Nvidia and SK Group Commit Over $500 Billion to AI Infrastructure and Memory Development

Unprecedented Scale in AI Investment

Nvidia and South Korea's SK Group announced a joint initiative worth more than $500 billion on Friday, targeting the buildout of large-scale AI data centers and development of next-generation memory technology. The partnership positions Seoul as a major hub in the global race to build computing infrastructure for artificial intelligence workloads.

The agreement spans multiple SK subsidiaries and establishes a long-term supply relationship between Nvidia and SK Hynix for high-bandwidth memory, a critical bottleneck in AI chip performance. SK Hynix will secure supply of its next-generation memory products to Nvidia while the two companies jointly develop HBM technology tailored for AI training, autonomous agents, and physical AI applications, according to Nvidia.

Data Center Plans Take Shape

SK Telecom plans to construct a 2-gigawatt AI data center powered by Nvidia's upcoming Vera Rubin chips and SK Hynix's HBM4 high-bandwidth memory, with the first facility scheduled to come online in 2027, Nvidia said. The scale of the power commitment reflects the enormous energy demands of training and running large AI models. For context, a 2-gigawatt facility would rank among the largest data centers globally by power capacity.

The Vera Rubin chip represents Nvidia's next-generation architecture following its current Blackwell platform. Pairing it with HBM4 memory, which is still under development, signals both companies are planning infrastructure years ahead of current technology availability.

Expanding Existing Footprint

Separately, Nvidia announced it will work with South Korean internet company Naver and infrastructure investor Brookfield to expand Naver's existing AI data center in South Korea. The announcement did not specify investment amounts or timelines for this expansion, though Naver has been building out AI capabilities to compete with global cloud providers in the region.

The collaboration with Brookfield, a Canadian asset manager with significant infrastructure holdings, suggests capital for these projects will come from a mix of corporate balance sheets and institutional investors seeking exposure to AI infrastructure.

Memory Supply as Strategic Leverage

The supply agreement with SK Hynix addresses one of Nvidia's most pressing supply chain constraints. High-bandwidth memory has been in tight supply as demand for AI accelerators has surged, with SK Hynix and Samsung controlling the majority of global HBM production. Securing long-term supply gives Nvidia more predictability in ramping chip production and puts competitive pressure on rivals who lack similar arrangements.

HBM4, the next iteration of the technology, is expected to offer significant improvements in bandwidth and power efficiency over current HBM3E memory used in Nvidia's latest chips. Joint development allows both companies to optimize the interface between processor and memory, a key determinant of overall system performance in AI workloads.

South Korea's AI Ambitions

The initiative underscores South Korea's strategy to position itself as a cornerstone of the AI supply chain, leveraging its dominance in memory manufacturing and telecommunications infrastructure. The country already hosts significant semiconductor fabrication capacity and has been courting foreign chipmakers and hyperscalers to build data centers locally.

SK Group, one of South Korea's largest conglomerates, operates businesses spanning semiconductors, telecommunications, energy, and chemicals. The diversified structure allows it to marshal resources across subsidiaries for large infrastructure projects that would challenge standalone companies.

What Comes Next

Execution risk remains substantial given the scale and complexity of the commitments. Building multi-gigawatt data centers requires not only capital but also securing power supply, cooling infrastructure, and regulatory approvals in a country where land and energy resources are constrained. The 2027 timeline for the first SK Telecom facility is aggressive given typical data center construction cycles.

Meanwhile, memory technology development carries its own uncertainties. HBM4 is still in the engineering phase, and any delays in bringing it to volume production could cascade through the data center deployment schedule. Both companies will need to coordinate product roadmaps tightly to meet the announced timelines.

The partnership reflects a broader pattern of vertical integration and long-term supply agreements reshaping the AI hardware industry, as companies seek to lock in scarce resources and reduce exposure to spot market volatility.