Jensen Huang says NVIDIA is now an AI Infrastructure Company and Taiwan is at the heart of it

"NVIDIA is no longer just a tech company," declared its flamboyant co-founder and CEO, Jensen Huang, at his Computex 2025 keynote while flanked by the familiar glow of massive LED screens and in his usual leather jacket. It wasn’t just keynote rhetoric. Jensen wanted everyone to understand that NVIDIA had evolved into something else entirely: an “AI infrastructure company”. That phrase wasn’t thrown around lightly. Infrastructure is foundational – roads, power grids, the internet. His argument was that AI (and if you don’t know it by now, you will soon), and specifically NVIDIA’s architecture, was becoming as essential as those utilities.

Jensen spent the rest of the keynote driving home what he meant, and why it mattered. At the core of this grand vision is CUDA-X Everywhere. For most, CUDA has been synonymous with GPU acceleration, a way to make graphics and computations faster. But his ambition for CUDA-X is bigger than just GPU boosts. He sees it as the backbone of AI computing. Libraries built around CUDA are extending across every major computing domain: quantum computing, genomics, digital twins, even 6G networks. Jensen’s message was clear: if AI is going to power everything from robots to self-driving cars, he wants CUDA to run the show.

The NVIDIA CEO’s vision for CUDA-X is almost missionary in its ambition. First widely talked about in 2019, CUDA-X has been an almost inseparable and necessary component to accelerate AI and data science, as reported by my editor-in-chief. Jensen described it as the keystone for what he calls AI factories, where data centres are purpose-built to train, deploy, and optimise AI models at a scale previously unimaginable. It’s about reach as much as it’s about speed. CUDA-X Everywhere is exactly what it sounds like: a push to make CUDA an omnipresent layer of the AI stack, deeply integrated into every conceivable application that demands accelerated computing. From industrial automation to real-time language processing, CUDA-X is Jensen’s bet that NVIDIA can dominate the AI era the way Intel once did for CPUs.

But this global vision zoomed into something more tangible when Jensen announced NVIDIA's plans to build “AI for Taiwan”. In collaboration with Foxconn, TSMC, and the Taiwanese government, NVIDIA will be building a massive AI factory supercomputer right on the island. It’s hard not to think of this as part of a calculated step to turn Taiwan into the nexus of AI development in Asia. It’s a natural extension of what I wrote about recently in my Computex 2025 global press conference coverage, where I talked about how Taiwan is more than just a hub for chip manufacturing; it's the very foundation of our digital lives – a point I wrote about recently in my Computex 2025 blog. Jensen was quick to highlight Taiwan's unparalleled manufacturing prowess, citing its dominance in semiconductor production as a key driver behind the decision. To him, this AI supercomputer isn't just about local innovation; it’s a calculated move to solidify Taiwan’s standing as a key player in global tech infrastructure.

In Jensen’s own words, the AI supercomputer project is designed to be "an AI engine for the world." This isn’t just about leveraging Taiwan’s existing capabilities as much as it’s about transforming it into the heartbeat of global AI development. He was explicit that this is a strategic alignment to ensure that AI growth is localised in regions where chip manufacturing already thrives. "We’re building AI where AI is built", he said, and driving home the idea that innovation should sprout from the places best equipped to handle it.

If Taiwan is the proving ground, then NVLink Fusion is the toolkit. One of the keynote’s most practical announcements, NVLink Fusion is NVIDIA’s way of letting companies build their own AI infrastructure with NVIDIA’s tech as the backbone. It’s modular, it’s scalable, and crucially, it’s open – shattering its traditionally walled garden approach to AI infrastructure. Jensen described it almost like building blocks, where you can slot in NVIDIA GPUs, CPUs, your own ASICs, and have them all work together seamlessly. This pivot is timely. With Microsoft and Amazon ramping up their in-house chip design efforts, NVIDIA's dominance in the data centre market is facing real competition. Huang might prefer if you bought the entire NVIDIA stack, but he's happy enough if you’re buying something – anything – from NVIDIA.

Then there was Physical AI, a concept that almost stole the show with its sci-fi implications. We’ve all heard of digital twins – virtual replicas of real-world objects. But NVIDIA is pushing this concept further by training robots entirely in simulation. Using their Omniverse platform, these machines can be taught how to move, interact, and even anticipate real-world variables without ever touching physical space. Jensen showed off digital twins of factories, autonomous vehicles, and robotic arms, all learning in controlled environments before deployment. The implications are massive. Manufacturing lines could be tested and optimised without downtime. Self-driving cars could run millions of miles virtually before ever hitting the road. Omniverse as a simulation ground is a critical part of this puzzle. NVIDIA’s pitch was that Omniverse could become the de facto platform for developing not just virtual worlds, but practical, real-world-ready AI models. It’s NVIDIA saying, "Why learn in the real world when you can master it virtually first?"

But perhaps the most eyebrow-raising announcement came at the end of his keynote: NVIDIA Constellation. It’s not just another data centre. It’s an AI hub, set to rise in Taiwan as part of NVIDIA’s grand vision to make the island a global epicentre for AI development. Jensen painted it as more than just a corporate outpost for its Taiwan headquarters. It’s also intended to be a research and development powerhouse, a place where the next wave of AI technology will be built, tested, and shipped out to the world. If Taiwan is becoming an AI nerve centre, Constellation will likely be its beating heart. Jensen called it a "commitment to building AI infrastructure where it matters most”.

Jensen then wrapped up his keynote with the kind of confident optimism you’d expect from someone who genuinely believes his company is reshaping the world. And maybe he’s right. The pieces are all there: CUDA-X providing the computational backbone, NVLink Fusion making it customisable, Physical AI pushing the boundaries of robotics, and Taiwan as the proving ground. The only thing left is to see if the rest of the world catches up to his vision. But if NVIDIA has its way, catching up won’t be enough – you’ll have to be running alongside them.

Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.