Jensen Huang’s recent tour of South Korea was far more than a routine corporate visit; it was a calculated move to fortify Nvidia’s dominance across the entire AI hardware stack. By signing three distinct agreements with SK Hynix, Naver, and Doosan Group, Nvidia is expanding its reach beyond GPU sales into the critical realms of advanced memory, sovereign cloud infrastructure, and the industrial engineering of data centers. These partnerships signal a shift in Nvidia's strategy, moving from being a component supplier to becoming the lead architect of the physical AI era.
Central to this visit was Huang’s sobering assessment of the global semiconductor landscape. He warned that the industry is facing a structural deficit in wafers, advanced packaging, and silicon photonics that will persist for "many years." By locking in a multi-year development agreement with SK Hynix for High Bandwidth Memory (HBM), Nvidia is attempting to front-run a scarcity that threatens to throttle the scaling of its next-generation Blackwell chips and beyond.
The collaboration with Naver and Doosan highlights the broadening of the AI bottleneck. Naver, a leader in the Korean internet space, is transitioning from renting external compute to building its own core AI infrastructure using Nvidia’s latest technology. Meanwhile, the involvement of Doosan Group focuses on the heavy-duty physical requirements of massive data centers—energy systems, cooling, and robotics—proving that power and heat management are now just as vital as the chips themselves.
This aggressive integration into the South Korean tech ecosystem underscores a new phase in the global AI race. The competition has moved beyond software model training to a brute-force struggle for physical infrastructure. By securing these long-term commitments, Nvidia is ensuring that its partners remain at the front of the line for scarce resources, while simultaneously setting the standards for the giga-watt scale data centers of the future.