The rapid expansion of artificial intelligence in recent years has put increasing pressure on the world's computing infrastructure. Companies like OpenAI and Anthropic alongside giants such as Google and Microsoft are making billion-dollar investments in this technology, and AI applications have spread from users' daily lives to the activities of large companies and government entities.
According to ITNA, this rapid growth comes with significant costs. the data centers that handle AI request processing require massive amounts of electricity, and a large portion of that energy is spent keeping servers cool.
High electricity consumption, in addition to increasing pressure on power grids and the likelihood of higher consumer electricity bills, can also increase greenhouse gas emissions in areas where electricity generation depends on fossil fuels.
In such circumstances, one idea under consideration is moving data centers underwater and using the ocean's natural capacity to cool equipment.
Microsoft's experiment; why was the underwater data center project stopped?
Microsoft was among the first large companies to practically test the idea of locating data centers deep in the sea. In 2018, the company deployed a fully sealed data center on the seabed near the coast of Scotland as part of a project called "Project Natick."
The aim of this experiment was to examine the possibility of using cold seawater to reduce the need for conventional cooling systems and to evaluate equipment performance in an underwater environment. Although the experiment's results were judged technically successful, Microsoft ultimately decided not to continue commercial development of the technology.
According to Reuters, lack of sufficient customer demand and concerns about costs were major factors in shelving the plan.
This experience showed that technical success of a technology does not necessarily mean it can be used economically and at scale.
Repairing and upgrading servers; a serious challenge in the deep sea
One of the biggest obstacles to developing underwater data centers is the difficulty of maintenance and upgrades.
In onshore data centers, technicians can access faulty servers, replace damaged parts, or upgrade equipment on site.
But in underwater facilities, such direct access does not exist. If servers fail, repairs must be performed remotely or equipment must be brought to the surface for inspection and repair.
This limitation is especially important for AI data centers, because the processors and chips used in this field are advancing rapidly and companies may be forced to upgrade their hardware at relatively short intervals.
In some cases, replacing an entire underwater unit can be more cost-effective than removing and replacing individual server chips; a factor that reduces the flexibility of these centers compared to land-based facilities.
Do underwater data centers harm the environment?
Using seawater to cool servers at first glance seems like an energy-efficient and more environmentally friendly solution, but this method is not without consequences. The heat produced by the equipment is ultimately transferred to the surrounding water and can raise the temperature of the marine environment.
An increase in water temperature, depending on its magnitude and extent, may affect dissolved oxygen and the living conditions of marine organisms.
In a case similar to the Project Natick data center, the temperature increase in the water a few meters below the facility may be only a few thousandths of a degree; a very small change that is likely limited in the scale of a single data center.
However, deploying a large number of data centers in a small area could have more cumulative effects on the surrounding environment.
On the other hand, the ocean also poses threats to equipment. Saltwater can cause corrosion of external parts and the buildup of marine organisms like barnacles on heat exchangers can reduce heat transfer efficiency.
Therefore, while the ocean can solve part of the cooling problem for data centers, it brings new challenges in maintenance, equipment durability, and environmental protection.
Moving data centers to space; a more ambitious idea
Alongside underwater plans, some tech companies are exploring a very different solution: moving AI processing infrastructure to space.
In this idea, data centers would be deployed on satellites in low Earth orbit and in the future even on or in lunar orbit or on the Moon's surface.
Such installations could use solar energy and overcome some of the limitations associated with terrestrial infrastructure.
Communications between these centers and Earth would also be established via high-speed laser links so that data can be sent for processing and results returned to ground users.
Google, under Project Suncatcher, is exploring the possibility of building AI computing infrastructure in space. According to the plan reported, the company intends to launch the first of three experimental satellites this year.
The company Starcloud is also working on developing space AI infrastructure and has announced collaboration with the space company Firefly to test the technology needed for future lunar-orbit data centers.
SpaceX, as part of a plan called "Starmind," is also seeking to create a network of AI-processing satellites in low Earth orbit.
According to the reported schedule, the company intends to demonstrate the technology in 2027.
SpaceX's experience in rocket launches and operating the Starlink satellite network could be an important advantage for implementing such a plan.
Why is building data centers in space harder than in the ocean?
Despite the appeal of processing information in Earth orbit, significant technical and economic barriers stand in the way.
Jensen Huang, CEO of NVIDIA, is among those who have pointed out the practical difficulties of implementing such projects.
One of the most important problems is the very high cost of transporting heavy computing equipment into orbit.
Moreover, contrary to common perception, cooling equipment in space is not simple. In the vacuum of space there is no convection to carry away heat, and systems must primarily use thermal radiation to get rid of generated heat.
High-energy space radiation can also damage AI chips and other electronic components.
Ultimately, repairing and upgrading space equipment will be even more difficult than underwater data centers, and any maintenance operations may incur high costs and complexity.
The future of data centers; ocean or space?
The growing demand for AI services has pushed tech companies to explore solutions that until a few years ago seemed more like science fiction.
Underwater data centers can leverage the ocean's natural cooling capacity, while space facilities offer the possibility of using solar energy and developing off-Earth processing infrastructure.
However, none of these approaches has yet become a definitive answer to AI's increasing needs for energy and processing capacity.
Microsoft's experience shows that even technically successful projects can be abandoned due to high costs and operational limitations. Space projects, too, must overcome even more complex barriers before reaching wide-scale operation.
As a result, the future of these technologies will depend not only on the ability to cool and power servers, but also on implementation costs, the possibility of repair and upgrade, and their environmental impacts.