Over a single weekend in Seoul, Nvidia CEO Jensen Huang ran what looked like a diplomatic tour and functioned more like a parts list. Five back-to-back meetings on Sunday: cold noodles with Hyundai's executive chair, two gaming-café sessions with the heads of Krafton and NCSoft, a ceremonial first pitch at Jamsil Baseball Stadium, then fried chicken in Gangnam with SK Group leadership at the same Kkanbu Chicken table where the October 2025 photograph with Samsung and Hyundai went viral. On Monday came formal partnership announcements with LG Group and Doosan. Read together, the schedule maps every layer of what Nvidia now calls "physical AI," the technology stack that will move robots and autonomous machines from demos to factories.
The headline deal is silicon. SK hynix is currently estimated to hold 60 to 70 percent of the HBM4 memory allocation for Vera Rubin, Nvidia's next-generation accelerator, and that share is not negotiable in the short term. HBM4 doubles the memory interface width of its predecessor to 2,048 bits. Each Vera Rubin NVL72 rack carries 20.7 terabytes of HBM4 delivering 1.6 petabytes per second of aggregate bandwidth, a more than 2.7-fold improvement over the previous Blackwell generation. That bandwidth is the difference between large training models fitting on-chip and models constantly round-tripping to external memory. SK hynix's yield, refined across multiple GPU generations, is why Huang signed an HBM4E wafer the week before with the note "Please make more."
The stop that needs more explaining is the gaming detour. Why is the CEO of a robotics-and-AI company spending an afternoon in PC bangs with the heads of Krafton and NCSoft? The reason is physics simulation. To learn the laws of the physical world, robots and autonomous vehicles need to run millions of simulated trial-and-error attempts before they ever touch real hardware. Game studios have been building exactly that simulation infrastructure for two decades. Unreal Engine and Unity, the same frameworks that govern how a falling crate behaves in PUBG, obey the Newtonian rules a robot arm has to learn. Krafton has already spun off Ludo Robotics in California to commercialise the transition. Huang is buying access to a simulation pipeline no robotics company can match on its own.
The robotics deals fill in the rest. Doosan Robotics is folding Nvidia's Isaac framework, Cosmos world models, Newton physics engine and Jetson Thor compute into an "Agentic Robot OS" for industrial depalletising and sanding work. LG's broader agreement covers humanoid, logistics and industrial robots, plus a "physical AI data factory" that uses Cosmos to generate synthetic training data, addressing the single biggest bottleneck holding humanoids back. Hyundai, which already uses Jetson Thor in its industrial lines, is layering on autonomous driving and software-defined vehicle work. Doosan Enerbility is on the hook for the gas turbines and small modular reactors needed to power the data centres that train all of it.
Step back, and the strategy resolves into something other countries cannot easily replicate. Nvidia is not just buying Korean components. It is assembling, in one geography, a vertically coherent physical AI ecosystem: memory silicon from SK hynix, simulation software from the gaming studios, robotics hardware from Hyundai, Doosan and LG, mechatronics from LG Innotek, and a synthetic-data factory layered on top. Jeff Kim of KB Securities reduced the calculus to four words: "Nvidia needs Korea."
The implication, if the strategy works, is a sharper division of labour in the global AI economy. The United States retains the model and accelerator design layer. China presses forward with deploy-first humanoid production at scale. Korea slots in as Nvidia's preferred industrial-physical AI partner, with its gaming sector providing a simulation pipeline that turns into a strategic asset rather than a consumer hobby. There is also a vulnerability inside the design. The more layers Nvidia consolidates inside one country, the more exposed the whole stack becomes to anything that disrupts that country's industrial base.