AI’s Infrastructure Dilemma
AI services may feel like software that lives in the cloud, but they depend on large, physical facilities packed with powerful computers. These AI data centers process the information used to train and run AI systems, and their rapid growth is turning electricity, water and grid capacity into central questions about the future of the technology.
The tension is straightforward: AI needs more infrastructure to grow, but building and operating that infrastructure can put new pressure on the communities and energy systems that support it. The same technology might also help manage some of those pressures by improving how electricity is produced, distributed and used. Whether that potential becomes a real solution remains an open question.
AI data centers are estimated to need five to 10 times more electricity than conventional data centers, while also requiring extensive cooling and reliable backup power.
In South Korea, candidates in the June 2026 local elections promoted attracting AI industries to their regions. Data centers can be presented as a symbol of investment and technological growth, but the benefits and burdens may not be shared equally. A facility can require major public infrastructure without creating a similarly large number of permanent local jobs.
Investment Is Only One Part of the Local Impact
A data center is not a stand-alone building. It needs a dependable power connection, cooling systems, water infrastructure and backup arrangements. Building this ecosystem takes significant capital, which is why large projects are often developed with multiple corporate or financial partners rather than by one company alone.
The employment case can also be more complicated than a headline investment figure suggests. A data center may involve many workers during construction, but its long-term operations are largely automated and may require a comparatively limited permanent workforce. Residents may therefore question whether the jobs and economic benefits justify the demands placed on local power and water systems.
For communities asked to host a facility, the practical questions are immediate: Will electricity become more expensive or harder to secure? How much water will the site use? What will happen to the heat it releases? And will local residents share in the economic gains? Addressing those concerns early can help determine whether a project becomes part of a region’s development or a source of lasting conflict.
Why AI Data Centers Need So Much Power
Conventional data centers already use electricity around the clock to run servers and keep them cool. AI workloads can raise demand because they use large amounts of computing power to train models and respond to users. As companies and public agencies increasingly use both their own AI systems and outside services, known as a hybrid strategy (combining in-house tools with external models), the total volume of processing may grow further.
The scale of the challenge is not limited to the computers themselves. AI equipment releases considerable heat, so facilities need cooling systems capable of keeping hardware within safe operating temperatures. They also need dependable backup capacity: if a data center suddenly stops operating, services that people and organizations rely on can be disrupted.
That risk is familiar in South Korea. A fire at the SK C&C data center in Pangyo in 2022 interrupted KakaoTalk and other mobile services, demonstrating how a failure at one facility can affect everyday communications. Since then, the incident has also served as a reminder that resilient data centers need plans for fires, power interruptions and other emergencies—not just enough electricity for normal operations.
Backup Power Has Its Own Costs
Facilities may use energy storage systems (ESS) (large batteries that store electricity for later use) and backup generators to keep operating during an outage. These safeguards can improve reliability, but they add equipment, cost and energy needs. If fossil-fuel generators are used, they can also increase emissions when they run.
The wider energy picture matters, too. South Korea’s semiconductor industry is expected to need more resources over time for manufacturing and research. As AI data centers and semiconductor facilities expand alongside other electricity users, planners must weigh competing demands rather than assessing each project in isolation.
Cooling Creates Pressure on Water Supplies
Many older data centers can cool equipment mainly with air. More powerful AI hardware produces more heat, encouraging the use of liquid cooling (using fluid to carry heat away from computer equipment). Water-based systems can be effective, but they make access to water, the effects on nearby ecosystems and the energy needed to move and reuse water important parts of the discussion.
The issue is not simply whether a facility draws water. Communities also need to understand where the water comes from, how much is required during hot or dry periods, and how it is returned or reused. A project that appears manageable under average conditions may be more contentious if it competes with household, agricultural or industrial needs during a shortage.
The impact of an AI data center is not measured by electricity use alone. Cooling, backup power, water infrastructure and the local grid all matter.
Can AI Help Solve the Problem It Creates?
AI is also being explored as a tool for running electricity systems more efficiently. As power networks add renewable sources and smaller, distributed generators, their output can be harder to balance. Distributed generation (electricity produced by many smaller sources near where it is used) can include renewable power, but its availability may change with weather and time of day.
The Korea Electric Power Corporation’s AI research group has been working on ways to use AI and data-based decision-making to help manage a more complex electricity system. The work includes cooperation with the Korea Institute of Energy Technology and efforts to support the reliable operation of an expanded renewable-energy supply. The article describes a goal of supporting 100 gigawatts of renewable energy capacity by 2030; a gigawatt is a measure of power equal to one billion watts.
In practice, software that forecasts demand and renewable generation could help grid operators decide when and where electricity is needed. For example, if wind and solar output are expected to fall while demand rises, a more accurate forecast could give operators time to adjust supply or storage. Such tools can support planning, but they do not create new electricity on their own or remove the need for adequate transmission lines and power sources.
More Efficient Power Delivery
In July 2026, LS Electric announced joint development work with German semiconductor company Infineon Technologies on direct-current power technologies suited to AI data centers. Direct current (DC) is electricity that flows in one direction; equipment in many power systems converts electricity between DC and alternating current, and each conversion can lose some energy.
Reducing the number of conversion steps could limit these losses and improve the efficiency of a facility’s power system. AI can also be used in the design and operation of infrastructure, including to identify inefficiencies and help coordinate equipment. This creates a potentially useful feedback loop: AI increases demand, while AI-supported systems may help reduce waste in meeting that demand.
But improved efficiency is not the same as lower total energy use. If AI services become cheaper or more widely adopted, overall demand could still rise even when each task uses less power. That possibility makes efficiency essential, but not sufficient on its own.
Turning Waste Heat Into a Resource
Data centers release heat that is normally treated as a problem to remove. One possible alternative is a heat pump (a system that moves heat from one place to another), which could help transfer waste heat to nearby buildings or industrial sites. Research discussions in South Korea have considered using data-center heat in district heating and industrial facilities.
The idea could make more sense where a data center is close to a steady source of heat demand, such as a neighborhood heating network or an industrial complex. It would still require pipes, control systems and a reliable match between when the facility produces heat and when nearby users need it. Heat reuse is therefore a promising design option, not a universal fix.
Cleaner Power and New Cooling Options
For now, the need for reliable electricity can conflict with climate goals. Reports have described increased reliance on diesel generators as data-center operators seek power that is available when the grid cannot immediately meet demand. Diesel may provide backup or flexible supply, but burning it produces emissions and can undermine commitments to run facilities on cleaner energy.
Potential alternatives include hydrogen fuel cells and other low-emission technologies, although their broader use depends on cost, reliability and the availability of the fuels and infrastructure they need. Companies including Microsoft and Naver have also explored investment and research related to seawater cooling or wave-powered electricity. These approaches may suit particular locations, but they are not readily applicable to every data center.
No single technology can resolve the full set of challenges. A seawater cooling system, for example, still needs to be assessed for its local environmental effects and energy requirements. A renewable power contract can support cleaner generation, but the facility still needs dependable electricity when the sun is not shining or the wind is not blowing.
What a Responsible Data Center Plan Should Include
As projects are proposed, a credible plan should look beyond construction cost and computing capacity. It should explain the facility’s likely needs and how those needs will be managed over time.
- Power: Estimate electricity demand, explain how it will be supplied, and account for backup systems and grid upgrades.
- Water and cooling: Disclose expected water use, identify the source and describe plans for reuse and environmental protection.
- Emissions: Show how the project will limit reliance on fossil fuels and measure the emissions linked to its power supply.
- Community benefits: Set out realistic expectations for permanent jobs, local investment and public infrastructure costs.
- Resilience: Explain how the facility will respond to fires, equipment failures and power interruptions without placing essential services at unnecessary risk.
The Real Test Is a Fair Transition
AI data centers are becoming essential infrastructure for services that people and organizations increasingly use. At the same time, their power and water requirements can make them unwelcome neighbors, particularly if local communities see few lasting benefits. The challenge is not simply to approve or reject new facilities; it is to decide where they belong, what standards they must meet and who should bear the costs.
AI-based grid management, more efficient power systems, heat reuse and new cooling methods could all contribute. Yet many remain under development or depend on local conditions, and none can by itself guarantee that AI’s growing electricity demand will be met sustainably. Companies, energy providers, governments and residents will need to coordinate investment and set clear expectations before conflicts become entrenched.
The decisive question is whether the infrastructure can be built in a way that delivers dependable digital services without leaving local communities with disproportionate costs. If AI is to help manage the energy shock it creates, the technology will need to be matched by cleaner power, transparent planning and practical protections for the people and ecosystems around each facility.
