SK Group has unveiled one of the most ambitious business plans in the global race for artificial intelligence. The South Korean conglomerate projects to mobilize around 2.1 trillion won, approximately $1.36 trillion USD at the exchange rate used in the announcement, to expand advanced memory production and build a network of data centers up to 15 GW in South Korea. The figure outlines a long-term strategy to be executed in phases, depending on demand, funding, permits, and corporate approvals.
Key points of SK’s AI mega-plan in 20 seconds
- Projected Investment: about 2.1 trillion won, allocated between semiconductors and AI data centers.
- Chip production: SK Hynix plans to dedicate 1.1 trillion won to new factories and equipment for DRAM, NAND, HBM, and advanced packaging.
- Data centers: SK Telecom aims to develop up to 15 GW of capacity in South Korea by 2035.
- Initial targets: the group intends to begin progressively opening around 5 GW starting in 2029.
- Initial project: the Ulsan data center, developed with AWS, will start from 100 MW and is scheduled to be operational in the second half of 2027.
- A conditional figure: not all capital will come directly from SK or be approved for immediate execution. The group considers technological partners, project financing, external investors, and clients who sign long-term capacity agreements.
The strategy combines two of the industry’s most demanded resources for AI: memory modules powering accelerators and facilities housing thousands of GPUs. SK Group aims to shift from just selling components to participating more broadly across the value chain, from chip manufacturing to leasing computing capacity.
During a presentation, conglomerate chairman Chey Tae-won emphasized that South Korea should not just export AI products. His proposal advocates for developing the necessary infrastructure domestically to produce and commercialize AI-based services, with large data centers that the group describes as “intelligence factories.”
However, the scale of these announcements warrants some caution. The 2.1 trillion won is not a fixed investment for the coming months. Instead, it reflects several long-term industrial programs, some extending to 2035 and subject to market conditions. SK Hynix has stated that investments will be phased and will require approval from its boards of directors.
SK Hynix accelerates memory production for AI
The industrial component primarily falls to SK Hynix, one of the world’s largest memory manufacturers and a leading supplier of HBM for AI accelerators. The company plans to invest approximately 1.1 trillion won in its Yongin and Cheongju facilities, as well as establish a new manufacturing base in southwestern Korea.
The largest portion, around 600 billion won, will go to the Yongin semiconductor complex. SK Hynix has advanced the completion of its fourth factory at this site to 2033—twelve years earlier than previously planned, which had scheduled completion for 2045.
Yongin is expected to become one of the world’s largest DRAM production hubs. These memories are used both in conventional servers and in HBM modules accompanying GPUs and other accelerators. Their ability to transfer large data volumes rapidly makes them indispensable in systems dedicated to training and deploying AI models.
Cheongju will receive another 100 billion won. The investment includes new NAND memory facilities, manufacturing equipment, and an expansion of advanced packaging capabilities, including the final HBM assembly phase. Packaging has gained importance because the performance of a system now depends not only on individual chip design but also on how memory and processors are integrated within the same assembly.
The remaining 400 billion won is reserved for a new complex in southwestern Korea. Its exact location has not yet been disclosed. The company will look for a site with sufficient land, access to electricity, water, and logistics infrastructure—three prerequisites for any new semiconductor factory.
This expansion reflects a bet on sustained growth in demand for AI memory, but also exposes SK Hynix to the industry’s cyclical nature. Building factories takes years and costs tens of billions, while prices for DRAM and NAND can fluctuate rapidly when supply exceeds demand. Phased implementation will allow adjustments if market conditions change.
SK Telecom aims to build up to 15 GW of data centers
The second key pillar of the strategy will be led by SK Telecom. The operator intends to evolve from its traditional telecom business into an infrastructure provider offering data centers, cloud services, and on-demand access to GPUs.
Its most ambitious goal is to reach up to 15 GW of data center capacity for AI in South Korea by 2035. This would be an extraordinary scale—even for a market that’s rapidly expanding its projects. For reference, many announced campuses in Europe range from tens to several hundred megawatts.
Deployment will start with the Ulsan data center, built in collaboration with Amazon Web Services. The first phase will have 100 MW capacity and is expected to be operational in the second half of 2027. SK has already indicated plans to expand the site to over 1 GW and turn it into the core of a cluster exceeding 2 GW in the southeastern Yeongnam region.
The company is also considering adding 1 GW in southwestern Korea. With these and other developments, SK Telecom plans to progressively open around 5 GW of capacity starting in 2029, aiming to reach the maximum of 15 GW over the next decade.
Not all this capacity is contracted or has a confirmed location yet. SK Telecom will evaluate each project based on land and electricity availability, key clients, and the possibility of agreements with major tech firms.
Funding will be another critical factor. SK estimates that a 1 GW AI data center may require around 70 trillion won due to costs of servers, GPUs, memory, electrical systems, and cooling. Extrapolating this directly to 15 GW helps explain why the group talks about investments approaching a quadrillion won.
SK Telecom may not shoulder all costs alone. The company plans to involve strategic partners, international capital, long-term contracts with major clients, and project-specific financing structures.
The group is already collaborating with AWS in Ulsan and with NVIDIA on an AI platform called AI Factory. This platform will house over 50,000 GPUs and support industrial applications such as digital twins, robotics, simulations, and manufacturing process models. Its first phase is scheduled for 2027.
A strategy linking chips, energy, and computing capacity
SK Group’s primary advantage lies in the integrated business ecosystem it can leverage for the project. SK Hynix produces memory; SK Telecom and SK Broadband have expertise in connectivity and data centers; other subsidiaries are involved in energy, construction, chemicals, and industrial infrastructure.
This integration can reduce reliance on external suppliers and facilitate the design of rack-friendly complexes. AI-specific servers consume significantly more power and generate more heat than traditional enterprise workloads, necessitating specialized electrical, liquid cooling, and backup systems.
South Korea also pursues a national objective. The government seeks to leverage the country’s semiconductor leadership to attract data center capacity typically located in the U.S. or other Asian markets. Having native memory HBM, advanced industry, and local operators offers advantages, but growth will be limited by available energy and the time required to build new lines and substations.
This announcement positions SK as one of the few groups with a comprehensive AI infrastructure strategy. Still, turning goals into operational factories and data centers will take more than a decade, requiring significant capital and clients willing to lease capacity. The true measure of the project will be in how many gigawatts and production lines are actually operational in the end.
Frequently Asked Questions
How much will SK Group invest in artificial intelligence?
The group projects around 2.1 trillion won, approximately $1.36 trillion USD based on the exchange rate used in the announcement. It’s a long-term strategy, not an immediate expenditure.
How is the announced investment distributed?
SK Hynix plans about 1.1 trillion won for semiconductor manufacturing, while data centers linked to SK Telecom could allocate close to one trillion won.
Will SK Telecom have 15 GW operational by 2035?
That is the maximum target announced. The development will happen in phases and depends on demand, power supply, clients, funding, and necessary permits.
Why is HBM memory important for AI?
HBM offers very high bandwidth and enables rapid data transfer to GPUs and accelerators. It is used in servers dedicated to training and deploying large AI models.
