Jensen Huang Declares AI the New Electricity at COMPUTEX 2025
NVIDIA’s Vision: Factories of the Future Run on Tokens, Not Widgets
At COMPUTEX 2025 in Taipei, NVIDIA CEO Jensen Huang took the stage before a packed audience of 4,000, framing artificial intelligence as the next foundational shift in human progress—akin to electrification or the birth of the internet. “AI isn’t just software,” he asserted. “It’s infrastructure. And data centers? They’re the factories of this revolution.”
“Imagine a factory where raw energy goes in and tokens—the currency of AI’s output—come out. That’s what an AI data center is,” Huang said, sketching a future where computation replaces assembly lines.
The keynote doubled as a roadmap for NVIDIA’s expanding empire. Huang revealed that CUDA-X, the company’s accelerated computing platform, now underpins everything from 6G wireless research to quantum supercomputing hybrids. “Every industry is finding new ways to harness it,” he noted, citing a 300% year-over-year surge in CUDA-powered AI model training.
From Agentic AI to Robotics: The Blackwell Era
Huang outlined three waves of AI evolution: agentic systems (which reason and act), physical AI (which interprets real-world environments), and finally, general-purpose robotics. Each leap, he argued, demands exponentially more power—hence NVIDIA’s Grace Blackwell NVL72 systems, which he called “the engines of this transition.”
Collaboration took center stage as Huang announced a partnership with Foxconn and Taiwan’s government to construct an AI factory supercomputer using Blackwell architecture. The system, designed for local researchers and manufacturers like TSMC, aims to turbocharge Taiwan’s position in the AI supply chain. “This isn’t just about chips,” Huang emphasized. “It’s about sovereignty in the age of AI.”
“Taiwan’s role in AI is irreplaceable. The opportunity here is extraordinary—once in a lifetime,” he added, moments before unveiling NVIDIA’s new Taipei office, dubbed Constellation.
Breaking Bottlenecks with NVLink Fusion
For hyperscalers wrestling with AI scalability, Huang introduced NVLink Fusion, a reimagined architecture to streamline data flow between GPUs and CPUs. “Think of it as a neural network for your data center,” he said, claiming it could reduce latency by 40% in massive language model training.
As the speech closed, Huang returned to his grand thesis: AI will redefine value itself. “Tokens might soon be worth more than the energy used to create them,” he mused. The crowd erupted—a fitting end for a talk that framed Taipei not just as a tech hub, but as the forge of a new industrial age.
