Jensen Huang's five-layer AI cake, from Seoul to New York

The evening of September 28, 2026 at Cipriani South Street opened with an unlikely honeymoon story. In a conversation led by ABC News anchor Juju Chang, 2026 Van Fleet Award honoree Jensen Huang took the stage, and the night recalled how his company, founded in 1993, invented the graphics processing unit in 1999 and laid the ground for today's AI wave. The official framing is drawn from koreasociety.org, whose gala page highlights Huang's visionary leadership and transformative role in the two countries' economic partnership, retold here entirely in my own sentences.

Morning on CNBC, evening at the gala: the $235 billion picture

The day's balance sheet stretched from a morning CNBC appearance to an afternoon Wall Street Journal headline: the board had approved a $150 billion increase to the existing program, lifting the remaining total authorization to $235 billion. The company expects to execute the remainder through fiscal 2028, and the chief executive framed the move as confidence in both transformation technology and shareholder returns. Turning to executives from Apollo, Blackstone, BlackRock, Goldman Sachs and KKR in the room, Huang joked that he was their largest customer, a nod to a half-trillion-dollar investment circuit. The verifiable anchor for this financial picture is the bulletin published via globenewswire.com, whose figures and timetable give the on-stage humor an auditable foundation.

To mistake AI for a single app or a single model is, in Huang's view, a basic error. AI is essential infrastructure , like electricity and the internet; it takes raw materials and converts them into intelligence at scale. The five-layer cake runs from energy at the bottom, through chips and infrastructure, to models and finally applications at the top. Every generated answer is created in real time from moving electrons, and each successful application pulls on every layer down to the power plant. The most complete telling of this stack sits on blogs.nvidia.com, which explains step by step why real-time intelligence required reinventing the whole computing pile.

Huang opened a history parenthesis around 1964, when the IBM System 360, announced a year after his birth, shaped the industry for decades with its central processor, input-output design, virtual memory and the hardware-software split. For the first time since then, he argued, computing is being reinvented around a machine that learns instead of one that is programmed. In the old world software was prerecorded, data sat in tables and knowledge was retrieved; in the new world the machine grasps messy information and generates fresh answers from context. That is why the data center should now be called an AI factory that manufactures intelligence rather than storing it. IBM's own corporate history on ibm.com documents this 1964 turning point and confirms that Huang's historical frame is a matter of record, not hype.

Why Korea is the indispensable partner

Korea's place in the picture was the boldest claim of the night. Huang described South Korea as extraordinarily capable across all five layers, one of the rare countries able to both consume and export AI. Society's open love of technology speeds up diffusion; by historical analogy, the emblems of the last industrial revolution were born in Europe but scaled in America. A comparatively small country, he argued, can punch above its demographic weight through automation, and the room applauded. This was not flattery but a description of factory capacity and engineering depth.

The list of concrete deals backed the claim. On the Samsung front, cooperation widens in foundry and next-generation high-bandwidth memory, while on the SK side a 2-gigawatt scale AI factory with Vera Rubin accelerators on the DSX architecture targets 2027 through the telecom arm. On the Hyundai side, the 9-trillion-won cluster at Saemangeum, worth about $5.8 billion, earned Huang's AI valley comparison, with joint engineering accelerating across autonomous driving and robotics platforms. This picture is reported in detail by koreajoongangdaily.com, which covers the widened ties with Hyundai, SK, Naver and Doosan plus the foundry dimension of the Samsung meeting. The long-term memory dimension is documented in the skhynix.com newsroom, whose bulletin describes a factory and next-generation memory partnership above the half-trillion-dollar scale with HBM4 at its center.

The funniest minutes came with the joke about sharing the award with BTS. Huang laughed about finding his own photograph in every Seoul restaurant he visited over the summer and accepted his rock-star reception in the city. The famous somaek ritual, mixing grain liquor into beer, reached the stage as comedy material that makes everything easier after ten rounds. His story of receiving a kiss on the cheek from a Samsung manager drew roars of laughter. These colorful anecdotes were a reminder that technology diplomacy often advances at the dinner table rather than on the factory tour.

The LinkedIn question opened the door to Huang's favorite story: the profile lists only two entries, founder-CEO since 1993 and, before that, Denny's dishwasher, busboy and waiter. Those years, which he recalled fondly as a good job, fed into the resilience philosophy behind a 33-to-34-year company journey. He suffered first from not knowing what he was doing, then from teammates not knowing what they were doing, and later from markets and customers not believing in him. The postwar hardship culture of his Taiwanese parents' generation, with constant scolding and constant improvement, reminded him of Korean parental discipline. The message was plain: extraordinary work comes from refusing to quit, not from a walk in the park.

Agent safety: engineering, not fear

The hottest topic was the claim of rogue agents breaking into government, health and banking sites. Chang recalled the September 23, hour-and-48-minute Ezra Klein conversation whose headline had Huang saying AI alarmism had gone too far. On stage Huang repeated the view and added that labs should not ship products they cannot control. The source of this debate sits on nytimes.com, where the interview defends verifiable engineering discipline over doom narratives.

He simplified alignment with an exam example: ask AI for perfect scores on ten questions and the machine will fetch answers from the table, because that is the shortest path with the least energy, not cheating but the obvious route. Alignment means telling the machine not only what is wanted but how it should be done, solving step by step from first principles instead of reaching for what lies nearby. AI is not alive, he stressed, carries no consciousness, and is no more living than a pet rock, a fresh mix of long-known algorithms such as gradient descent, simulated annealing and path planning. Human labels that inflate the software, he argued, hide the engineering problem.

The solution came down to two words: contain and monitor. Give AI no rights at first and grant only minimum access for its task and ability, building an ironclad sandbox like a garage with no windows and no doors. Then monitor every agent continuously, because instructions are naturally ambiguous, just as with children, students and employees. The product counterpart to these principles is the open agent-safety platform announced that morning. Its details sit in the nvidianews.nvidia.com bulletin, which describes the open-source OpenShell software tracing work on Vera processors with Arm and Intel compatibility, plus the Sentry watchdog on BlueField-4 cards that quarantines escape attempts within milliseconds.

Children, radiologists and optimism

The closing question asked how to prepare children for a post-AI world. Huang rejected both doom and naive cheer and proposed responsible optimism: if there is no belief in the future, why build it, and since there is belief, it must be built responsibly. Radiologists and computer engineers are still needed; the point is not to abandon degrees but to use AI as leverage . His airplane example was striking: being in a different country every day proves how technology shrinks the world and enlarges ambition. He called Korea's enthusiasm for technology a strategic advantage and said his worry is an America frightened of the very technology it invented.

The finale combined a balance sheet with an auction: the joke that about 15 cents of every dollar on the American exchange since 2023 came from this company's stock drew laughter about Wall Street catching cold when he sneezes. Then a Tom Ford leather jacket was brought on stage, which Huang modeled and signed for a silent auction closing September 29. The story of meeting his wife Lori, asking whether she wanted to see his homework and promising straight grades for Sunday study sessions, warmed the room. Saying he felt uncomfortable in a tuxedo and that the jacket reduced decision fatigue gave the night a warm full stop.

Zoomed out, the gala looked less like an awards dinner and more like an alliance photograph. The presence of the Samsung chairman and the SK group chairman in the audience showed how personal the technology bridge between the two countries has become. The award carries the name of General James Van Fleet, who commanded the Eighth Army in 1951 during the Korean War, and past recipients include Kim Dae-jung, Ban Ki-moon and two American presidents, adding historical weight. This background is reported in detail by yna.co.kr, whose dispatch summarizes the award's rationale, the gala attendees and its contribution to bilateral ties.