SK Group shifts focus to efficiency in AI infrastructure, chips

SK Group highlighted its focus on efficiency in AI infrastructure and memory chips at the 2025 SK AI Summit in Seoul on November 3.

Chairman Chey Tae-won said the company is prioritizing efficiency over scale to address the sharp rise in demand for AI computing power, which is outpacing current supply.

Chey cited challenges in scaling production of high-bandwidth memory (HBM) chips, noting long lead times and unpredictable demand.

SK hynix, the group’s memory chip affiliate, plans to start operations at a new HBM facility in Cheongju next year, and aims to complete a large-scale memory cluster by 2027.

SK Group is working with Nvidia to automate chip production using AI-powered systems and digital platforms.

The summit included participation from global partners such as Nvidia, AWS, and Schneider Electric, as well as several Korean AI startups.

SK Telecom announced new initiatives to expand its AI data center business, and develop energy-efficient solutions with other SK affiliates.

🔗 Source: The Korea Times

🧠 Food for thought

Implications, context, and why it matters.

SK hynix HBM expansion looks aggressive amid technical and supply risks

SK hynix finished HBM4 (the fourth generation of High Bandwidth Memory (HBM)) in September 2025 1, with mass production slated for Q4 2025 2 and the M15X fab in Cheongju targeting construction completion by end of 2025 3.
The 2027 goal for a “large-scale memory cluster” requires parallel ramps at Yongin in South Korea and Indiana in the U.S. 2. That often dents yields (the share of chips that meet specifications) and complicates logistics.
With 62% HBM share 2 and 2025 supply nearly sold out 4, contracts with Nvidia plus other clients cover 2025 2, so delays in 2027 could slow AI rollouts Chey flagged.
Chey’s efficiency push contrasts with rapid adds, so SK hynix may lean on Advanced MR-MUF (a packaging process called mass reflow molded underfill that stacks HBM dies while reducing warpage and defects) 1 to protect yields.

Korean AI data center builders can tap efficiency gains through cooling and power partners

Three AI data center efforts are underway. A 3 gigawatt (GW) site breaks ground in 2025 for 2028 completion 5 6, a BlackRock-backed hyperscale hub (a very large data center campus) is planned within a decade 7, and OpenAI may build a facility 8.
The 3 GW project calls out “advanced cooling infrastructure” with large energy load swings 5. That opens work for liquid cooling (coolant circulated to chip cold plates) plus immersion cooling (submerging servers in dielectric fluid), along with distributed energy resource management (DERM) vendors whose software orchestrates on-site batteries and generators with renewables.
SK Telecom plans efficient AI data center solutions with SK affiliates. That creates bids for power distribution and renewable integration, plus Data Center Infrastructure Management (DCIM) tools that cut Power Usage Effectiveness (PUE).
For builders, the 2025 to 2028 window for the 3 GW site 5 plus the BlackRock hub’s renewable needs 7 demand Korean partners with grid compliance within 12 to 18 months.