SK Group and NVIDIA Sign a 500-Billion-Dollar AI Infrastructure Deal

Half a trillion dollars. That is the headline number attached to the expanded partnership between South Korea's SK Group and NVIDIA, and even by the standards of 2026's increasingly gigantic AI investment announcements, it is a figure that demands attention.

The two companies announced plans for a comprehensive, multi-year collaboration covering AI factory construction and next-generation memory development. SK Group, whose subsidiaries include SK Hynix, one of the world's dominant producers of high-bandwidth memory (HBM) chips, brings something to this deal that money alone cannot buy: the manufacturing capability to produce the memory that NVIDIA's GPUs depend on to function at peak performance.

What is actually in the deal

The partnership spans several distinct pillars. On the infrastructure side, SK and NVIDIA are working together to establish AI factories, the dense clusters of GPU servers that power the training and inference workloads behind large language models and other foundation AI systems. These factories require not just GPUs but entire ecosystems of networking, cooling, and power management, and SK Group has the corporate breadth to contribute across multiple layers of that stack.

The memory angle is arguably even more strategically significant. SK Hynix is one of only three companies in the world capable of producing HBM at scale, alongside Samsung and Micron. As NVIDIA's Blackwell and now Vera Rubin GPU generations have pushed HBM consumption to new highs, locking in a deep supply relationship with SK Hynix is not just good business, it is a structural necessity.

For SK, the partnership is a signal of intent about where the group sees its future. South Korea has been navigating a complicated position in the global AI supply chain, with enormous chip manufacturing expertise but facing pressure from US export controls, Chinese competition, and the relentless pace of GPU architecture cycles.

Why this matters beyond the headline number

The 500-billion-dollar figure will naturally attract scepticism, and it should. These large partnership announcements often blend firm commitments with aspirational investment targets spread across many years and multiple subsidiaries. The actual capital deployed in year one will be a fraction of that number. What matters more is the strategic signal: two of the most important companies in AI hardware are deepening their interdependence at a moment when the geopolitics of semiconductor supply chains are intensely fraught.

The deal also reinforces a pattern that has become clear over the past 18 months. NVIDIA is not simply selling GPUs. It is assembling a coalition of manufacturing, cloud, and infrastructure partners who are collectively building the physical layer of the AI economy. SK Group joining that coalition at this scale puts it in the same tier as deals NVIDIA has done with SoftBank, Saudi Aramco's affiliates, and major US hyperscalers.

For consumers and enterprises downstream, the practical implication is that the supply of AI compute should continue to expand, which historically puts downward pressure on the cost of running AI workloads. The bigger the installed base of Vera Rubin and future NVIDIA architectures, the cheaper tokens become. That is not a guarantee, but it is the direction of travel.

South Korea's government has been actively encouraging its tech giants to position themselves at the centre of the global AI build-out, offering tax incentives and regulatory support for semiconductor investment. The SK-NVIDIA deal is the most visible result of that strategy to date, and it almost certainly will not be the last.