Jensen Huang Returns to Taiwan With a Trillion-Dollar Vision

Taiwan is two-thirds mountains yet produces more than 90% of the world's most advanced semiconductors. Every computer, smartphone, GPU, and AI data center depends on this one island's manufacturing capacity, a concentration that is virtually impossible to replicate anywhere else in the world.

Huang compares Taiwan's dominance to ancient mysteries—Göbekli Tepe, Egyptian pyramids—in terms of how it was built. The technological and geopolitical implications ripple across decades and centuries, making this single island a fulcrum for global AI competition.

Lex advocates a model where the US and Taiwan compete while maintaining their alliance. Understanding each other's culture and history, along with respect for both, are essential; so too is working together across interlinked economies, technologies, and supply chains.

He emphasizes that human-to-human connections—talking, marrying, eating together—anchor the relationship. These ties make division harder and give people skin in the game on both sides.

NVIDIA's new server chip successor to Grace Blackwell achieves full liquid cooling and radically faster assembly. The previous generation was only 80% liquid cooled; the new version is 100% liquid cooled.

Assembly time has collapsed from two hours to five minutes. Jensen notes that things easy to build are also reliable, linking manufacturing speed to product quality.

The Vera CPU marks a fundamental shift in NVIDIA's approach to processors. x86 hyperscale CPUs were designed to be rented, optimizing for maximum cores because the old economics measured cost per core. AI changes this entirely: it does not rent cores. Instead, it generates tokens and sells them.

The new purpose is to complete the exchange loop—inference—as fast as possible. More computing power purchased means more tokens generated and more revenue captured. The Vera CPU shows 1.8x performance over x86 architectures, reflecting this new optimization target.

This economic inversion reshapes how semiconductor companies think about product design. Every component must serve token generation speed, not core density.

Jensen is always nervous before public speaking. He doesn't take it for granted and wants to do a good job, but everything comes together at the last moment, making it hard to practice. No teleprompter is used during his keynotes.

The stakes are enormous: NVIDIA aims to build and sell a few trillion dollars of computing over the next several years, and the entire Taiwan island and ecosystem must be part of this buildout. Every company needs to invest hundreds of billions of dollars, and his words influence whether partners step up.

Gamers are where NVIDIA's GPU business began, and Jensen remains committed to that origin. Some of the most passionate and happiest humans on earth are gamers, and that foundation cannot be abandoned no matter how high the company climbs.

Lex observed consistent humility from engineers in both China and Taiwan. No arrogance emerged from young engineers who are reshaping global technology, a contrast he found striking to patterns he has witnessed elsewhere.

Jensen and Lex discuss what this humility might reflect—perhaps a cultural emphasis on collective contribution, or a recognition that the work is so much larger than any individual.

CoWoS is not a household name in most of the world, yet it became part of a child's vocabulary in Taiwan.

This anecdote reveals what Taiwanese culture considers cool and valuable. Engineering achievement ranks higher than consumer luxury, a hierarchy that reflects decades of building the world's most advanced chip manufacturing ecosystem.

NVIDIA currently has 1,000 people based in Taiwan, spanning supply chain management, systems engineering, operations engineering, fab work, packaging engineering, and test engineering. These are foundational hardware roles that cannot be done remotely.

The company plans to build a new facility that replicates its current structure and will hold 4,000 people, requiring the hiring of another 3,000 employees. This expansion cements NVIDIA's commitment to the island as manufacturing partnerships deepen.

The cook's knife cut him right across the face, and his father ran him to the hospital.

Despite the injury, Jensen maintained fond memories of the night market and loved returning there. The incident never soured his attachment to the place or its culture.

Jensen visited Shanghai-based AI companies like MiniMax and traveled to Hangzhou to see Unitree robotics. Every AI company he encountered reported the same problem: they are compute starved and want more computing power.

Constraints force them to create incredibly efficient architectures. Scarcity and severe restrictions drive architectural innovation that would not emerge from abundance. Despite the limitations, these companies continue to secure funding and operate successfully.

This pattern mirrors a broader principle: necessity, when it does not kill the company, reshapes how engineers think about their problems.

Jensen does not experience the responsibility of millions looking up to him as a burden. He has always been nice to people and wanted them to succeed, even in school, and was happy to share his homework with other students.

He likes seeing colleagues at other companies succeed, not just his own. Forty-three thousand NVIDIA employees know he wants them to win, and he wants each of them to win in return. This creates a reciprocal dynamic where success compounds through goodwill.

The weight does not press down because the foundation beneath it was laid long before—in small acts of kindness that now scale across an entire company and industry.

Lex frames a core strategic question for the US: Does America flourish if its economic and technological flywheel includes China, or does it flourish if China is excluded? This is an open question for America and for humanity.

There is a line that can be crossed, Lex argues, where geopolitical competition between friends becomes enemy competition with potential to destroy both sides' ecosystems. Huang agrees that the focus should be avoiding conflict while maintaining competition.

Competitive relationships have strategies leading toward conflict and strategies avoiding it. Interdependence through shared hardware, software, universities, friendships, and scientific collaboration makes division harder and raises the cost of conflict.

AI stacks are built on shared hardware and software across countries, not isolated national systems. Scientists use each other's universities, conduct joint research, and engage in friendly competition among different companies. The ecosystem fundamentally operates as collaboration despite surface-level competitive dynamics.

Greater interdependence and shared infrastructure make it harder to divide peoples and facilitate conflict. Energy leadership is distributed globally—some countries lead in energy while others excel in semiconductor design, software, or algorithms. No single nation owns the entire stack.

This distribution of capabilities and mutual reliance creates friction against the kind of clean decoupling that geopolitical rivalry might demand. Breaking the AI ecosystem would require breaking the infrastructure everyone uses and the relationships everyone depends on.

The fruit has a texture like a combination of apple, pear, and mango—a surprise to visitors who assume the color indicates immaturity.

Pork floss and pig's feet are both delicious Taiwanese dishes that suffer from English names with no appeal to Western audiences. Jensen and Lex note that Taiwan is not very good at marketing its traditional foods, a gap they contrast with how OpenAI names things clearly and perfectly.

The disconnect between actual taste and perceived value based on naming is acute. A dish can be extraordinary and still fail to find international audiences because the label repels before the food has a chance to persuade.

Jensen finds it fascinating that his brain cannot reconcile the disconnect between what his nose tells him and what his mouth experiences.

Lex notes that it smells really bad despite the good taste. Nature played a trick, presenting a fruit with every sensory signal pointing away from it except the one that matters most once you overcome the first barrier.

In 1996, the fabless semiconductor industry was newly born, with competitors including Cirrus Logic, S3, Weitek, and Western Digital. Most foundries at the time required licensing of technology as a condition of partnership; SGS-Thomson refused NVIDIA's business.

TSMC was a tiny $300 million company with no strings attached—a company sounding smaller than its actual scale, yet bold enough to take risks. Jensen left voicemails at TSMC's US office, found via Yellow Pages, calling multiple times with no response. Morris Chang called back around 6 PM; Jensen's team was playing PlayStation in the background when the call came through.

Chang invited Jensen to Taiwan, marking his first return to the island in 30 years. This single conversation set in motion a three-decade partnership that would define both companies' futures.

TSMC and UMC were literal neighbors, across a street within the same campus. Jensen visited both and considered them equally excellent companies, yet he chose TSMC for reasons beyond capability.

Morris Chang said 'we can't do that' to many of Jensen's requests, signaling a company built on core principles. Unlike a partner willing to do anything, Chang focused conversations on corporate governance, integrity, partnership, and long-term thinking. He promised: 'If we were to work together, I promise you we would jump through hoops for you.'

Thirty years later, TSMC kept that word. The partnership endured not because TSMC was faster or cheaper, but because Chang's integrity meant his promises were real and his refusals carried weight. A partner who says no to bad ideas and yes to good ones, reliably, is worth more than one who agrees to everything.

Every employee and TSMC must know that Jensen wants them to win. This intent shows up in daily partnership, business dealings, and always helping partners become stronger. Mutual desire to succeed creates effort; when both choose to work together, both step up.

Bad behavior—playing partners against each other, taking all profit, extracting value without reciprocity—is 'remembered on the island forever.' The reputation cost of betrayal is permanent, so the only rational long-term strategy is to build trust and demonstrate that your success depends on theirs.