Currently, about 80% of the 9,000 data centers in the world are built in climates that are not optimal for the equipment to operate efficiently. By 2030, their number could triple.
Currently, about 80% of the 9,000 data centers in the world are built in climates that are not optimal for the equipment to operate efficiently. By 2030, their number could triple.
According to the Rest of World report, the optimal temperature range for a data center is between 18°C and 27°C. However, of the 8,808 active data centers, 7,000 are built in locations where the climate typically falls outside this range. In addition, 600 (or about 10% of all data centers) are built in locations where temperatures regularly exceed the moderate limit of 27°C, PC Gamer reports.
According to the International Energy Agency, data centers are already consuming a lot of energy, collectively absorbing 1,5% of total global consumption (or 415 terawatt-hours) in 2024. To this figure must be added the work of the cooling system, which has to cope with the climate in which it is located, which adds more strain on the local power grid.
The study highlights India, where about a third of the 213 data centers are built in areas where it is extremely hot. Some of these data centers operate on an unstable local power grid, and the additional load from cooling needs potentially increases the risk of outages.
Experts predict that within the next five years, on-chip cooling and immersion cooling will become standard features. For example, large-scale commercial deployments of seawater cooling—like the «AI Atlantis» off the coast of Hainan Province, China—will become possible and widespread over the next decade. However, even if all of this happens, newer data centers will be the first to adopt it, while older facilities will either face expensive upgrades or become obsolete—and ultimately cooling will still be a problem.
According to a previous report by the IEA, energy consumption by AI could quadruple over the next few years, and companies are looking for alternative energy sources. AI company Lambda already operates a data center powered by hydrogen fuel cells. However, this use case may be difficult to scale up because the hydrogen supply infrastructure is not as well developed as, for example, gas supply infrastructure.
