residents were watching the rapid construction with concern. He pointed to the strain on local utilities.

The impact is going to be very, very visible.

Water is another constraint. Data centres often cool their servers by evaporating water, removing heat as the water turns into vapour. Authorities in Johor and Selangor have slowed or restricted some projects because supplies are already under pressure.

Johor's instruction to defer certain expansion projects using water cooling until 2027 places a practical limit on how quickly some facilities can grow. Malaysia's experience connects three parts of the same expansion: attracting investment, supplying electricity and preserving enough water for other users.

Water use extends beyond the server building

Coal generation releases carbon dioxide and air pollutants, but the environmental burden of artificial intelligence also includes cooling water, land, transmission infrastructure, backup generators and construction materials. Electricity generation can itself require water, adding to the demands of the computing facility.

Researchers at the University of California, Riverside and the University of Texas at Arlington estimated that global artificial intelligence demand could require between 4.2 billion and 6.6 billion cubic metres of water withdrawals in 2027. Their modelling includes water used directly for cooling and water associated with electricity generation. The projection is a range, with uncertainty because complete measurements from every facility are unavailable.

Water withdrawal means water taken from a river, reservoir, aquifer or municipal system. Water consumption is the portion that does not quickly return to that source, often because it evaporates. A facility reporting modest withdrawals can still put pressure on a local watershed, depending on how much water is consumed and the supplies available nearby.

India illustrates the local risks. Many of its data centre hubs are in areas already facing water stress. Under some cooling arrangements, a facility with a capacity of 30 megawatts can draw more than two million litres a day.

In Gautam Buddha Nagar district in Uttar Pradesh, groundwater extraction has exceeded estimated sustainable levels. Residents in nearby communities have turned to expensive water deliveries after wells ran dry. Concentrating new water demand in such places adds pressure to supplies already serving households and businesses.

India seeks manufacturing benefits beyond computing

Andhra Pradesh is the proposed location for a Google data centre with a capacity of one gigawatt, part of a project valued at $15 billion. India is planning some of the region's largest artificial intelligence facilities, and state officials see opportunities beyond housing servers.

Nara Lokesh, Andhra Pradesh's information technology minister, said the investment should help attract manufacturers of power equipment, air conditioning systems and construction materials. He described his ambition for a wider industrial base.

For me, it's not merely about the data centre.

The power electronics, the air conditioning, the building materials guy, the whole nine yards. We are now bringing all of them to my state to do the manufacturing.

Construction can bring a large temporary workforce. Once open, however, data centres are highly automated. Research cited in the region suggests that a major facility may employ only a few hundred permanent workers.

For host governments, the economic calculation therefore extends beyond the initial investment announcement. Tax revenue, skilled employment and technology transfer must be weighed against the value of the land, electricity and water committed to a project. Lokesh's manufacturing ambitions reflect an effort to spread the economic benefits beyond the server building.

Residents push for a say in expansion

Public discussion has often lagged behind construction. Several Asian countries have incomplete rules for environmental reviews of data centres, and communities may receive little information about expected electricity demand, water use or pollution before projects are approved.

In February 2026, residents in Johor held what was described as Malaysia's first public protest against the rapid expansion of data centres. Participants raised concerns about pollution and water scarcity. Similar disputes have emerged in Indonesia, where residents have questioned agreements involving community land and local water supplies.

Central Asia faces related pressures. A proposed computing hub in Ekibastuz, Kazakhstan, could require 300 megawatts and rely on coal. The city already experiences industrial air pollution.

In Uzbekistan, a planned facility with a capacity of 300 megawatts in Karakalpakstan is expected to use electricity from a gas plant. The region faces serious water and energy shortages and has been shaped by the loss of the Aral Sea.

Several dates show how quickly investment plans and resource concerns are converging:

• First half of 2024: Malaysia announced about 850 megawatts of potential data centre electricity demand.
• February 2026: Johor residents protested over data centre expansion, pollution and water scarcity.
• 2027: Certain expansion projects using water cooling in Johor have been told to wait until this year; researchers also project global AI water withdrawals of 4.2 billion to 6.6 billion cubic metres for this year.
• 2030: Malaysia plans six to eight gigawatts of additional gas generation, while a regional assessment projects data centre electricity demand of 68 terawatt hours.

A small share of global electricity demand can still translate into a large local burden when facilities cluster in a dry basin or a weak power system. Ekibastuz and Karakalpakstan illustrate why the location of new computing capacity matters as much as its size.

Storage and cooling choices can reduce pressure

Renewable generation is part of the response, but solar panels and wind turbines alone cannot guarantee continuous electricity. Power systems also need flexibility: the ability to store electricity, shift demand or bring another source online when renewable output falls.

Storage that can supply electricity for long periods could allow data centres to use renewable power beyond the periods conventional batteries support. Storage can also ease grid pressure during peak hours. Batteries, pumped hydro, thermal storage and newer chemical technologies are being tested for this role.

Some technology companies are signing contracts that connect computing demand with new power capacity. Agreements in the United States include renewable electricity supported by storage, onsite fuel cells and virtual power plants. Virtual power plants link batteries and other flexible devices so they can participate together in electricity markets. Few such arrangements have yet been deployed at a similar scale across Asia.

Nuclear generation is another option under discussion, particularly small modular reactors. They could provide steady electricity with low operating emissions. Southeast Asia has no operating commercial nuclear reactors today, and licensing, safety, financing, construction time, waste management and public acceptance would need to be addressed before nuclear power could serve the data centre boom.

Cooling design can reduce water demand. Closed loop liquid systems reuse water rather than constantly evaporating fresh supplies. Facilities can also use treated wastewater, air cooling or designs that avoid potable water. Each option involves tradeoffs: air cooling may require more electricity in hot weather, while treating and transporting wastewater also uses energy.

A facility in an area with plentiful water, additional renewable generation and closed loop cooling has a different environmental footprint from one in a drought affected region powered by an old coal plant. Requiring developers to publish electricity demand, water withdrawals, water consumption and emissions for each site before approval would make those local costs more visible.

Regional power sharing offers another option

Asia's energy security concerns help explain coal's appeal. Oil price shocks, shipping risks and geopolitical conflict can rapidly raise imported fuel costs. Coal produced within the region appears more predictable to planners, even when its health and climate costs are much higher.

The ASEAN Power Grid has already enabled electricity exchanges across borders, including through the Laos, Thailand, Malaysia and Singapore integration project. Expanded transmission links could let countries share hydropower, solar and wind resources rather than build isolated fossil fuel capacity for every new industrial load.

Households and existing businesses also need a place in power planning. Data centres can generate investment and tax revenue, but special tariffs and public subsidies may transfer infrastructure costs to other electricity users. Transparent contracts, pollution rules, water limits and developer contributions to grid upgrades could give communities a clearer share of the benefits.

Air conditioning will add another major demand on Southeast Asian grids. The International Energy Agency expects the region's stock of air conditioners to increase from almost 50 million units in 2020 to around 300 million by 2040. Cooling demand is projected to reach roughly 300 terawatt hours, close to the combined current electricity use of Indonesia and Singapore.

Malaysia's plan for a sovereign artificial intelligence cloud reflects a desire to gain domestic skills, research capacity and control over digital infrastructure. That ambition will be harder to fulfil if the country becomes mainly a cheap location for overseas companies while bearing the costs of their electricity, water and land use.

Today's contracts can shape decades of pollution

Artificial intelligence has real uses in medicine, manufacturing, weather forecasting, energy management and public services. Data centres support those benefits. Matching their growth with adequate power systems and environmental rules is central to protecting nearby communities.

Coal is filling the immediate gap because it is available, dispatchable and familiar. Dispatchable generation can supply electricity when called upon, making it attractive to operators needing continuous service. Long contracts and facilities financed around existing plants can turn that immediate response into a commitment lasting decades, increasing air pollution, water stress and exposure to future carbon rules.

A responsible expansion would connect new computing capacity to additional clean electricity, storage and stronger transmission. Recycled water, published local resource balances and community involvement before construction would help governments judge whether a project fits its location. Where basic water or power conditions cannot be met, a pause may be safer than allowing demand to outrun local capacity.

Key Points

• Coal remains a major fallback for Asia's expanding AI data centres as clean electricity infrastructure struggles to keep pace.
• Grid connection delays are encouraging private generation, batteries, microgrids and other direct power arrangements.
• Malaysia's data centre electricity demand could rise eightfold, from 8.5 terawatt hours in 2024 to 68 terawatt hours in 2030.
• Water constraints have slowed some projects in Johor and Selangor, while communities elsewhere face pressure on already limited supplies.
• Johor residents protested in February 2026; certain expansion projects using water cooling have been deferred until 2027.
• Storage, improved cooling, regional electricity trade and additional clean generation could reduce reliance on fossil fuels.
• Contracts lasting decades can bind new computing investment to coal or gas, making local approval and resource planning consequential now.