Nvidia CEO Jensen Huang Lays Out Taiwan’s AI Future, $150B Spending, and a Critical Power Warning

Vera Rubin Is in Full Production — and Huang Flew to Lock It In

The operational headline of the visit was production certainty. Huang confirmed that Vera Rubin "has entered full production" and that Nvidia has "orders in hand" . He called the platform "the most ambitious endeavor in the history of our company" .

The Vera Rubin supply chain is twice as large as the previous generation Grace Blackwell platform, with HBM4 memory supplied by SK hynix, Samsung Electronics, and Micron . More than 1 million rack components for Vera Rubin are assembled across 25 factory sites in Taiwan alone .

Behind the keynote announcements, the trip's real purpose was a face-to-face meeting with TSMC Chairman C.C. Wei to lock in CoWoS advanced packaging capacity for the Vera Rubin ramp . The production timeline is tight: the Vera Rubin ramp overlaps with the tail end of Grace Blackwell GB300 production, meaning TSMC must sustain one platform's volume while simultaneously spinning up another .

Huang's visit also produced a product surprise: the unveiling of RTX Spark, an Arm-based Windows PC superchip that pushes Nvidia into the consumer computing space it has largely avoided . At the same GTC Taipei keynote, he declared that the AI industry has entered a "full-fledged commercialization phase" .

The Demand Picture: $1 Trillion by 2027

Huang reiterated a demand outlook that has steadily escalated. Earlier in 2026 at GTC San Jose, he projected that AI infrastructure demand could reach at least $1 trillion by 2027, more than doubling his previous $500 billion estimate for the Blackwell and Vera Rubin generation . "The inference inflection point has arrived," he said, describing a structural shift toward inference-driven computing that is pushing capital expenditure beyond previous models .

During his Taiwan departure remarks, Huang said AI is driving a surge in computing demand expected to generate trillions of dollars in value in the coming years . He has separately told suppliers that Nvidia alone could require TSMC to more than double its capacity over the next decade .

The Bottleneck No One Was Talking About: Power

The most striking warning Huang delivered before leaving Taiwan was not about chips, capacity, or geopolitics — it was about electricity.

"Taiwan needs more electrical power to continue capitalizing on AI business opportunities and maintain economic prosperity," Huang said, calling energy "the foundation of industrial growth" . He framed this as a global challenge rather than a problem unique to Taiwan, but the location of the remark was significant: compute demand in Taiwan has "skyrocketed" because that is where the AI factories are being built .

The juxtaposition was blunt. Nvidia will spend roughly $150 billion a year on an island whose power infrastructure Huang publicly flagged as a constraint. For a CEO who spends his visits thanking partners and celebrating production milestones, the energy warning read as a deliberate signal to policymakers.

What the 14-Day Visit Signals

Huang's trip compressed multiple messages into a tight timeline: a production milestone (Vera Rubin), a spending commitment ($150 billion annually), a supply-chain dependency (150 partners, 2 million parts per system), and a structural risk (power). The visit was less a celebration and more a working trip to secure capacity for what Huang called "the largest product launch, probably in the history of Taiwan" .

The departure remarks on June 5 closed the loop. Taiwan is essential, Nvidia is all-in, but without more power, the island's ability to capitalize on the AI buildout has a ceiling. Huang put that ceiling in public view before boarding his flight.