Fei-Fei Li's World Labs raised $1B from A16Z, Nvidia to advance its world models(bloomberg.com)
bloomberg.com
Fei-Fei Li's World Labs raised $1B from A16Z, Nvidia to advance its world models
https://www.bloomberg.com/news/articles/2026-02-18/ai-pioneer-fei-fei-li-s-startup-world-labs-raises-1-billion
17 comments
Headline suggests A16Z, Nvidia as investors, but the article and their own blog suggests Autodesk, AMD etc. as the bigger investors
This site probably has fans or bots of the egg head submitting and upvoting things.
Clickbait.
The interesting thing to me about their world models is that it's like a static point cloud model, as opposed to deepmind's genie 3 which feels more like a video generation model at its base. I see the video generation base as generally superior but far more expensive. The current approach for world labs is likely based on the expertise of the founders, but I don't see how it can scale and match what genie 3 does even if it does offer state persistence as an advantage.
Yes, Marble (from World Labs) feels like it's generating Gaussian Splats or similar. I guess it's more compatible and easier to use for 3d asset generation and reusing in other software.
Very exciting times ahead!
related:
https://news.ycombinator.com/item?id=45923326
https://entropytown.com/articles/2025-11-13-world-model-lecu...
I still want to know the definition of a world model, because it seems that what they are doing now is a very basic model of GS (not sure if it is self-developed) and a GS browser.
https://news.ycombinator.com/item?id=45923326
https://entropytown.com/articles/2025-11-13-world-model-lecu...
I still want to know the definition of a world model, because it seems that what they are doing now is a very basic model of GS (not sure if it is self-developed) and a GS browser.
World model is a 3d game engine, which uses neural net to compute screen pixels instead of rasterizing triangles with textures or ray-tracing.
It is different from video generators, because it also accepts control inputs (ie. keyboard, mouse).
It is different from video generators, because it also accepts control inputs (ie. keyboard, mouse).
I think a robot would want to know a little more about the world though? For instance, picking up a spatula requires less force than picking up a an anvil. So you would want to know about the mass of things.
What are the industries that would truly benefit from good world models? Obviously there are benefits for Robotics and Self Driving but what else am
I missing?
entertainment: gaming, videos, VR
learning
AGI: DeepMind is trying to build general agents within Genie (+SIMA)
learning
AGI: DeepMind is trying to build general agents within Genie (+SIMA)
depending on what you are using them for they are quite good for planning 3d operations.
However the definition of "world models" is a bit nebulous. World labs looks kinda like sprites in a 3d world.
Which also leads me to suggest that in no way are they worth $1b, espcially as google are doing to destroy them (possibly meta, but I doubt they are going to be any good anytime soon)
However the definition of "world models" is a bit nebulous. World labs looks kinda like sprites in a 3d world.
Which also leads me to suggest that in no way are they worth $1b, espcially as google are doing to destroy them (possibly meta, but I doubt they are going to be any good anytime soon)
Robotics is an enormous TAM, and would be all this company needs to justify this type of valuation.
Also, video games, and movie / tv / video content generation.
Also, video games, and movie / tv / video content generation.
Robotics benefits nearly every industry dependent on robotics, being able to expand your market base to countries where individuals have lower incomes obviously provides a boost to companies that wish to sell more product.
It also allows for companies to reclaim their supply chains within the country of manufacture (for high income countries)
It also allows for companies to reclaim their supply chains within the country of manufacture (for high income countries)
World models are not a new idea, it comes from the "model-free" and "model-based" reinforcement learning paradigms that have been around forever
Model-free paradigms (which we do now without world models) does not actually model what _happens_ when you take an action, they simply model how good or bad an action is. This is highly data inefficient but asymptotically performs much better than model-based RL because you don't have modeling biases.
Model-based RL, where world-models come in, models the transition matrix T(s, a, s') meaning, I'm in state s and I take action a, what is my belief about my new state? By doing this you can do long-term planning, so it's not just useful for robotics and video generation but for reasoning and planning more broadly. It's also highly data efficient, and right now, for robotics, that is absolutely the name of the game.
What you will see is: approximately zero robots, then approximately one crappy robot (once you get performance + reliability to jusssst cross the boundary where you can market it, even at a loss! and people will buy it and put it in their homes). Once that happens you get the magic: data flywheel for robotics, and things start _rapidly_ improving.
Robotics is where it is because it lacks the volume of data we have on the internet. for robotics today it's not only e.g. egocentric video that we need but also _sensor-specific_ and _robot-specific_ data (e.g. robot A has a different build + components than robot B)
Model-free paradigms (which we do now without world models) does not actually model what _happens_ when you take an action, they simply model how good or bad an action is. This is highly data inefficient but asymptotically performs much better than model-based RL because you don't have modeling biases.
Model-based RL, where world-models come in, models the transition matrix T(s, a, s') meaning, I'm in state s and I take action a, what is my belief about my new state? By doing this you can do long-term planning, so it's not just useful for robotics and video generation but for reasoning and planning more broadly. It's also highly data efficient, and right now, for robotics, that is absolutely the name of the game.
What you will see is: approximately zero robots, then approximately one crappy robot (once you get performance + reliability to jusssst cross the boundary where you can market it, even at a loss! and people will buy it and put it in their homes). Once that happens you get the magic: data flywheel for robotics, and things start _rapidly_ improving.
Robotics is where it is because it lacks the volume of data we have on the internet. for robotics today it's not only e.g. egocentric video that we need but also _sensor-specific_ and _robot-specific_ data (e.g. robot A has a different build + components than robot B)
I am not trying to be mean but this does not smell right to me, getting a solution too early for a problem vibes.
NVidia has to spend the money while the music still plays. NVidias money doesn't exist. Nothing has meaning, the value of money depends on who holds it.