Competition over investment in artificial intelligence data centers is expanding into competition over securing electricity, deployment speed and computing service capabilities. In South Korea, companies are also moving to shorten AI infrastructure delivery times by combining factory-prefabricated modular data centers with cloud services.
According to the information technology industry on October 27, the Asia-Pacific data center development pipeline compiled by global real estate services firm Cushman & Wakefield reached a record 26.5 GW in the first half of 2026. About half of the capacity under construction was concentrated in Southeast Asia, while capacity under construction in Johor, Malaysia, increased by 91% in just half a year.
An image from Cushman & Wakefield's Asia-Pacific Data Center Update for the first half of 2026.
Contracts take six weeks, construction takes years
Even as investment grows, supplying computing resources takes considerable time because electricity must be secured and data centers built. Goldman Sachs identified electricity as a major constraint on data center construction schedules and delivery capacity in Asia. It explained that contracts with new AI customers can be signed in six weeks, but data center construction takes years, creating a lag between demand and supply.
Modular and prefabricated methods are spreading to reduce this gap. While servers and power and cooling equipment are prefabricated in factories, site preparation and foundation work proceed on location to shorten deployment times. Modules can be added to match the required capacity, but actual service launch dates depend on site conditions such as electricity connections and permits.
Overseas, U.S. company Crusoe has introduced its Spark modular AI data center. Meta installed GPU clusters in tent-like structures using aluminum frames and specialized membranes. As demand for AI computing grows rapidly, reducing the time required to complete buildings and equipment has become a competitive factor.
An introductory image of Spark Liquid, the liquid-cooled modular AI data center unveiled by Crusoe.
Electricity availability reshapes locations
The criteria for selecting locations have also become more complex. Operating an AI data center requires a reliable supply of the latest accelerators as well as buildings and electricity. Semiconductor export controls, supply chain conditions and geopolitical risks also influence investment decisions. South Korea has opportunities to use its accelerator procurement and construction capabilities, but must also assess regional electricity supply conditions.
In South Korea, the ability to secure electricity has also become a key factor determining development locations. CBRE Korea said the domestic data center market was shifting from a location-focused model to one in which development follows electricity availability. By March 2026, applications for data center electricity supply in the Seoul metropolitan area had reached 522, totaling 33,592 MW, but only 10 had received final approval.
The imbalance between demand concentrated in the Seoul metropolitan area and electricity supply is leading to development outside that area. This involves building data centers in regions with available electricity capacity and providing computing services over networks. In manufacturing, defense and the public sector, there are also moves to locate computing facilities nearby to meet on-site AI inference demand.
AI PMDC, a liquid-cooled modular AI data center built by Elice Group.
From construction to cloud services and computing exports
South Korean companies are also entering the modular data center business. iA has unveiled its NEOCUBE modular AI data center and is pursuing demonstrations. Modular data center manufacturer HDMS has begun a demonstration project with Jeonju to build containerized AI data centers near industrial sites. Collaboration combining domestically developed AI chips, cooling and cloud technologies is also expanding.
Elice Group combines modular data centers with its own cloud platform to supply everything from infrastructure construction to computing services. Its AI PMDC design involves preinstalling servers and power and cooling equipment in container modules. According to the company, modules are manufactured in parallel at a factory while electricity connections are secured and sites prepared, taking 12 to 16 weeks from a construction request to service launch.
Elice Group built a liquid-cooled AI PMDC based on NVIDIA B200 hardware in 2025. It combined this with its proprietary ECI virtualization software, which groups GPUs and NPUs into a single resource pool, and GPU cloud services. The model provides operating software and services so customers can access the computing resources they need after the physical equipment is installed.
Domestic deployment is also under way. A GPU- and NPU-based edge AI data center is being built at the Myeongji-Noksan National Industrial Complex in Busan. In June 2026, Elice Group was selected as one of multiple suppliers for the Artificial Intelligence Industry Cluster Agency's 2026 national AI data center upgrade project in Gwangju.
Elice Group also proposed an export model supplying computing resources secured in South Korea to overseas customers. Its plan is to combine data center construction with cloud operations technology to generate service revenue. Realizing this will require not only deployment speed but also stable operations, competitive computing service prices and overseas customer acquisition.
Elice Group Chief Executive Kim Jae-won said, “AI data center investment is taking place simultaneously around the world, including Southeast Asia, so it is difficult to differentiate through infrastructure construction alone,” adding, “If everything from infrastructure construction to computing services can be completed domestically, this can also expand into an export industry supplying those computing resources overseas.”
Original Korean article: Edaily
English brief by Korea Wind Intelligence, koreawindintel.com
