SK Group is strategically planning to build an AI data center testbed in the United States.
The primary motivation for this move is to gain direct, hands-on experience in data center design and operations. In their existing Ulsan AI Data Center project, SK partners with Amazon Web Services (AWS), but a key detail is that AWS operates its "AI Zone" independently. This structure means crucial operational know-how doesn't transfer to SK, creating a knowledge gap the company is now determined to fill on its own.
This isn't just a standalone project; it's a key piece of a larger global strategy. First, Chairman Chey Tae-won recently announced plans to build an "AI factory" in Japan by 2028-2029. This signals that expanding AI infrastructure overseas is a top priority for the group. A U.S. testbed would serve as a pilot program to standardize technologies, talent development, and operational procedures before rolling them out on a larger, multi-regional scale.
The timing is also driven by several converging factors. Second, SK hynix, a world leader in HBM (High Bandwidth Memory) for AI, provides the industrial logic. As the AI memory supercycle continues, it makes strategic sense for SK to integrate "downstream" into the data center infrastructure where its chips are used. Third, favorable U.S. industrial policy, such as the CHIPS Act incentives, creates a welcoming environment. Finally, with data centers driving a surge in U.S. electricity demand, mastering the art of securing power and building efficient facilities is becoming a critical competitive advantage.
Concrete steps are already being taken. SK hynix America recently purchased a property in San Jose, California, for nearly $50 million. This existing building provides a physical footprint, potentially accelerating the timeline for getting a pilot facility up and running compared to building on undeveloped land. This move transforms the "California testbed" idea from speculation into a tangible plan.
In essence, the U.S. testbed is a calculated move for SK Group to build its own capabilities. It's about transforming from a component supplier and capital partner into a full-stack player that controls the design, construction, and operation of the critical AI infrastructure of the future.
- HBM (High Bandwidth Memory): A type of high-performance memory used in GPUs and other processors for AI and high-performance computing. It stacks memory chips vertically to achieve higher bandwidth and lower power consumption.
- Testbed: A controlled environment used for testing new technologies, products, or processes. In this case, it's a smaller, experimental data center to refine designs and operations before building larger facilities.
- CHIPS Act: A U.S. law aimed at boosting domestic semiconductor manufacturing and research through financial incentives and subsidies.