Hyundai E&C has closed the contract for the Pohang AI Data Center, a 40-megawatt facility ordered by AI Factory Pohang PFV that will be the first in the country to use a fully precast concrete structure and direct liquid cooling. The project targets a 16.5-month build - roughly half the time of conventional methods - putting schedule compression at the center of the commercial equation for GPU-as-a-service infrastructure.
The center rises in the Gwangmyeong general industrial complex in Ocheon-eup, Nam-gu, Pohang, with an IT load of 32 megawatts. Construction began in June, and operations are slated for the second half of next year. Neo AI Cloud, the project developer, designed the site as expandable infrastructure that can house next-generation GPUs and link GPUaaS with AI platform services for domestic and international companies.
Full precast build cuts frame work from a year to two months
The 16.5-month schedule forced a departure from standard practice. Hyundai E&C applied precast concrete methods to the entire structure - slabs, walls, stairs, beams, and columns. Some commercial data centers in South Korea have used precast components for individual elements, but this is the first project to prefabricate the full concrete envelope, the company said. Frame work that would take about a year with cast-in-place methods was compressed to two months.
Schedule pressure was addressed before ground broke. Hyundai E&C ran preconstruction services alongside a fast-track approach that overlapped design, permitting, and construction preparation. Work started roughly one month after the company joined the project.
Specifications built around Nvidia's Vera Rubin chip
The facility was engineered for AI servers, not general-purpose computing. It provides an ultra-high-density environment with power density roughly 10 times that of a typical cloud data center, configured to handle GPU servers based on Nvidia's latest AI chip, Vera Rubin. A direct liquid cooling system was installed to manage the heat load from densely packed GPUs - a first for a South Korean data center, according to Hyundai E&C. Power usage effectiveness, or PUE, was designed at 1.25.
Time-to-market as the new competitive axis
"With the spread of generative AI and hyperscale AI models, competitiveness in the data center market is shifting beyond scale to a race over how quickly services can start, what we call time-to-market," a Hyundai E&C official said. "By completing the Pohang AI Data Center on schedule, Hyundai E&C will demonstrate not only the country's best construction capability but also a distinct edge in securing profitability, schedule and quality at the same time."
Hyundai E&C has delivered 19 data centers since 2004, starting with the Korea Financial Telecommunications and Clearings Institute's Bundang center and including KT's Mokdong internet data center, the Pacific Sunny data center in Yongin, and NAVER's Gak Sejong data center. The company said it plans to develop data center models tied to renewable energy and small modular reactors, matched to regional energy sources.
Why this matters for operations, real estate and construction professionals
The Pohang project signals that schedule compression is becoming a primary differentiator in AI infrastructure builds. The shift from cast-in-place to full precast for data centers - and the parallel fast-track permitting and design process - creates a template that general contractors and developers will need to evaluate for their own pipelines. For real estate and operations teams, the 1.25 PUE target and direct liquid cooling specification set a new domestic benchmark for energy performance in GPU-dense facilities, which will influence lease terms, operating costs, and tenant requirements as GPUaaS demand grows. Professionals managing data center portfolios or construction delivery should watch whether the 16.5-month timeline holds and how the precast supply chain scales for follow-on projects.
Your membership also unlocks:
