There are two kinds of business trips. The first is when you go somewhere to sell something. The second is when you go somewhere to buy something. Jensen Huang's visit to Seoul was unambiguously the latter. What he wanted was semiconductors—or more precisely, the kind of semiconductors that cannot be bought in sufficient quantities anywhere on earth.

The moment Huang stepped off his plane at Gimpo Airport on June 5th, he stated his purpose plainly. "The primary objective is supply-chain co-ordination," he told waiting journalists. Nvidia needs to produce AI chips in "enormous quantities," he said, and high-bandwidth memory (HBM)—the critical component that makes those chips work—remains in short supply. The chief executive of the world's most valuable company had come, in effect, to knock on his suppliers' doors. Rather like a purchasing manager visiting a parts vendor, only arriving by private jet.

The diplomacy of the dinner table

Over four days, Huang met Chey Tae-won, chairman of SK Group, three times. Not in boardrooms, but at a pork-belly restaurant, a fried-chicken joint and a cold-noodle house. That Huang has clearly mastered the Korean custom of conducting serious business over a meal speaks to how carefully this visit was prepared. The conclusions reached across those dinner tables were formalised at a joint press conference. SK Hynix and Nvidia announced co-development of next-generation HBM and a strengthened long-term technology partnership. Examining a sample of SK Hynix's HBM4, Huang declared it "truly beautiful." Few chief executives describe a semiconductor package in aesthetic terms.

But the story did not end there. On the fourth and final day of his visit, Huang's last official engagement was with Samsung Electronics. After meeting Jeon Young-hyun, vice-chairman and head of Samsung's semiconductor division, Jeon told reporters: "We've co-operated for a long time, but today I feel we had the best conversation yet." The talks covered HBM4 and foundry work in the near term, and joint development of HBM4E and HBM5—along with collaboration on autonomous-driving and AI chips—over the longer horizon. The apparent contradiction—Huang heaping praise on SK Hynix in the morning, then having his "best conversation" with Samsung in the afternoon—is no contradiction at all. To Nvidia, these two companies are not rivals to be chosen between; they are two suppliers it needs simultaneously.

A scarce and indispensable material

To understand why Huang felt compelled to travel to Seoul in person, one must first understand what HBM actually is. High-bandwidth memory is the essential component of AI accelerators, processing data dozens of times faster than conventional DRAM. Each of Nvidia's next-generation AI chips, the Vera Rubin, requires dozens of HBM chips stacked on top of it. Only three companies on the planet can make them: SK Hynix, Samsung, and America's Micron. Huang's announcement on arrival—that all three had passed the qualification tests for HBM4—was the signal that this critically scarce supply chain was finally coming online.

The underlying problem, however, is that demand overwhelms supply by a vast margin. Goldman Sachs forecasts that leading technology companies will spend $1.1 trillion on AI capital investment next year alone, much of which will ultimately flow into purchasing HBM. When Huang says he intends to use HBM "in massive quantities," it is not rhetorical flourish.

The foundry courtship

One aspect of the visit received less attention than it deserved, yet may prove more consequential in the long run: foundry co-operation. Jeon revealed that Samsung is already manufacturing Nvidia's autonomous-driving chips and Grok chips using 4-nanometre and 8-nanometre processes, and that discussions on next-generation process nodes are under way.

Samsung's contract chip-making business, known as a foundry (a plant that fabricates chips designed by others), has struggled for years in the shadow of Taiwan's TSMC. Yield problems cost it major contracts; losses mounted. But a deepening relationship with the world's largest fabless chip designer—a company that designs chips but does not manufacture them—could shift the entire centre of gravity of that business. A single customer of sufficient scale can rewrite a struggling division's fortunes.

Caution is warranted, nonetheless. Huang described SK Hynix as the partner that "will remain our largest memory supplier going forward." For Samsung to secure a firm place in Nvidia's core supply chain, results will matter far more than words. Jeon's own comment—"we will demonstrate this through results in due course"—was an implicit acknowledgement of precisely that reality. Yet as long as Nvidia is intent on diversifying its supply chain, the window of opportunity for Samsung remains open.

South Korea's position

During his visit, Huang called South Korea's manufacturing capability "world-class" and identified the country, alongside the United States and China, as one of the three leading nations in AI. Even allowing for diplomatic flattery, there is substance behind the remark. Huang met not only Samsung and SK, but also Hyundai Motor and LG, to discuss collaboration on physical AI—the hardware and robotics that will underpin the next wave of automation. The coming era of robots demands not just sophisticated software but world-class capabilities in physical manufacturing. South Korea, Huang confirmed over pork belly, is one of the few countries that possesses both.

The emperor spent four days in Seoul, then departed. He left behind signed agreements, a handful of photographs and a brief, sharp rally in South Korean semiconductor share prices. But the more significant legacy is simpler: he came at all. The fact that the world's busiest chief executive cleared his schedule to fly to Seoul is itself a statement—that the global AI supply chain cannot bypass South Korea. His itinerary made the argument that no press release could.

What to watch: Whether Samsung's foundry secures Nvidia contracts for next-generation process nodes. Should collaboration at the 2-nanometre level become concrete, the timeline for Samsung's foundry division to return to profitability could move considerably closer.

*Nvidia and South Korea's two semiconductor giants compared*

**SK Hynix** | **Samsung Electronics**

HBM status | Nvidia's largest memory partner | HBM4 qualified; supply commencing

Foundry | Not applicable | 4nm and 8nm collaboration ongoing

Joint development | Multi-year technology partnership signed | HBM4E and HBM5 under discussion

Huang's verdict | "Beautiful"; "largest partner" | "Best conversation we've had"