On the 1st of this month, OpenAI, the developer of ChatGPT, announced its collaboration with Samsung Electronics and SK Hynix on a large-scale data center project in the United States called 'Stargate.' As part of this, Samsung C&T and Samsung Heavy Industries plan to develop a 'floating data center' on the sea with OpenAI.



The demand for the floating data center that Samsung and OpenAI aim to develop is rapidly increasing. According to market research firm Global Information, the market size of floating data centers, which was USD 224.2 million (approximately KRW 321.7 billion) in 2023, is projected to grow more than threefold to USD 732.6 million (approximately KRW 1.0513 trillion) by 2033.

The advantages of floating data centers are clear. Firstly, being afloat on the sea allows the use of cold seawater for cooling purposes nearby. It is known that about 35% of the total energy used by ground data centers is consumed by servers, and about 50% is used for cooling. The floating data center can significantly reduce the power used for cooling, thus lowering data center operating costs.

Being able to move on the sea is also a major advantage. It can be relocated near desired areas when needed, reducing 'latency' in data transmission processes. Since land is not utilized, there is no need to secure sites, and various regulations associated with it can be avoided. However, waterproof and fireproof designs are essential to prevent corrosion from high salinity and humidity of seawater, and maintaining stability against mechanical vibrations from waves is technically challenging.

Prior to Samsung and OpenAI, Japan's Mitsui O.S.K. Lines is collaborating with Turkey's Karpowership to build a floating data center. They are developing a data center with a maximum capacity of 73 megawatts (MW) and aim to begin trial operations by 2027. They plan to reduce the construction period, which would take four years on land, to one year by modifying used ships.



China has begun operating the world's first underwater data center. The South China Morning Post (SCMP) reported on the 7th that a commercial data center was established in Lingshui County, Hainan Province, China. This data center, weighing 1,300 tons, equivalent to 1,000 passenger cars, is submerged at a depth of 35 meters.

Fu Ding, the project manager of Shenzhen HighCloud Data Center Technology, which carried out the project, explained, "Seawater helps lower the temperature," adding that "compared to ground data centers, underwater data centers reduce energy consumption, lowering operating costs." The company plans to start full-scale operations by the end of this year.



Before China, Microsoft (MS) conducted 'Project Natick,' submerging 855 servers at a depth of 36.5 meters off the coast of Scotland from 2018 and operating them for two years. Of the 855 servers contained in a metal container, only six failed, confirming that it operated about eight times more stably than ground data centers. However, the project officially concluded in 2024 and did not lead to commercialization.

"Within the next 10 to 20 years, gigawatt (GW) scale data centers will be built in space. Within 20 to 30 years, the cost of space data centers can be offered cheaper than Earth data centers."

Jeff Bezos, founder of Amazon, mentioned the feasibility of space data centers while attending the 'Italian Tech Week' in Turin, Italy, on the 4th (local time). Bezos, who founded the space development company Blue Origin, is developing space launch vehicles.

Although it may sound like a pipe dream, space data centers are closer than expected. The American company StarCloud plans to launch a small data center 'StarCloud-1' equipped with Nvidia's graphics processing unit (GPU) 'H100' into low Earth orbit by the end of this year.



Here, it will be verified whether Google's artificial intelligence (AI) agent 'Gemini' operates properly. If Gemini is confirmed to operate without issues, it will prove the commercial viability of the data center. StarCloud plans to launch the commercial small data center 'StarCloud-2' next year. Additionally, Lone Star Data Holdings, and Thales Alenia Space, a joint venture between European aerospace companies Thales and Leonardo, are also pursuing data center construction in low Earth orbit.

The reason big tech companies are pioneering data center sites from the seabed to space is due to the rapidly increasing server capacity demand from artificial intelligence (AI) development, leading to a severe shortage of data centers. Bloomberg News reported on the 10th (local time) that Microsoft's data center shortage problem will last longer than initially expected. In July this year, Amy Hood, MS Chief Financial Officer (CFO), mentioned that this capacity shortage would continue until the end of this year, but Bloomberg News, citing sources, stated it would last until at least 2026.

However, it is not feasible to indefinitely increase data centers on Earth. The acceleration of global warming due to greenhouse gases emitted by data centers has been consistently pointed out, and significant costs are involved in cooling facilities to dissipate the immense heat. In the long term, as suitable sites for expanding data center scale are sought, various locations such as the seabed, sea surface, and space are being considered as next-generation data center sites.

Popular News