Cooling AI data centers exceeds the energy requirement of most large cities. Plans are considered to build nuclear power stations to supply their energy. Russia and China have found different solutions than are discussed in the west: Russia has built AI data centres in Siberia and China places them on the bottom of the ocean to cool down the servers. Both ways are very efficient, while cooling the data centres can be achieved at minimal cost. Probably the Russian solution makes servicing and expanding the data centres somewhat less difficult and cheaper, because the data centres can be reached over land, while Siberian temperatures are 40 degrees below zero (in degree Celsius) 9 months of the year. Russia's hydro power generation takes place east of the Ural mountain chain, not too far from where the data centres are placed, so cheap required additional power will not encounter many difficulties.
 |
| One of Russia's hundreds of AI data centre in Siberia |
However, the Chinese solution to cool data centres relies on the coolness of ocean water all year long. There are even plans under way by the two countries to collaborate in operating efficiently cooled data centres, which would theoretically perfectly fit within the cooperation framework of the BRICS alliance. Russia and China have recently begun to make cross border traffic a lot easier, which will develop in favour of collaboration in many respects. The west is increasingly falling behind in AI development compared to China, after already threatening to lose the chip war, as well as in the quantum computer race. In addition Taiwan's top ranking chip engineers are moving to Shenzhen in increasing numbers, which means that the very top of the computer industry definitely is developing in favour of China, while the west is strangled in numerous meetings, disorderly and costly supply logistics, that consume more money than the extremely structured logistic strategy often (partly) funded by China's government subsidy programs.
How China cools its data centres
Project Overview: China has been deploying innovative underwater data centers to support the growing demands of artificial intelligence for data. Their submerged facilities are designed to store large numbers of servers in sealed, pressurized steel capsules placed on the ocean floor.
Location and Depth: The projects are located in various spots, including off the coast of Hainan and near Shanghai. Depending on the specific site, these modules sit at varying depths, with some reports mentioning approximately 35 to 40 meters (around 115 feet) below the surface.
Technological Goals: The primary driver for placing these data centers underwater is energy efficiency. By utilizing the cold seawater, the system benefits from natural cooling, which significantly reduces the energy consumption typically required by land-based data centers to keep servers from overheating.
Scale and Scope: There are reports of significant deployments, including references to 1,300 tons of AI server infrastructure, with some sources citing even higher figures like 2,000 servers. This is part of a broader trend to leverage sustainable energy sources, such as wind power, to run these facilities. A logical thing to investigate would perhaps be to use energy supplying systems that are driven by ebb and flow, since they could be placed nearby the data centre.
Operational Status: The search results indicate that some of these commercial underwater data centers have already gone live as of May 2026, marking a significant step from experimental concepts to real-world infrastructure at significantly lower energy consumption cost.
 |
| China's AI data centre on the ocean floor, powered by wind energy |
How Russia cools its AI data centres
One of the advantages of Siberia is its freezing cold and the nearby abundant hydropower is extremely cheap - while permafrost temperatures of minus 40 degrees Celsius are perfectly suited to cool hot computer server equipment. The Kremlin is planning to turn the frozen wastes of its mythical tundra into one of the biggest AI data hubs on the planet and China has already signed up to be a customer.
Russia has 194 commercial data centres, according to industry data, and until recently 85% of those were stationed in Moscow. Those figures are changing. Siberia and Russia's Far East now account for more than 15% of the national data centre footprint - and the measure is accelerating, driven by a combination of geography, energy economics and geopolitical alignment that is attracting significant Chinese corporate interest.
The fundamental proposition is simple. Cold air cools servers. Siberia has cold air for eight to nine months of the year. Data centres in hot and humid southern China spend significant resources on active cooling systems that consume additional energy and reduce overall power efficiency. In Siberia, nature provides the cooling for free for the majority of the calendar.
 |
| One lane inside a Siberian AI data centre |
The electricity price advantage is where the Siberian proposition becomes most compelling. In Russia's Far East special economic zones, electricity costs between $0.045 and $0.065 per kilowatt-hour. Russia’s landscape beyond the Ural mountains is crisscrossed by large rivers in which numerous hydropower dams were built in Soviet times to power the so-called mono-city natural resource based habitations, but not a lot of other activity is taking place there.
Eastern China's equivalent cost of electricity is approximately two to two and a half times higher. The consequence for operational economics is significant: running a 10 megawatt server farm costs approximately $475,000 per month in Siberia, compared with over $1.1million per month in Shanghai for an equivalent measure of supply of electrical power.
That gap matters enormously in a business where electricity is the dominant operational cost. Data centres globally are estimated to account for 1 to 5% of total electricity consumption, and as AI workloads have grown more intensive - requiring high-performance computing clusters running continuously for model training.
Russia freed up an estimated 1.5 to 2 gigawatts of electricity capacity in Siberia and the Far East by cracking down on illegal cryptocurrency mining, which had been consuming between 2.5 and 3 gigawatts - primarily in these regions - before enforcement action was intensified. That freed capacity is now being redirected toward cooling the servers that process the legitimate data centre infrastructure, with cleaner provenance and state support.