Huangzilla in Taipei: The NVIDIA Keynote That Redefines 'Thinking Big'

Huangzilla in Taipei: The NVIDIA Keynote That Redefines 'Thinking Big'

Just when we thought we had seen it all, Jensen Huang shows up and says, 'Hold my leather jacket for a wee bit, will you? I need to unfold my master plan.' What the NVIDIA CEO unleashed at Computex 2025 wasn’t your typical tech talk. It was a manifesto. A tribute to over-the-top scale. And, let’s face it, a parade of hardware and ideas so mind-blowing we could have mistaken them for Interstellar scenes rendered in real time.

This live keynote, presented at the COMPUTEX 2025 tech event in Taipei, was a showcase of NVIDIA’s vision for the future of artificial intelligence, data centres, robotics, and accelerated computing. But it was also a bold declaration: according to Huang, the world is entering the age of 'AI factories', and there’s no turning back.

From Selling Chips to Creating Parallel Universes

NVIDIA is no longer just a tech company. It is, according to Huang, 'an essential infrastructure company.' AI is being positioned as the third industrial revolution, and data centres as the new power plants that, instead of electricity, generate 'tokens.' Tokens, people! As if they were digital oxygen bottles feeding AI agents that are already working right alongside us.

And like every good revolution, Huang rolled out names that sound like Star Wars characters: Grace Blackwell, DGX Spark, CX8... And let’s be honest, the guy pitches it with such passion that we feel like we may want to re-mortgage our flat just to buy a DGX station and render our grocery list in 8K.

AI doesn’t just 'respond' any more

It reasons, and soon will look us in the eye while doing it. Welcome to the era of agentic AI. These are artificial intelligences that understand, think, and act. Huang explained it simply: 'An agentic AI is a digital robot.' Like broken poor Siri, but with a brain, a task planner, and full access to our calendar. The immediate future of our organisations, according to NVIDIA, will involve these digital agents not just helping, but actively collaborating.

Next up: physical AI. Robots with common sense, spatial memory and, hopefully, better manners than some humans. They are currently trained in simulators with physics engines so precise that if you throw a croquette, they'll know whether it bounced off the cat or just landed on the floor.

Omniverse: minus the Marvel Multiverse

Simulated reality is no longer science fiction. Huang introduced the Omniverse as the virtual gym for robots. It’s where the automata we’ll see in future factories or bringing us breakfast (fingers crossed), get trained. Every piece, every circuit, every clumsy robot stroll is simulated before a single screw is tightened in the real world. Efficiency, pixel by pixel.

And of course, all of it is powered by the NVIDIA ecosystem: massive chips, libraries with spell-like names (CUDNN, Tensor RTLM, Dynamo, Cosmos...), and a stream of tokens that would make Bitcoin fans weep.

Hardware Highlights: Superchips, Racks, and Nerd Nirvana

It wouldn’t be an NVIDIA keynote without a hardware mic drop. Jensen came armed:

• Grace Blackwell Superchip: The heavyweight champion of the keynote. A monster fusion of CPU and GPU delivering 40 petaflops per node. That’s more power than 18,000 GPUs from 2018, squeezed into one computing unit. Grace Blackwell is what happens when Moore’s Law chugs three espressos and sprints.

• DGX Spark: The personal AI cloud you didn’t know you needed. One petaflop of power, 128 GB memory, and it fits under your desk. Same performance as a 300-pound DGX1 from 2016, now miniaturised for your (overheated) home office.

• MVLink & MVLink Fusion: NVIDIA’s answer to the question ‘What if your GPU could talk to everything?’ These are ultra-fast interconnects and modular link systems that let third-party chips plug into the NVIDIA ecosystem. It's plug-and-play for megalomaniacs.

• RTX Pro Enterprise Server: A workstation beast. x86 compatible, runs legacy IT, powers Omniverse, and yes, even runs Crysis. Uses new RTX Pro 6000 GPUs and CX8 switch chips for wicked-fast east-west traffic.

• Rack-scale Systems: Think racks cooled by liquid, chewing through 120kW and strung together with 70kg of cables. Each rack is a modular motherboard. Literal compute muscle. If you’re not sweating just by thinking of it, it’s not hot enough.

AI factories. Welcome to the Trillionaire Club

Forget traditional data centres. Now they’re called AI factories. Places where you pump in energy and tokens come out. Not just any tokens, but the kind that power generative models capable of translating anything into anything else. Text to video, video to text, dog to bird. Huang even paints a future where companies will measure productivity in tokens per hour, like bolts or pints of beer.

Naturally, all of this is backed by 120kW racks, liquid cooling, and cables weighing 70 kilos. Because if there’s no risk of a hernia, it’s not bleeding-edge tech.

Nvidia Constellation: from silicon to real estate

A surprise ending? Not really. After revolutionizing half a dozen industries, Huang announced the construction of a new mega-campus in Taipei: Nvidia Constellation. Because no keynote is complete without a Pharaonic building project. And, well, half of NVIDIA’s engineers no longer fit in the old office.

Wrap-Up: between awe and brain overload

If we found it overwhelming, we’re not alone. Huang took us on a journey from the GPU in our laptops to reasoning agents, robots learning from AI-generated dreams, and architectures so modular they make Lego look like the Stone Age.

But amidst the mega-factories and hyperchips, one essential question remains: are we building tools, or future digital deities?

Either way, if AI is going to shape our world, we’d better understand what’s going on. Or at least have a laugh while someone tells us, ‘This isn’t art, it’s simulation.'

Until the next keynote, where they’ll probably announce a fridge already connected to Blackwell that now explains to our brother-in-law how he should set up the table this Christmas.

To learn more: NVIDIA CEO Jensen Huang Keynote at COMPUTEX 2025