← Back to feed

SK Group and NVIDIA signed letters of intent on July 24 covering a $500-billion-plus strategic collaboration that spans Korea’s largest AI factory buildout and a long-term memory supply agreement. The announcement was made at the K-AI Summit in San Francisco, with South Korea’s President Lee Jae-myung in attendance alongside SK Group Chairman Chey Tae-won and NVIDIA CEO Jensen Huang.

The $500 billion figure covers aggregate commercial activity expected between the companies across multiple years, not a single payment. It bundles SK Telecom’s GPU and hardware purchases for AI data centers, SK Hynix’s memory supplies to NVIDIA under the long-term agreement, and revenue from the broader NVIDIA DSX ecosystem involved in building the factories.

The 2GW Factory

SK Telecom will build a 2-gigawatt AI factory — Korea’s largest — using NVIDIA’s DSX platform, which provides a complete blueprint combining accelerated computing hardware, networking, and software into a data center-scale system. The factory will deploy NVIDIA Vera Rubin accelerated computing powered by SK Hynix HBM4. The first AI factory is planned to come online in 2027.

The 2GW target positions Korea as a major sovereign AI infrastructure hub in Asia-Pacific. NAVER separately announced a 55MW-to-200MW expansion of its own NVIDIA DSX AI factory at GAK Sejong this week, in a separate deal with NVIDIA and Brookfield. The two announcements together make Korea one of the fastest-moving AI infrastructure markets outside the US, running on almost exclusively NVIDIA silicon.

HBM4 Supply Lock-In

The SK Hynix component of the deal formalizes and extends a long-term technical partnership already in place. SK Hynix currently supplies an estimated 60-70% of NVIDIA’s HBM requirements and holds the 60-70% share of the HBM4 market ahead of Samsung and Micron. The new agreement establishes codevelopment of next-generation AI memory solutions including HBM, spanning use cases from large language model training to agentic and physical AI.

For NVIDIA, locking in SK Hynix under a long-term supply framework is a supply chain risk management move. HBM is the binding constraint on GPU output — the ability to guarantee supply as Vera Rubin scales to tens of thousands of racks matters as much as the chip design itself.

Side Deals: Anthropic and AWS

Beyond the NVIDIA agreement, SK Telecom signed a separate memorandum of understanding with Anthropic for cooperation on building a gigawatt-class AI data center in Korea. SK Telecom also agreed with Amazon Web Services to combine capabilities in building and operating domestic AI data centers, with joint expansion planned across the region.

That makes SK Telecom a hub for multiple Western AI supply chains simultaneously: NVIDIA (compute), Anthropic (model layer), AWS (cloud infrastructure).

Jensen Huang’s Quote

“Korea has all the conditions to leap into a global AI powerhouse, including world-class networks and data centers, semiconductor technology leadership, and a vast industrial base.” Huang added that NVIDIA, SK Telecom, and SK Hynix together would “create a new generation of AI factories to lead Korea’s next growth.”

SK Group Chairman Chey Tae-won framed the investment as an identity shift: Korea moving from consuming AI to producing it. “Competitiveness in the AI era depends not only on how well AI is utilized but on how much intelligence is created.”

Key Numbers

• Total initiative value: $500B+ (aggregate, multi-year letters of intent)
• SK Telecom factory: 2GW, NVIDIA Vera Rubin DSX platform, SK Hynix HBM4
• First factory online: 2027
• SK Hynix HBM market share: ~60-70% of NVIDIA’s supply
• Separate: SK Telecom MOU with Anthropic for GW-class Korean data center
• Separate: SK Telecom + AWS data center collaboration
• Context: NAVER also expanding NVIDIA DSX factory from 55MW to 200MW this week