Synopsis: The AI revolution looks invisible and effortless on our screens. But behind the scenes, something far bigger is unfolding. As nations race to host the future of computing, hidden pressures are building in unexpected places. This article uncovers what the digital boom is quietly demanding, and who might ultimately pay the price.
Artificial intelligence is expanding at a speed few industries have ever seen. New AI models are launched every few months, companies are pouring billions into computing infrastructure, and countries are competing aggressively to host the data centres that power this digital revolution. Governments see them as symbols of progress, proof that their nation is part of the future. Behind every chatbot response, AI image, and cloud-based service, there are vast warehouses filled with servers running day and night. The AI boom may look invisible on our screens, but in the physical world, it is anything but light.
At the same time, something less visible is happening. Electricity demand is surging in regions hosting large data centres, putting pressure on local power grids and, in some cases, contributing to higher energy costs. In water-stressed areas, massive cooling systems are drawing on already limited freshwater supplies. Communities are beginning to question why homes are asked to conserve energy and water while industrial server farms continue to expand. The digital economy promises convenience and growth, but what happens when the cloud starts drying up the ground beneath it?
What Is a Data Centre Really?
Data centres did not begin as the massive, industrial facilities we see today. They first appeared in the early 1940s, when computers were extremely large, complicated machines that were difficult to operate and maintain. Early systems, known as mainframes, required multiple bulky components that had to be physically connected using large bundles of cables.
These machines consumed enormous amounts of electricity and generated significant heat, which meant constant cooling was necessary to prevent overheating. Because of this complexity, companies placed all their computing equipment in a single dedicated room. That room became what we now call a data centre. At the time, every company built and managed its own facility.
As technology improved, computers became smaller and more energy-efficient. However, while hardware became more compact, information technology systems grew far more complex. The volume of data created and stored by businesses increased at an extraordinary pace. Virtualization allowed software to operate separately from physical machines. Advances in networking made it possible for applications to run on hardware located far away from the user. In simple terms, even though computers became smaller, the digital world around them became much bigger and more demanding.
Today, a data centre is a physical facility that houses computing machines and all the supporting hardware they require. It includes servers, storage devices, and networking equipment. It is the backbone that holds a company’s digital information. Every website, online service, internal software system, and digital transaction depends on this physical infrastructure.
All businesses, whether small or large, need computing equipment to run websites, deliver services, sell products, manage accounts, handle human resources, and oversee operations. As a company expands, its digital needs grow rapidly. Managing equipment spread across multiple offices becomes expensive and difficult. To solve this, companies centralize their hardware in data centres where it can be maintained more efficiently and at lower cost. Instead of building their own facilities, many now rely on third-party data centres to store and manage their digital systems.
The Electricity Crisis – When Servers Compete With Homes
Data centres do not just store information. They consume enormous amounts of electricity to keep that information moving. According to the International Energy Agency, around 60 percent of a data centre’s total electricity use goes directly to powering servers that process and store digital data. This demand becomes even heavier inside AI-focused hyperscale facilities.
These centres use advanced chips capable of performing trillions of calculations every second. Compared to traditional servers, these chips require two to four times more electricity to operate. As artificial intelligence expands, energy demand does not rise gradually, it accelerates sharply.
What makes this more serious is not just how much electricity data centres use, but where and how they use it. They are often concentrated in specific regions, creating local grid stress. In many places, the question is no longer about digital growth alone. It is about whether power grids built for homes and factories can handle this new wave of demand.
Ireland
Ireland’s early data centre growth was not driven by artificial intelligence. It began with cloud storage, banking systems, and the personal data networks that supported Big Tech’s earlier expansion. Since the post-war era, Ireland has built its economic strategy around attracting foreign multinational companies through tax incentives and infrastructure support. This approach proved successful during the rise of the ICT industry in the 1980s and 1990s, when American technology companies moved operations to Ireland to serve European markets.
After the 2008 global financial crisis, investment became increasingly concentrated in major technology firms. As global demand for computing power grew, Dublin experienced steady data centre expansion. The city became a digital bridge linking offshore headquarters of technology companies to the United States through undersea internet cables.
By 2017, data centre growth was not only accepted but formally included in national industrial planning. A Government Enterprise Strategy aimed to align rising enterprise electricity demand with generation capacity and transmission planning. Dublin went on to become Europe’s second largest data centre hub, with 1,150 megawatts in operation, just behind London at 1,189 megawatts, and ahead of other major European markets in the first half of 2025.
However, rapid growth created pressure on Ireland’s electricity grid, particularly around the Dublin region. As demand increased faster than supply upgrades, the regulator effectively imposed a moratorium on new data centre grid connections in 2021. While new action plans now aim to enable further expansion, the strain remains significant. Data centres accounted for 22 percent of Ireland’s total electricity consumption in 2024. EirGrid projects this could rise to 31 percent by 2034 under its median growth scenario. In a country of Ireland’s size, that represents a substantial share of national power being directed toward server farms.
United States
The United States hosts more than 4,000 data centres, according to industry estimates. A third of these are concentrated in just three states: Virginia with 643 facilities, Texas with 395, and California with 319. Major hubs such as northern Virginia, Dallas, Chicago and Phoenix dominate the landscape. About half of the data centres currently under construction are being added to existing large clusters, reinforcing geographic concentration.
Electricity consumption is rising alongside this growth. Total U.S. electricity use reached a record level in 2024. That number could increase further if data centre expansion continues at the current pace. In 2024 alone, U.S. data centres consumed 183 terawatt-hours of electricity. This represented more than 4 percent of the country’s total electricity usage and was roughly equal to the annual power demand of Pakistan. By 2030, this figure is projected to grow by 133 percent, reaching 426 terawatt-hours.
AI adds another layer of intensity. A typical AI-focused hyperscale facility consumes as much electricity each year as 100,000 households. The largest new facilities currently under development are expected to consume 20 times that amount. When concentrated in specific regions, this demand places heavy pressure on local grids. In 2023, data centres used around 26 percent of Virginia’s total electricity supply. Other states also saw significant shares: 15 percent in North Dakota, 12 percent in Nebraska, 11 percent in Iowa, and 11 percent in Oregon. These numbers show how server farms can become major power consumers within regional systems.
APAC
The Asia-Pacific region faces what can be described as a direct clash between AI expansion and limited power infrastructure. Electricity consumption linked to data centres in the region is projected to rise from 320 terawatt-hours in 2024 to 780 terawatt-hours by 2030, a 165 percent increase.
At the same time, development pipelines are expanding rapidly. In the first half of 2025 alone, nearly 2,300 megawatts were added to the pipeline. Operational capacity stands at about 12.7 gigawatts, with 3.2 gigawatts under construction and 13.3 gigawatts in planning. Bank of America expects APAC data centre capacity to double within five years, adding 2 gigawatts annually, which is twice the growth rate seen between 2018 and 2023.
Yet power availability remains the main obstacle. Nearly half of industry respondents cite electricity access as the primary barrier to completing projects. Only 32 percent of projected demand is expected to be met by renewable energy.
The region’s power grids were designed for dispersed residential and industrial loads, not for concentrated multi-hundred-megawatt campuses. A single NVIDIA GB200 deployment consumes 30 megawatts continuously, which is more than entire business districts in many Asian cities. Grid operators are receiving requests for 500 megawatt connections in areas where total substation capacity may only reach 200 megawatts. This creates a zero-sum situation where adding one large AI facility can reduce available power for thousands of homes.
Financial investment alone cannot solve the issue. Regulatory processes, geographic limitations, and long infrastructure timelines slow progress. Oracle withdrew from a planned 150 megawatt facility in Singapore after failing to secure power allocation despite two years of negotiations. Microsoft has chosen to build its own power plants in Indonesia instead of waiting for grid upgrades.
Country-level examples highlight the imbalance. Thailand’s data centre electricity demand rose 400 percent between 2020 and 2024, while power generation increased by only 8 percent. Vietnam attracts hyperscale investment due to low land and labour costs, yet experiences weekly power cuts during peak summer periods. Indonesia’s Java-Bali grid operates at 95 percent capacity even before accounting for new data centre additions. Across the region, annual electricity demand is already growing at 4.5 percent without including AI’s accelerated requirements.
South Korea benefits from nuclear energy, which provides 28 percent of its power generation and offers stable supply for data centres. However, policy shifts toward renewable energy create uncertainty about future nuclear expansion. Samsung’s semiconductor facilities in Pyeongtaek already consume 1 gigawatt continuously, and AI chip production is expected to add another 500 megawatts by 2026. Concentrated industrial demand contributed to grid instability and blackouts in Seoul during the 2023 heat waves.
Singapore faces transmission limitations. Its 230 kilovolt network cannot handle the 400 kilovolt connections required by data centres above 100 megawatts. Upgrading just 50 kilometres of high-voltage lines would cost around 2 billion dollars and take five years. The city-state’s limited land area forces transmission lines underground, which costs ten times more than overhead lines.
A hidden bottleneck across APAC is substation capacity. A 500 megawatt data centre requires a dedicated 500 kilovolt substation costing around 200 million dollars and taking three years to build. Environmental approvals can add another 12 to 18 months in developed markets. These timelines make rapid scaling extremely difficult.
Netherlands
Amsterdam has traditionally been one of Europe’s leading data centre hubs, alongside Frankfurt, London, Paris and Dublin. However, growth has slowed significantly due to electricity constraints.
The city council recently announced that new data centre applications will only be considered from 2035 onwards because of grid congestion. Around 200 megawatts of new capacity are currently under construction, but another 200 megawatts planned beyond that faces uncertainty.
The broader implications are economic. The sector supports between 150,000 and 250,000 jobs in the wider digital infrastructure industry and generates approximately 26 billion euros in annual turnover. Dutch data centres currently account for 3.3 percent of national electricity consumption. With the rise of generative AI applications, which require far more processing power than traditional computing, this share could increase significantly.
Constraints include network congestion, limited physical space, and public concerns about rising energy use. The debate in the Netherlands highlights a central tension: digital infrastructure brings economic benefits, but its energy footprint can restrict future growth.
The Water Problem – When the Cloud Dries the River
Cooling the Machines That Power the Cloud
After electricity for servers, the second major energy burden inside a data centre comes from cooling systems. Depending on efficiency levels, cooling accounts for around 7 percent of total energy use in advanced hyperscale facilities and can exceed 30 percent in less efficient enterprise setups. As artificial intelligence, high-performance computing, and cloud workloads increase rack density, cooling becomes essential to keep infrastructure stable and operational.
Cooling systems protect servers from overheating and ensure reliability, performance, and longer equipment life. The most widely used method is air cooling. It relies on air conditioning systems, fans, and ventilation to circulate air and remove heat. Because it is simple and scalable, it remains common, especially for lower-density workloads such as AI inferencing. According to a report from Cushman & Wakefield, Standard air systems typically support up to 20 kilowatts per rack, while enhanced versions can manage around 35 kilowatts. Beyond that threshold, liquid or hybrid cooling solutions are usually required to maintain efficiency, manage power usage effectiveness, and limit carbon impact.
Liquid cooling is increasingly adopted for high-density environments. Liquid immersion cooling submerges servers in a non-conductive fluid that directly absorbs heat. This allows higher computing density, reduces noise, and lowers mechanical wear, making it suitable for AI and high-performance computing workloads.
Direct contact liquid cooling circulates liquid through cold plates attached to CPUs and GPUs, removing heat precisely without immersing the entire system. Direct-to-chip cooling similarly targets high-heat components using coolant-filled plates. Although effective in thermal control and reducing reliance on air systems, shorter service cycles have limited widespread adoption in some cases.
Billions of Gallons Behind the Screens
Cooling systems often require large amounts of water. Some designs use more than others, but overall consumption is significant. In 2023, data centres in the U.S. directly used approximately 17 billion gallons of water. Hyperscale and colocation facilities accounted for 84 percent of that total, according to a 2024 Berkeley Lab report commissioned by the U.S. Department of Energy.
Looking ahead, hyperscale data centres alone are projected to consume between 16 billion and 33 billion gallons of water annually by 2028. These figures exclude indirect water consumption, such as water used in electricity generation or semiconductor manufacturing.
Researchers at the University of California, Riverside estimate that a single 100-word AI prompt can consume about 519 millilitres of water. Much of the water used in cooling towers does not return to the system. It absorbs heat and is released into the air as vapour, meaning significant volumes evaporate rather than being recycled.
Protests, Pushback and Public Anger
The United States offers an early example of how water use can trigger community resistance. In Virginia, home to the world’s largest data centre cluster, residents have protested new approvals. Concerns include groundwater depletion, noise, and rising electricity tariffs.
In parts of Arizona and Georgia, local governments have tightened zoning rules following public opposition. In drought-prone Arizona, investigative reports revealed that large facilities were consuming millions of gallons each year while residents faced water restrictions. Community resistance has slowed or stopped multiple projects.
43 Percent in High Water Stress Zones
Water risk is not limited to individual communities. According to an S&P Global analysis using Sustainable1 Physical Risk and DCKB datasets, 43 percent of data centres globally operate in areas classified as having high water stress in the current decade. High water stress areas are defined as locations with a water stress index of 40 or higher on a scale of 1 to 100, indicating high competition for limited water resources.
All data centres in Middle Eastern countries fall into high water stress zones. Similar exposure exists in parts of Europe, including Belgium, Greece, and Spain, and in Latin America, including Chile, Peru, and Mexico. In these regions, the average water stress index ranges between 80 and 100, far above the defined high-stress threshold.
The United States and China, which lead globally in data centre operational capacity, show different exposure levels. Around 60 percent of China’s data centre assets are located in high water stress areas, with an average index of 59. In comparison, 38 percent of U.S. assets face high water stress, with an average index of 43.
The Hidden Water Cost of Electricity
Water use does not end with cooling towers. Around half of the electricity currently used by U.S. data centres comes from fossil fuel power plants. These facilities consume large quantities of water to generate steam that drives turbines. This means water is used both directly inside data centres and indirectly through the energy systems that power them.
As artificial intelligence expands, the cloud becomes increasingly physical. Behind every AI response lies infrastructure that draws from rivers, groundwater, and public water supplies. In regions already facing scarcity, this growing demand is creating visible tension between digital expansion and basic human needs.
Land, Housing Pressure and Other Hidden Costs
Data centres require constant and reliable electricity, and many depend on gas-fired power for routine operations along with diesel generators for emergency backup. Both sources carry environmental and health risks. Some facilities install gas-powered systems on-site for daily use, leading to continuous greenhouse gas emissions and air pollution. In Memphis, Tennessee, more than 30 natural gas turbines are being installed at xAI’s Colossus data centre for regular operations.
Local residents and the NAACP have filed a notice of intent to sue under the Clean Air Act, arguing that the project could worsen already poor air quality in a city with high asthma rates and long-standing environmental health disparities. Diesel backup generators, though used during emergencies, release harmful pollutants such as fine particulate matter (PM2.5) and nitrogen oxides (NOx), which are linked to respiratory disease, heart disease, and asthma. These generators can emit 200 to 600 times more nitrogen oxides than natural gas plants. One analysis in Virginia estimated that even limited backup generator use could be associated with nearly 300 million dollars in annual public health costs and around 14,000 asthma-related health impacts across multiple states.
Beyond air pollution, data centres generate significant electronic waste. These facilities contain thousands of components with different life spans. Servers, batteries, and networking equipment usually need replacement every three to five years, while infrastructure such as cooling systems and generators may last more than a decade. In a 2020 survey, 42 percent of IT managers reported replacing servers every two to three years. Given the global scale of data centre operations, this results in large volumes of discarded equipment.
Server longevity depends on design, workload intensity, maintenance practices, and operating conditions. However, many upgrades happen not because equipment has failed, but because operators want improved performance, higher energy efficiency, and greater scalability offered by newer technologies. Short upgrade cycles, especially in high-performance facilities, significantly increase e-waste.
Data centres also require substantial land and infrastructure, often competing with agriculture, housing, and other commercial uses. In some areas, rapid expansion contributes to rising land prices, gentrification, and pressure on local infrastructure. Planning for green data centres requires careful zoning that accounts for power grid and water capacity, proximity to renewable energy sources, and the impact on surrounding communities and industries. While renewable energy adoption is promoted as a sustainability solution, solar and wind farms themselves require vast areas of land.
As a result, the physical footprint of digital infrastructure extends beyond server buildings, affecting land markets, public health, waste systems, and local development patterns.
The AI boom is often presented as clean, invisible and futuristic. On our screens, it feels effortless. But in the real world, it is built on heavy infrastructure that consumes electricity, water, land and public resources at an enormous scale. Data centres are not just buildings full of computers. They are industrial facilities that compete directly with homes, farms and small businesses for basic necessities like power and water.
Countries invite them for jobs, investment and technological leadership. And yes, they do bring economic benefits. But the costs are becoming harder to ignore. In Ireland, data centres now take a large share of national electricity. In parts of the United States and Asia, they are straining regional grids. In water-stressed areas, cooling systems draw from the same sources communities depend on. In some places, projects are being delayed or opposed because local infrastructure simply cannot keep up.
The bigger question is not whether AI will continue to grow. It will. The real question is whether infrastructure planning, environmental regulation and energy systems will grow at the same pace. If they do not, the digital future may begin to clash with everyday life in very visible ways, higher power bills, water shortages, land pressure and public health concerns.
The cloud is not floating in the sky. It sits on land, runs on electricity and drinks water. If countries want the benefits of artificial intelligence, they must also plan for its physical footprint. Otherwise, the race to host the future could quietly weaken the foundations that people depend on today.
Disclaimer: The views and investment tips expressed by investment experts/broking houses/rating agencies on tradebrains.in are their own, and not that of the website or its management. Investing in equities poses a risk of financial losses. Investors must therefore exercise due caution while investing or trading in stocks. Trade Brains Technologies Private Limited or the author are not liable for any losses caused as a result of the decision based on this article. Please consult your investment advisor before investing.
