The rapid expansion of artificial intelligence applications has driven a surge in demand for data center capacity. Analysts estimated the global data center market at about $270 billion in 2025, with forecasts projecting growth from roughly $300 billion in 2026 to $700 billion by 2034 — an average annual growth rate of about 11%. North America held a 38% share of the market in 2025.
Large-scale server facilities require substantial electricity and, in many cooling designs, significant volumes of water. Building and operating such centers also involves high upfront capital expenditure and ongoing operating costs (power, cooling, security, staffing). Grid connection lead times can stretch to years in many places, and transmission bottlenecks in dense urban areas limit short-term expansion. Volatile energy prices, tightening environmental rules and local opposition (NIMBY) further complicate development.
Water stress and drought — many planned sites are in dry regions
Cooling systems, especially evaporative approaches, can be heavy water consumers. According to reporting cited in the source, a majority of planned data centers are slated for regions that experienced drought in the past year: of 809 planned sites, 517 are located in places that were persistently drought-affected over the last year. The Brookings Institution warns that the rapid spread of generative AI is driving unsustainable increases in data center energy and water demand; one estimate cited suggests a single AI-based search can consume ten times the electricity of a traditional Google search.
These pressures are pushing operators and regulators to improve hardware efficiency (liquid cooling, more efficient chips), reduce water consumption and optimize energy use.
Technical and siting responses: underground, underwater and more efficient cooling
Cooling alone can account for up to about 40% of a data center’s power draw, making it a prime target for efficiency gains. New approaches include warm liquid cooling — for example some NVIDIA AI servers are designed to operate with warm coolant (up to around 45 °C), which can materially raise efficiency. Indirect evaporative cooling systems that use a semi-permeable membrane to separate water from airflow have been developed to cut both electricity and water use.
Some operators are exploiting natural conditions for passive advantages. In Norway, the Lefdal Mine Datacenter repurposed an abandoned underground olivine mine: subsurface temperatures help reduce cooling costs and preserve surface green spaces. The underwater concept — notably demonstrated by the Natick project off Scotland — remains of interest. Recently, China installed a large commercial subsea data center near Shanghai powered by offshore wind, combining marine cooling and renewable power.
Underwater deployments carry ecological risks (seafloor disturbance, local warming), but experts cited in the source argue these risks are likely manageable with monitoring. Rick Stafford of Bournemouth University is quoted saying: “a subsea data center is probably a good idea. While cooling with seawater produces locally elevated temperatures, it will not be unmanageable.” The source also notes Microsoft paused some sea-based plans, though interest in the concept persists.
Power variability, storage and the IEA perspective
The International Energy Agency (IEA) highlights that AI training and inference produce large, rapid power swings compared with traditional data center workloads, making energy storage an important component of reliable supply. The IEA projects roughly 20–25 GW of battery storage could be deployed in data centers globally by 2030; with appropriate incentives these assets could also provide grid services. The IEA stresses that the impact of rising electricity demand on prices depends on local fundamentals and policy: in some systems it could trigger new generation and upward price pressure, while in others it could improve utilization of existing assets and reduce prices.
Supply-chain constraints — long lead times for electrical gear, switchgear, transformers and high-capacity cooling systems, as well as semiconductor shortages — slow project timelines. Growing cyber-security risks also require operators to invest in defensive architecture, increasing operating expenses.
EU regulatory steps
The European Union has moved to tighten oversight. A mandatory reporting framework and a central EU database require all data centers with at least 500 kW capacity to publicly report energy and water consumption, share of renewables, temperature settings and waste heat reuse. The first reporting deadline was in September 2024; annual submissions are due by May 15 each year.
The European Commission issued a new energy efficiency package with a labeling/qualification system in the first half of 2026, and plans to finalize mandatory minimum performance standards for both new and existing centers by 2027, based on needs assessments.
If technological innovations and regulatory measures are implemented at scale, data center electricity and water demand could decline significantly. The sector’s trajectory will, however, depend on the speed of efficiency improvements, the roll-out of storage and renewables, and regulatory enforcement. Ensuring ethical and sustainable AI operation — including addressing the market power of large tech firms and less transparent actors — remains a policy challenge.
Conclusion
Growing AI workloads create substantial pressure on energy and water systems and force a combination of technical, infrastructural and regulatory responses: more efficient cooling and hardware, on-site energy storage, alternative siting such as underground and underwater facilities, and stricter reporting and minimum standards. Coordinated progress across these domains will determine whether the industry can scale without exacerbating environmental and grid stresses.
