NVIDIA "AI Factories" -- More Than Clever Marketing?

NVIDIA (NVDA) founder and CEO Jensen Huang gave the opening keynote at the annual Computex conference in his home country of Taiwan on Monday.

It was another special moment for a small nation with stunning technological prowess in electronics and semiconductors. He said it was his 30th such visit and he was proud to acknowledge his parents in the audience.

Jensen opened the event describing his strong relationships with Taiwan Semiconductor (TSM), Foxconn Hon Hai, and dozens of other Taiwanese tech companies and partners. And he closed with a dramatic new video showing NVIDIA’s Santa Clara office “launching” into space and landing in Taiwan.

The big reveal was NVIDIA Constellation, a brand new Taiwan office for NVIDIA’s growing Taiwan workforce.

Huang emphasized that the work Taiwanese companies are doing has changed the world. He thanked NVIDIA’s ecosystem partners and described the industry’s opportunity as “extraordinary” and “once in a lifetime.”

“We are in fact creating a whole new industry to support AI factories, AI agents, and robotics, with one architecture,” he said.

Jensen took the stage at a packed Taipei Music Center Monday to kick off COMPUTEX 2025, a week-long event of technology speakers, demos, and workshops.

As usual with his keynotes, NVIDIA kick-started the event with 7 press releases and 11 separate blogs on new innovations, products, and partnerships.

But I want to highlight the one major theme from Jensen's talk that many investors still find hard to grasp: what exactly is an AI factory?

The Path to Improving Humanity's Reach

Since Huang revealed his "AI Factory" roadmap at CES earlier this year, he has clearly laid out the steps from Generative AI, to Agentic AI, to Physical AI.

The simplest way to describe why NVIDIA now talks about "AI Factories" is to understand what supercomputers with large clusters of GPUs used to be called a decade ago: "massively parallel architectures."

I made my first vlog about this technology in March of 2017 titled "Get Your MPA in Deep Learning." Back then, it seemed the innovations couldn't move any faster, even as IBM and Salesforce were bringing Watson and Einstein together using NVIDIA technologies.

This has been one of the key features of GPUs that they could be combined by the thousands to do what they do best: compute simultaneously, or "in parallel," without the constraints and bottlenecks of traditional CPU computing.

As the technology and its speed have accelerated by 1,000x and new use cases across science, engineering, medicine, energy, and commerce are discovered every day, building the “MPAs” faster, for wider availability to more enterprises, has become the bottleneck.

The human appetite for turning data into knowledge -- and then into useful intelligence -- has grown exponentially now that we see what can be done to improve life for more people at a pace we never imagined, from curing cancer and resolving global conflict to solving poverty and pollution.

All of these “intelligence” possibilities are only constrained by our physical systems, the datacenters that can be upgraded by NVIDIA GPU dynamics, to solve all these problems at lightning speed. It's why Oracle (ORCL) CEO Larry Ellison mused last year "We have 162 datacenters... we'd like to have 1,000 to 2,000 more."

The Robots (and their Agents) Are Coming

And this is not a stretch when you imagine the AI infrastructure required to train and operate millions of autonomous vehicles and robots. Jensen believes that robotics will be the next multi-trillion-dollar industry.

When Isaac GR00T was unveiled last year as the next generation platform for training humanoid robots using synthetic data, I was once again in awe of Jensen’s long-term vision. Companies like BMW were already using Omniverse “digital twins” to simulate new factory operations, so it only made sense that robots could be trained on synthetic data to make them safe, reliable, and affordable before they ever get near real humans.

In Taipei, Huang introduced new tools to speed the development of humanoid robots: the Isaac GR00T-Dreams blueprint will help generate synthetic training data using Omniverse and Cosmos platforms to create neural trajectories, enabling robots to adapt to new environments and tasks.

The cross-over here between “agentic AI” and “physical AI” is mind-blowing. Agents are “essentially digital robots,” Huang said, able to “perceive, understand and plan.” To speed up the development of physical robots, the industry needs to train robots in a simulated environment. And NVIDIA has partnered with DeepMind and Disney to build Newton, the world’s most advanced physics training engine for robotics.

AI as Electricity X the Internet

Here's how the master described the situation himself last night...

"AI is now infrastructure, and this infrastructure, just like the internet, just like electricity, needs factories,” Huang said. “These factories are essentially what we build today.”

"They’re not data centers of the past. These AI data centers, if you will, are improperly described. They are, in fact, AI factories. You apply energy to it, and it produces something incredibly valuable, and these things are called tokens.”

I admit, "tokens" sounds sort of boring and soulless. But they are simply the units of data and information that can be combined exponentially to create new knowledge and intelligent solutions. And when enterprises have NVIDIA power tools like the GB200 NVLink72 exascale supercomputer, they will be creating, combining, and transforming tokens at higher bandwidths than the entire Internet operates.

Jensen took some noted pleasure in explaining that math last night when he showed off the NVLink “spine” and said “This moves more traffic than the entire Internet."

NVIDIA also announced DGX Cloud Lepton, a global GPU compute marketplace connecting developers to over 10,000 GPUs via partners like CoreWeave (CRWV), Foxconn, and SoftBank. It supports on-demand and long-term compute needs for agentic and physical AI applications for companies and research labs that need access to the power tools without investing heavy capital in their own infrastructure.

And yes, Jensen has a vested interest in promoting the AI revolution because it sells more GPUs.

But if you think about the world-changing power of AI as just like electricity or the Internet, you realize that if NVIDIA stumbled tomorrow, a thousand other companies would fill the void to capture the potential of what is unfolding in what I call the 5th Industrial Revolution.

Recall that it took only a century to go from widespread electricity to the ubiquitous iPhone.

Now, by harnessing AI factories, we are about to witness centuries of change in only a decade. And that innovation will add trillions of dollars in productivity, solutions, and intelligence to the global economy every year.

Kevin Cook has been intensely following NVIDIA innovations since 2016. He runs the Zacks TAZR Trader portfolio where the AI revolution is always the center of discussion -- when he’s not talking about Bitcoin.

Want the latest recommendations from Zacks Investment Research? Today, you can download 7 Best Stocks for the Next 30 Days. Click to get this free report

This article originally published on Zacks Investment Research (zacks.com).

Zacks Investment Research