Google DeepMind Genie 3 vs NVIDIA Cosmos

Google DeepMind Genie 3 vs NVIDIA Cosmos: Welcome to the World Model Wars

User avatar placeholder
Written by Aarti Dattani

August 13, 2025

Here’s the inside info: Google and NVIDIA have both decided it’s time to start building their own “worlds.” 

And no, these aren’t Sims worlds where you make your neighbour’s house mysteriously disappear-these hyper-realistic AI-powered and physics-centric worlds are meant to train robots, self-driving cars, and perhaps even AI brains that borderline too well for our taste.

Google DeepMind Genie 3 vs NVIDIA Cosmos

First off: Google (a.k.a. DeepMind) is putting the hard yards into “World Models,” an AI sandbox where physics, environments, and characters behave like entities would in the real world. 

It’s all in keeping with the vision toward developing AGI (artificial general intelligence)

Of course, being Google after all, they’re integrating this into other major projects, such as Gemini 2.5, VEO3, and Genie 3 World Model, which means it’s now so much more than just a sandbox.

Then there are the good ol’ boys from NVIDIA, and they are coming through with a shiny new toy called Cosmos

And Cosmos isn’t just one thing; it’s a whole suite of world foundation models with operatives called Cosmos Reason, Cosmos Transfer-2, and Cosmos Predict. 

In simple terms, it can create photorealistic, physics-accurate video simulations, reason about them, and even transfer them into different scenarios.

This is where they are integrated with CARLA and the Omniverse simulation platform, utilizing their absurdly powerful GPUs and cloud systems.

If Google is designing the city in its entirety, NVIDIA comes in with trucks full of bricks and cranes, along with the personnel to go ahead and erect it.

The interesting part is that both are after what one would call physical AI: the robots, the self-driving cars, industrial machines, and so forth. 

In other words, they are working at the same end of teaching AI to ‘live’ in a virtual world before letting them loose into ours. 

So-called rivals?

There appears to be some uncomfortable overlap. Google has the research acumen, NVIDIA has its hardware and developer tools, and each is competing to become the flaky developer paradise. 

It’s like watching Apple and Samsung launch their flagship devices in the same week, with smiles on stage, and daggers back-stage. 

The stakes? Huge. 

Whoever cracks it first shall have the front seat in the development of robotics and AI-driven industries. 

And between you and me, if we start seeing these worlds leak into VR games, I’m dubbing this as the onset of the “World Model Wars.”

The question remains: are these worlds going to be friendly to us…or just a training ground for the machines?

Image placeholder

Aarti Dattani

Aarti Dattani is a content writer by profession with 5 years of experience writing in various niches. She is passionate about AI and loves to learn and get to the depth of it. At AI Gimmick, Aarti aims to bring to you readers a simplified way to understand the difficult technical terms and help you explore new tools and trends.

Leave a Comment