Nvidia and SK Group announced a $500 billion strategic partnership covering long-term HBM memory supply, a 2 GW AI data center, and next-gen AI infrastructure development. The deal, disclosed Tuesday, positions SK Group as a cornerstone supplier for Nvidia's data center buildout through the end of the decade.
Key facts
- $500 billion combined investment over ~5 years
- 2 GW AI data center in South Korea, construction 2027
- HBM4 supply agreement through 2030
- Follows $1B Nvidia investment in Naver last week
- SK Hynix supplies ~80% of Nvidia's HBM
The partnership includes a multi-year agreement for SK Hynix to supply HBM4 memory for Nvidia's next-generation AI accelerators through 2030 According to Tom's Hardware. SK Group will build a 2 GW AI data center in South Korea, with construction expected to begin in 2027. The facility will house Nvidia's Blackwell and future Vera Rubin GPU clusters, per the announcement.
Memory supply chain lock-in
The HBM4 commitment extends the existing SK Hynix-Nvidia relationship beyond the current HBM3E cycle, which has faced yield challenges. SK Hynix currently supplies roughly 80% of Nvidia's high-bandwidth memory, and this deal locks that share through at least 2030. The $500 billion figure represents combined investment across both companies over approximately five years, per the announcement — Nvidia did not disclose its specific capital commitment.
Strategic context
The partnership builds on Nvidia's existing relationship with SK Group, which included a $1 billion investment in Naver for a Korean AI data center announced last week. It also follows Nvidia's recent $1 billion Naver investment and expansion of its SK Group pact for a Korea AI hub [per our prior reporting]. The 2 GW data center is comparable in scale to AMD's recently announced 2 GW GPU supply deal with Anthropic, announced July 22.
What's at stake
Nvidia needs guaranteed memory supply to meet demand for its next-generation AI accelerators, particularly as hyperscalers like Microsoft and Meta build out 1 GW+ clusters. The 2 GW facility alone could house roughly 200,000 H100-equivalent GPUs at current power densities, though Nvidia did not specify the exact configuration. SK Group gains a long-term customer for its HBM production lines, which are currently running near capacity.
What to watch
Watch for Nvidia's Q2 FY2027 earnings in August, where Jensen Huang may disclose the specific capital allocation for the SK Group partnership. Also track SK Hynix HBM4 yield rates — if they fall below 60%, the 2030 supply commitment could face renegotiation.
Source: tomshardware.com
