Jen-Hsun Huang responds to Google TPU competition in Taiwan: Nvidia GPUs have higher versatility, and Nvidia is definitely not afraid of falling behind in the market.

Jen-Hsun Huang is visiting Taiwan for the fifth time this year, relying on the “fungibility” theory to maintain GPU leadership and striving for a 50% expansion in TSMC's 3nm production capacity to counter the challenges of Google TPU and the supply chain pressures from the Trump administration. (Background: Nvidia's Q3 financial report shatters AI bubble theories, Jen-Hsun Huang: Blackwell demand is off the charts, Nvidia's after-hours trading jumped 5%) (Supplementary background: Jen-Hsun Huang shares a fried chicken dinner with Samsung and Hyundai leaders, igniting a frenzy in Korea's “chicken meme stocks”) The humid and warm night in Taipei in November saw Nvidia CEO Jen-Hsun Huang's steps noticeably quicken. On November 27, he stepped foot in Taiwan for the fifth time this year. Besides visiting the ailing TSMC founder Morris Chang, the public's focus was not on his classic leather jacket but on the next round of the chip war: under pressure from Google TPU and the US manufacturing policy, how he would maintain Nvidia's core discourse power in AI. He (Morris Chang) is doing well, and his condition is excellent… and I will be leaving soon, I will be going home today. “Fungibility” theory: GPU value Investors have recently been drawn to the efficiency of Google's self-developed ASIC, questioning whether GPUs can still maintain their premium. Jen-Hsun Huang rebutted during an interview in Taipei, reiterating that fungibility is an insurance for enterprise procurement. He pointed out that with AI models updated monthly, ASICs designed for single tasks may quickly become outdated after algorithm innovations; conversely, GPUs are like cash in a data center's pocket, ready to be deployed to the latest models, covering everything from training to inference. This assertion is not just empty talk. According to Spyglass analysis, Nvidia's B200 single-chip computing power can reach 9,000 TFLOPs, exceeding Google TPU's 4,614 TFLOPs. Jen-Hsun Huang emphasized that while performance differences are important, the CUDA ecosystem and software toolchain represent a “full generation” of leadership. In other words, what customers are buying is not just a piece of silicon, but an insurance policy against the unknown. Nvidia GPUs are highly fungible and can operate anywhere in the cloud; Nvidia is the only global system capable of running all AI models… no matter what needs arise, Nvidia can assist, and Nvidia's position is very stable and unique, but we still have to work hard every day. Production capacity victory: TSMC's 3nm monthly capacity to increase by 50%, Nvidia locks in 35,000 units. As demand for AI chips jumps from millions to tens of millions, design superiority is no longer the sole winning hand. Jen-Hsun Huang's dinner with TSMC founder Morris Chang ultimately resulted in more concrete capacity assurances. According to TechNode reports, the 3nm monthly production capacity of TSMC's Fab 18B will increase from 100,000 to 160,000 units, a 50% increase, with 35,000 units specifically reserved for Nvidia, corresponding to a future shipment plan of 20 million AI chips over the next five quarters. For Jen-Hsun Huang, this is not just a contract, but a declaration to the capital market that “without the supply chain, there is no innovation.” He candidly stated in front of the media: “Without TSMC, there is no Nvidia.” To supply chain experts, this statement is a cold reality: the only company in the world capable of expanding both advanced processes and packaging simultaneously is TSMC. Trump pressure test: A hybrid supply chain under the banner of “American manufacturing”. In 2025, Trump returns to the White House, and “American manufacturing” once again becomes politically correct. Nvidia has been promoting that Arizona's Fab 21 has produced the first batch of Blackwell chips, presenting a compliance report to Washington. However, Jen-Hsun Huang also revealed the reality in Taipei: even if wafers are completed in the United States, advanced packaging like CoWoS still relies on Taiwan, and server system assembly cannot do without partners like Foxconn and Wistron. This hybrid model of “wafers made in America, packaging done in Taiwan” may become the best balancing point for Nvidia to navigate international politics over the next four years. From Silicon Valley to Washington and back to Taipei, Jen-Hsun Huang is responding to the same question at every step: Amid the rapid evolution of models and the dual risks of geopolitical tug-of-war, can Nvidia still firmly hold its core in AI? He persuades customers with fungibility arguments and reassures investors with TSMC's capacity commitments. Facing competitors like Google pushing for vertical integration, he bets that the market will still be willing to pay for flexibility. TSMC thus becomes the key market maker in this gamble—without it, any strategy is merely a blueprint on a whiteboard. Related reports: Musk returns to the White House “Handshake and reconcile with Trump,” attending a banquet with Tim Cook, Cristiano Ronaldo, and Jen-Hsun Huang. Jen-Hsun Huang declares that in the AI era, electricians are in high demand, but do Taiwan's blue-collar workers really earn more? Jen-Hsun Huang flashes to Taiwan before the nuclear power referendum, shouting “nuclear energy is better”—is the nuclear power plot of President Lyle coming true? <Jen-Hsun Huang responds to competition from Google TPU: Nvidia GPUs have higher fungibility, and Nvidia is absolutely not afraid of falling behind in the market.> This article was first published in BlockTempo, the most influential blockchain news media.