Artificial intelligence did not come out of nowhere; It required unprecedented computing power to train complex models. The GPU manufacturer discovered almost by accident that the architecture designed to render video games was the key piece that the computing world was waiting for
Here we are in the middle of 2026, seeing how Nvidia not only dominates the chip sector, but is rolling out a roadmap to become the standard for what they call “physical AI.”
After CES 2026, it is clear that Jensen Huang’s ambition has no limits. Their goal is ambitious and, at the same time, strangely familiar: they want to be the Android of robotics. But what does this mean for us, for businesses, and for the way we will interact with technology? Let’s break it down calmly.
An ecosystem for robots
When we think of Android, we think of a system that allowed hundreds of phone manufacturers to offer cutting-edge technology without having to reinvent the wheel from scratch. Nvidia seeks exactly the same for robotics.
Until now, creating a robot was a fragmented, expensive and extremely complex process. Nvidia wants to standardize everything: from how the robot is trained to how it moves in the real world.
This vision is supported by what they call the “three-computer” architecture. Imagine this process as the growth of a child:
- The first computer (Cloud/Data Center): This is where the robot “studies”. Massive Nvidia infrastructures are used to train the AI models.
- The second computer (Simulation/Omniverse): This is the “safe playground.” Before touching the real ground, robots practice millions of times in ultra-realistic virtual environments (digital twins) where they cannot break anything or hurt anyone.
- The third computer (Edge/Jetson): It is the brain that the robot has incorporated to make decisions in real time while walking through a factory or helping us at home.
Project GROOT and the era of humanoids
One of the most exciting points of your research is Project GROOT (Generalist Robot 00 Technology). These are not just machines that move boxes, but humanoid robots designed to understand our environment, reason and adapt.
Version N1.6, introduced this year, is a significant leap. Thanks to vision-language-action (VLA) models, these robots no longer just execute rigid orders. They can now interpret what they see, understand context, and perform complex movements, such as holding a delicate object while walking across uneven terrain.
It is the definitive union between the “intellectual” intelligence of models like ChatGPT and the “motor” intelligence necessary to wash a dish or assist in a hospital.
Cosmos: Reasoning reaches the physical world
But how does a robot understand that a glass surface is fragile or that it should not push a door if there is someone behind it? This is where the new Cosmos model suite comes in, another of the big news for 2026.
Nvidia has released models such as Cosmos Reason 2, designed specifically for vision and language processing applied to physics. This allows machines to not only “see” pixels, but to “understand” the consequences of their actions in the physical world.
In addition, tools like Cosmos Transfer 2.5 allow you to generate synthetic data (virtual videos) so that robots continue learning without having to record thousands of hours in the real world, which accelerates development exponentially.
Hardware that fits in the palm of your hand (and is a beast)
Nvidia has also updated its Jetson platform, which is basically the central nervous system that is installed in robots. The new Jetson T4000 module, based on the Blackwell architecture, is capable of performing a huge amount of AI calculations with surprisingly low power consumption.
The best of all is the price. Development kits for engineers and students start at $500. This means that a startup or a group of university students in Spain can begin to prototype their own assistant robot with the same technology that a multinational uses.
A team effort: From Boston Dynamics to Mercedes-Benz
Nvidia knows it can’t do this alone. That is why its partnership strategy is so aggressive. At CES 2026 we have seen how giants like Boston Dynamics are integrating Nvidia technology into their famous Atlas robot.
Mercedes-Benz is also using the Drive AV platform for its autonomous vehicles that will hit US roads later this year.
They have even strengthened ties with Hugging Face, the largest open AI community in the world. This allows millions of developers to have free access to models like Cosmos or GROOT to experiment with no barriers to entry.
The challenges that lie ahead
Not everything is a bed of roses, and it is important that as a society we are aware of the challenges. The energy consumption of training these models is massive and could strain our infrastructures.
Additionally, there is the eternal question about job displacement and the safety of AI.
There is also the competition. Qualcomm has not stood by and has launched its platform Dragonwing to try to stand up to Nvidia’s dominance. This competition is good for us, the users, because it forces us to innovate faster and lower prices.
A future that is already here
Nvidia’s vision for 2026 paints a picture where robots will no longer be dumb machines locked in safety cages in factories.
Thanks to this “open platform” approach, we are about to see an explosion of robot assistants, smarter vehicles, and automation systems that truly “understand” the world around them.
We are experiencing the transition from computing that only happens behind a screen to computing that physically interacts with us. Nvidia wants to be the foundation on which this new world is built. And, judging by what we have seen, they have every chance to achieve it.
This post is also available in: