Huang Sat With Xi. Then He Met Samsung and SK.

The Seat Across From Xi

Jensen Huang sat directly opposite Xi Jinping at a state dinner in the East Room on September 24, 2026. Around the head table were Lisa Su of AMD, Elon Musk, and Tim Cook — all CEOs whose companies have deeply entangled supply chains and revenue streams in China. The image was carefully staged, a public relations gesture meant to signal that American technology leaders remained engaged with Beijing even as Washington restricted the flow of advanced semiconductors. But staging and substance diverged sharply in the days that followed.

The dinner itself was notable not only for its seating chart but for what it represented: a moment when the most powerful figure in AI hardware was physically positioned at the center of a geopolitical arrangement that no corporation can fully control. Huang’s presence in the East Room was not accidental. It was the product of weeks of coordination between the White House, the Department of Commerce, and NVIDIA’s government affairs team, all trying to manage the optics of American corporate exposure in China without appearing to legitimize Beijing’s technology ambitions.

Four days later, Huang was in New York preparing to receive the Van Fleet Award from the Korean Society, an event attended by Lee Jae-yong of Samsung and Choi Tae-won of SK Group. The sequence matters more than the symbolism. Washington was negotiating rules for AI chip exports to China. New York was where those rules would land on the supply chain. The two cities, separated by one flight and one political horizon, captured the full geometry of the problem: policy is made in one room, and consequence is absorbed in another.

What HBM Has to Do With Geopolitics

NVIDIA’s AI accelerators cannot function without high-bandwidth memory. Samsung and SK Hynix are the two companies that make most of it. This is not a relationship of convenience — it is structural. Every generation of NVIDIA GPU pulls more HBM from Korean fabs. Every US restriction on what China can buy changes the calculus for how much HBM those chips can move, and therefore how much revenue flows into Gyeonggi Province rather than Silicon Valley or Shenzhen.

The math is unforgiving. Chinese data centers have been buying modified NVIDIA chips — the A800, H800, and their successors — because domestic alternatives remain years behind. When the US tightens the nozzle on what qualifies as “advanced,” NVIDIA’s engineering team redesigns. Korean HBM teams redesign with it. The supply chain does not get to pick which side of the border its revenue comes from. More critically, it does not get to pause while policymakers argue about red lines.

What has emerged is a shadow architecture of compliance. NVIDIA engineers calculate the performance thresholds set by Washington and trim products to fit. Korean memory teams adjust yield targets and packaging configurations to accommodate chips that may ultimately be destined for San Francisco or Shanghai. The result is a feedback loop in which geopolitical decisions cascade through engineering specs, production schedules, and quarterly guidance. No single actor in that loop designed the system. All of them live inside it.

The Real Meeting Is in New York

The Chosunbiz report notes that the specifics of Huang’s conversations with Lee and Choi are not confirmed. That uncertainty is the story. These are meetings that happen in fragments — a dinner in Woodside, California in July with Choi, a handshake in San Francisco with Lee — layered over months of parallel negotiations. The Van Fleet gala is just the latest public node in a network that has no official agenda but enormous commercial weight. Public appearances serve as calibration markers, signals to investors, regulators, and competitors about where relationships stand without revealing where they are headed.

Choi and Huang have already discussed AI infrastructure, semiconductor cooperation, and physical AI — a term that encompasses robotics and autonomous systems that will depend on the same memory and compute stack. Lee and Huang met in San Francisco last July with President Lee Jae-myung present, a meeting that carried both commercial and diplomatic significance given the South Korean government’s own stance on export controls. None of these were casual. They were calibration sessions, each one gathering information about what the other side could tolerate and what it needed.

The unspoken question in all of these encounters is whether the Korean memory industry can preserve its access to the Chinese market without triggering US enforcement action. The answer, so far, appears to be that they are buying themselves a narrow corridor by aligning publicly with American policy while quietly ensuring that their product portfolio remains useful to Chinese buyers. It is a strategy that requires constant adjustment and generates constant risk.

Second-Order Effects: The Quiet Realignment

The immediate consequences of this sequence — dinner with Xi, then meetings with Korean chipmakers — are visible in stock prices and earnings calls. The deeper consequences are harder to track but more consequential. They include a gradual realignment of how semiconductor supply chains map onto geopolitical alliances.

For one, the Korean companies are internalizing a lesson that took China and the United States years to learn: memory is a strategic commodity regardless of whether you call it one. HBM is not a general-purpose product. It is a bottleneck with a near-monopoly provider base, and that concentration gives it leverage beyond what normal market dynamics would confer. Samsung and SK Hynix are responding by investing aggressively in next-generation HBM production, not solely to capture NVIDIA’s growing demand but to cement their position as不可替代 suppliers that neither Washington nor Beijing can afford to lose.

There is also a subtle but significant shift in how Chinese companies are hedging. Domestic memory producers, led by ChangXin Memory Technologies and Yangtze Memory Electronics, are accelerating development efforts that are still years from competing with HBM4 or HBM4e. The incentive structure is clear: every quarter of delay increases the likelihood that US restrictions will tighten further, making self-reliance not just a national security imperative but a commercial necessity. The Korean firms understand this dynamic because their customers are partly the companies they are competing against in the Chinese market and partly the companies Washington is trying to contain.

A third effect, less discussed but increasingly visible, is the emergence of a compliance ecosystem. Law firms, consulting groups, and former government officials now occupy a niche role that did not exist five years ago — helping semiconductor companies navigate the gap between what they want to sell, what US law permits, and what Chinese buyers will pay for. This ecosystem has its own incentives. It benefits from ambiguity, from the space between the letter of the law and the reality of trade. The meetings between Huang, Lee, and Choi are, in part, exercises in reading that space together.

Who Wins, Who Loses

The winners so far are clear: NVIDIA secures HBM supply regardless of where policy drifts. Samsung and SK Hynix keep selling into the world’s fastest-growing AI hardware cycle. Chinese AI companies, for now, still find pathways to something — modified chips, older generations, or domestic substitutes that are adequate if not competitive.

The losers are invisible in this framing. They are the policy architects in Washington who designed export controls assuming chip demand is infinitely divisible along political lines. It isn’t. Every restriction creates a parallel market. Every loophole creates a renegotiation. The Korean chaebols understand this because they live inside it, and their understanding is shaping how they prepare for the next round of restrictions, which will almost certainly come.

What is less visible is the third party in every one of these conversations: the United States government. Huang did not choose to sit across from Xi. He was invited as part of a delegation of American industry leaders whose China exposure the White House needed to manage, not hide. His subsequent meetings in New York are equally managed — not secret, precisely, but deliberately unannounced in scope and outcome. The government benefits from this arrangement because corporate engagement with China can continue at a level that does not trigger escalation, while the companies benefit because they retain access to markets that policy has not fully closed.

But this arrangement carries its own risk. If Washington perceives that corporate engagement is undermining the pressure campaign against Beijing, the legal framework can tighten abruptly. The Semiconductor Export Control Guidelines of 2024 already demonstrated how quickly acceptable boundaries can shift. There is no guarantee that the informal understandings being tested in private dinners and award ceremonies will survive a change in administration or a geopolitical incident.

The Timeline Ahead

HBM production ramps are measured in quarters, not speeches. Samsung and SK Hynix are investing billions in new lines that will come online over the next two years. Those investments assume demand from NVIDIA’s next-generation GPUs and continued, if modified, Chinese purchases. If Washington closes the modified-chip door entirely, the Korean memory business takes a hit. If Beijing accelerates domestic alternatives, the same thing happens on a slower clock.

The next inflection point will be visible in NVIDIA’s quarterly earnings calls and in any new US licensing actions targeting Chinese AI chip buyers. Those are the events that will determine whether the Lee-Huang-Choi triangle holds or fractures under policy pressure. Investors are already pricing in scenarios where HBM demand outpaces supply, but few are pricing in scenarios where supply and demand are both constrained by policy decisions made in rooms none of these CEOs actually control.

The Clearer Close

The image of Huang smiling across a dinner table from Xi will be the one that circulates. It is also the one that distracts from the actual mechanism of power. The rules for AI chip competition are being written in Washington and Beijing. But the factories that make those chips possible are in South Korea, and the contracts that fund them are negotiated in rooms like the one Huang is walking into on September 28.

Korean semiconductor companies are not pawns in this story. They are infrastructure. And infrastructure does not pick sides — it gets used by everyone, then asks who pays for maintenance.

The question hanging over New York is whether Samsung and SK will get a seat at the table when the next round of US-China chip negotiations reshapes their market, or whether they will continue to absorb the consequences of decisions made without them. The answer to that question will determine not just the trajectory of two corporations but the architecture of the global AI supply chain for the next decade. If they are brought in as stakeholders rather than casualties, the system may prove surprisingly resilient. If they are left to adjust alone, the next restriction cycle will look very different from the last — and the companies that designed it assumed they would never have to face the consequences.