The CES 2026 show, now underway in Las Vegas, has drawn the attention of the global tech industry.

As the industry’s annual marquee event, this year’s show has surfaced one trend that stunned just about everyone.

The first to set the tone was Huang, who was willing to bet NVIDIA’s trillion-dollar market value on physical AI. A look at the CES exhibitor list makes one thing clear: it is not just NVIDIA making that wager. Boston Dynamics, Unitree, AgiBot, Samsung, LG, Qualcomm, AMD and nearly every big name you can think of have pushed their chips in the same direction.

After a day at CES, one Chinese reporter was left speechless: “I now truly understand that one day in heaven is one year on earth. The world is about to be overturned.” “The machine age is really here. How can you think it has nothing to do with you? I’m advising every young person to get in, even if it starts with being a robot salesperson, just to claim a spot in this industry.”

Technology bloggers livestreaming from the show floor were stunned and went into full cheer mode.

Yes, the singularity has arrived.

01 Huang Sets Off the Show

Jan. 7, Las Vegas.

Standing on stage in his signature leather jacket, Huang made a statement that sent the tech world into a frenzy: “The ChatGPT moment for robots has arrived.”

At that moment, he laid out the full scope of his AI ambitions, and the scale of them was enough to send a chill through the room.

He has long argued that the next major AI wave will be physical AI.

Today’s AI is smart, but it is trapped on a screen. It needs a body to enter the real world.

That is physical AI.

AI has already learned nearly all of humanity’s written knowledge, or what you might call “reading thousands of books.”

But future robots will need interaction data from the physical world.

That means learning what it feels like to fall, how hard to grip an egg, or how much friction it takes to twist open a bottle cap. In other words, training a robot foundation model requires “traveling ten thousand miles.”

The problem is that training robots in the real world is too slow, too expensive, and too dangerous. Huang has zeroed in on that industry pain point.

He has been laying out his strategy piece by piece.

First, the brain. Huang used NVIDIA’s own massive compute to train a general-purpose robot brain: Project GR00T. Going forward, companies will not need to train robots from scratch. They can start with this “prebuilt brain” and fine-tune it with their own data.

Second, the heart. Huang then designed a tiny heart for robots: Thor. Today’s robots are split in two. The robot foundation model, or “brain,” has to live in the cloud because it is too large. The motion control layer, or “little brain,” sits on the device so it can react quickly. But that creates a fatal problem: latency.

If the network slows down, the robot falls over. So Huang built a super chip that forces the complex foundation model, which used to run in the cloud, into the robot’s chest. The result is a robot with “offline intelligence.” Even unplugged from the internet, it can keep moving.

Third: the training ground.

Huang then built a massive training arena: Omniverse.

Inside that virtual world, time can run 1,000 times faster.

Virtual robots can practice walking, carrying boxes and serving plates.

They fall? No problem. Reset the data. No machine repairs, and trial-and-error costs drop to zero.

Then, on Jan. 7 at CES, Huang unveiled the Rubin superchip platform, and his full map finally came into view.

To simulate physical-world environments, you need to move an enormous amount of data. Rubin was built for that virtual training ground.

Inference costs are cut to one-tenth of the original, training speed is up four times, and what used to cost 2 million yuan now costs just 200,000 yuan.

Only now does Huang’s ambition fully show itself: all robot makers can start with his robot brain, use the heart he has already built for them, and come train in his arena.

But once you start, you will not be able to break away from NVIDIA.

His ambition is to build the lowest-level infrastructure for the future robot world. From the moment a robot opens its eyes to sense, thinks through a model, or moves through compute, it will be paying NVIDIA.

This is no longer about selling shovels. It is about building the highway from robots to the physical world, with NVIDIA running the toll booths.

It is a brilliant move, and a trap that robot makers will have a hard time escaping.

02 Huang’s Edge

Huang and Musk are both betting on physical AI.

But I think Huang is operating on a higher level, and by several dimensions.

Musk wants to be the “Steve Jobs” of the physical AI era. He wants to build an Apple-like product at the extreme end of quality, which is why he has poured everything into the Optimus robot. His competitors are Boston Dynamics, Unitree, AgiBot and every other robot company.

Huang, by contrast, wants to be the “god” of the physical AI era. He is defining the laws of physics.

He does not build robots. He only supplies the brain, the heart and the training room.

He does not care whether Unitree wins or Optimus wins.

As long as there is a fight, everyone has to buy bullets from him.

What makes Huang so effective is that he sits much further upstream in the supply chain, and he does it with remarkable subtlety.

03 CES Keywords

When one company does physical AI, that is an experiment. When NVIDIA, Google, Amazon, Tesla, Hyundai and Samsung are all doing physical AI, that means the direction has been set.

We have seen Huang’s ambition. Now look at Qualcomm’s more understated move.

Huang’s Thor chip is powerful, but it also draws a lot of power, making it better suited to large humanoid robots.

Qualcomm, meanwhile, rolled out a small chip built for low power consumption and 5G connectivity.

Its bet is that the robots of the future will not all be giant bipeds. More of them will be medium and small, battery-powered, lightweight machines.

Think delivery carts, home robots and drones.

Google’s biggest highlight at CES this year was showing how robots can understand complex home environments.

If you point at a pile of clutter on the table and say, “I’m hungry,” the robot can identify the apple in the mess and hand it to you, rather than handing you the remote control.

The main upgrade to Google’s model is aimed at brute-forcing the key bottleneck in physical AI: understanding.

Appliance giants were not sitting still either.

LG unveiled a robot called CLOi, with a humanlike upper body and wheels on the lower half, capable of folding clothes, carrying plates and taking bowls out of a dishwasher.

It is a little slow, but the ambition is clear. LG does not just want to sell washing machines; it wants to sell you a “person” who can operate one for you.

John Deere showed a fully autonomous weeding robot at CES.

It does not need a human operator. It can identify crops from weeds on its own and spray pesticides with precision.

This is not a concept. It is physical AI already running across millions of acres of farmland. Even Caterpillar, which makes excavators, showed a remote-controlled unmanned excavator.

A worker in the Las Vegas convention hall operated the machine to dig at a mine 2,000 kilometers away.

Industry insiders shouted: “The essence of a factory is a giant robot!”

If American giants are competing on the “brain,” then China’s camp is competing on the “body.”

On the CES floor, Chinese robots were everywhere, and they drew some of the biggest crowds.

“A lot of new robot players from China are here, many of them from automation, and they have basically turned themselves into robot companies,” one foreign media outlet wrote. Pricing, too, has become brutally competitive.

Small quadrupeds or tabletop robots are only 6,000 to 10,000 yuan.

Small humanoids or lightweight service robots go for just 8,000 to 30,000 yuan. Full-size humanoids have even been pushed to under 100,000 yuan.

04 Media Frenzy

“When the smartest people in the world are all charging in the same direction, you cannot, and should not, ignore it,” one overseas reporter on the ground said.

A broad consensus emerged across mainstream media: the biggest turning point for AI is no longer bigger models, but moving out of the cloud and into the real world.

After looking at NVIDIA’s ambition, Reuters called it a marker of AI entering the “physical AI” stage. Business Insider kept repeating the same message: “Stop worrying about software demos. AI devices and robots are the real core of the future.”

The Verge and PBS argued that the real breakout at this CES was the deep integration of AI with chips, robots and autonomous systems, signaling that AI is moving from “systems that talk” to “systems that act.”

Nearly all the media treated this CES as the first year of physical AI and praised it as a singularity moment.

Why this year?

Because three singularities happened to collide at once: today’s foundation models finally let robots understand human language, which is the awakening of the brain.

China’s supply chain has made robot bodies cheap, which is the awakening of the body.

And Huang’s virtual training ground has removed the data bottleneck, which is the awakening of data.

The singularity is here. Are you ready?