Inside Jensen Huang's 14-Day Taiwan Blitz: The Supply Chain Is Now the AI Bottleneck

If doubling output for the second half of 2026 wasn’t enough, Huang issued a stark message to suppliers just before departing. On June 5, at Songshan Airport before boarding his private jet for South Korea, he told reporters Nvidia had already “arranged to boost capacity in the second half of this year to next year” but explicitly warned: “The second half of next year is going to be so much bigger than this year and so everybody must be prepared” . Taiwan News reported he directly told suppliers to “prepare for strong growth” extending from H2 2026 into 2027 .

The subtext was clear: the capacity Nvidia has secured is a starting point. He was publicly pressuring the ecosystem to continue expanding or risk becoming the bottleneck.

The Commitment: $150 Billion Annual Spend in Taiwan

The most eye-popping number of the trip was delivered earlier, on May 27, at a groundbreaking ceremony in Taipei for Nvidia’s new headquarters project. Huang announced Nvidia plans to invest around $150 billion a year in Taiwan, a staggering leap from the roughly $10 to $15 billion annually the company was spending just four to five years prior .

"Four years ago we spent close to zero here... now we spend about $150 billion a year," Huang said, according to Reuters and other outlets . The figure represents Nvidia’s total spending through Taiwan's ecosystem—chips, packaging, systems, advanced memory, and the massive new headquarters facility slated for completion in 2030 . To put the sum in perspective, it is larger than the GDP of most European Union member states .

The Framing: 'Taiwan is the Epicenter of the AI Revolution'

In that same May 27 speech, Huang delivered the phrase that would define the visit: “Taiwan is the epicentre of the AI revolution. This is where the chips come, packaging comes. This is where the ecosystem is” . He repeated the theme at nearly every stop, casting the island not merely as a manufacturing hub but as the singular point where the physical internet of AI is assembled.

At the June 2 Computex briefing, Huang made the relationship transactional and personal, stating, “We are now the biggest buyer of any company within the Taiwan ecosystem” . Nvidia, he later added, plans to expand from roughly 1,000 employees in Taiwan to a facility that can house 4,000, signaling a multi-decade commitment to local engineering talent .

The Context: The $1 Trillion Forecast Is the Demand Side

Several reports during the visit referenced a $1 trillion revenue projection. That forecast was not new. Huang originally made it during his GTC 2026 keynote in San Jose, California, on March 16, when he said he expects at least $1 trillion in lifetime sales from the Blackwell and Vera Rubin platforms through 2027 . The Taiwan visit did not announce the $1 trillion target—it showed how Nvidia plans to physically build the hardware to reach it.

The March forecast doubled a previous $500 billion projection made at GTC 2025, reflecting the arrival of what Huang calls "the inference inflection point," where AI models are not just being trained but are actively thinking in real time, consuming enormous compute resources .

The Memory Squeeze and the 'Please Make More' Moment

The trip also surfaced a critical subplot: memory supply. At Computex on June 2, SK Group Chairman Chey Tae-won announced SK hynix would double its total wafer production capacity over the next five years—a reversal from his March stance that adding capacity wasn't planned .

Hours later, Huang walked to the SK hynix booth, picked up a marker, and wrote “Please Make More” directly on an HBM4E wafer on display. It was a public, theatrical request from the world’s largest AI chip designer to its top high-bandwidth memory supplier, highlighting that even with expanded logic chip production, advanced memory remains a gating factor for scaling AI systems .

The Road Ahead: A Busy Second Half

Huang’s trip was bookended by product teases. At a developer gathering in Taipei on May 30, before the main Computex events, he previewed the rest of the year: "The second half of this year is going to be very, very busy with Grace Blackwell, Vera Rubin, and we have a surprise new product that we haven't told anyone about yet" . That surprise was later confirmed as the N1X—a next-generation platform extending Nvidia’s roadmap even further.

The takeaway from Huang’s 14-day tour is that Nvidia’s AI dominance is now a manufacturing and logistics problem as much as a design one. The CEO’s personal presence across multiple events was a direct signal to TSMC, SK hynix, and the broader Taiwanese supply chain that the AI demand curve is still accelerating, and the capacity to meet it must be built now.