OpenAI DESTROYED Sora

OpenAI killed Sora. Here's what's next. So OpenAI just killed Sora, one of the most hyped AI tools of the last two years. Gone.

And the reason why tells you everything about where AI is heading next. This isn't just news about one app getting shut down, this is about a billion dollar pivot that puts OpenAI in a direct race with Google, Meta and some of the biggest names in AI history. And whoever wins that race, they don't just win the AI industry, they win the physical world. Let me break it down.

So here's what happened on OpenAI with their shut down of the Sora app and its API. That was on March the 24th. No warning, just gone. Downloads had dropped 75% from their peak.

Total revenue from Sora was about 2.1 million dollars and this was a company valued at hundreds of billions. That's basically nothing. The internal world and the internal word was that Sora was a drag on resources. The GPUs were, and I'm quoting here, melting under the strain of generating video.

So OpenAI made a call. Stop burning compute on Sora and redirect the entire research team towards something much bigger. That something is called a world model. And before you think this is just another buzzword, stay with me because this shift changes everything about what AI can actually do.

Here's the difference between what Sora was and what a world model is. Sora made videos, beautiful videos. You typed a prompt, it generated footage, but it had no idea what was happening in that footage. Drop a ball in a Sora video and it might look like it bounces, but Sora doesn't know why it bounces.

It doesn't understand gravity. It doesn't know the ball exists five seconds later. It's just predicting pixels. A world model is completely different.

So a world model actually understands how the physical world works. Physics, you've got, for example, cause and effect, spatial relationships, and if you tell a world model turn left, it generates what happens when you turn left in real time, interactively. It maintains a consistent environment where objects stay where you put them. It can plan ahead by stimulating multiple futures internally.

Think of the difference like this. Sora was a really good painter. He could paint you a beautiful picture of a city. A world model is more like a city simulator.

You can walk around in it, make decisions, and the city responds. That's a jump OpenAI's team is now making. And here's what's interesting. The team leading this isn't a new hire.

It's the same people who built Sora. Bill Peebles, the head of Sora, is leading the world model push. He invented the diffusion transformer architecture, the underlying tech, the powers. Not just Sora, but much of a modern image and video generation.

So he's been saying internally that the real prize is, quote, automating the physical economy. Thomas Dimson, who built Instagram's original ranking algorithm, is the engineering lead. You've got Rohan Sahal, who's heading product. These are not junior researchers and they're not starting from zero.

Two years of training video models on the physical world gives you a massive head start on building a world model. The physical and the physics of this is already partially baked in. And this is why the pivot matters. OpenAI didn't lose Sora.

It turned Sora into a foundation for something more powerful. Now, who else is in this race? Because OpenAI is not alone. Not even close.

Google DeepMind released Genie 3 in August 2025. Right now it's the most technically impressive real-time interactive world model anyone has ever shipped. It generates navigatable 3D environments at 720p and 24 frames per second from a text prompt. You can explore them.

You can walk around. You can interact with objects. It maintains visual consistency for several minutes. Research scientist Jack Parker Holder said publicly that they think models are key on the path to AGI.

That's Google saying this out loud. And then there's Jan Lekum. If you don't know who he is, he's a Turing Award winner. Spent 12 years at Meta and is considered one of the three godfathers of modern AI.

He left Meta in December 2025 and started a new company called AMI Labs in Paris. Then he raised 1 billion and 30 million dollars in March 2026. His investors include Jeff Bezos, Eric Schmidt, NVIDIA, Samsung, and Toyota. His thesis is blunt.

He has said the path to super intelligence through large language models is, and I'm quoting directly, complete BS. He believes world models built on a completely different architecture are the only viable path to true AI. His company has no product yet. Possibly won't for years.

But a billion dollars says people believe him. Then there's Fei-Fei Li, the woman who is often called the godmother of AI. She founded World Labs, launched a product called Marble in November 2025, the first commercial world product anyone has ever shipped, and raised 1 billion dollars in February 2026 at a 5 billion dollar valuation. Autodesk put in 200 million dollars.

NVIDIA is an investor. Marble lets you generate persistent 3D environments from text or images. Environments you can download and use in Unreal Engine, Unity, or a VR headset. Li's quote on this, if large language models can teach machines to read and write, systems like Marble can teach them to see and build.

And then NVIDIA, Jensen Huang's Cosmos platform. Open source stat, CES 2026, has been downloaded over 2 million times. It was trained on 20 million hours of real-world footage. Companies like Figure AI and Uber are already using it to train robots and self-driving systems.

By the way, if you want to know how to use world models and AI automation that actually grow your business right now, not in five years, now, come join the AIprofitboredom.com. Link in the comments description and we will basically give you the roadmaps, the automations, the processes to generate more leads, save time, and grow with AI automation. So what does this actually mean for you? Well for agencies, freelancers, for creators, for e-commerce people, in the near term, world models are going to show up first in creative tools.

Marble is already letting designers generate fully editable 3D environments from a text description with one prompt. Explorable space, exportable assets, that's a workflow that used to take a team of 3D artists weeks. For anyone doing product photography, virtual staging, game ready assets, architectural visualization, this is coming for your workflow fast. For agencies who are maybe building content for brands, interactive 3D product demos are about to become as easy to generate as a video is today.

We're probably 12 to 18 months from that being mainstream. A bigger shift is in robotics and automation and this is the one most people aren't watching yet. World models are how AI learns to operate in the physical world. Right now companies are training robots by running them through real-world scenarios hundreds of thousands of times.

World models replace that. You simulate the environment, train in the simulation, and then deploy in reality. The human-eyed robot market is projected to hit 38 billion dollars by 2035. Accenture estimates that virtual twin technologies, the commercial version of this idea, could unlock 1.29 trillion dollars in economic value by 2030.

And I'm not saying this to scare you, I'm saying it so you understand why every major player in AI just pivoted to the same thing at the same time. Here's the honest picture. World models are very early. The World Model Bench benchmark from Nuri IPS, 2025, found that even the best world models follow instructions correctly only 61% of the time.

And Google Genie 3 holds consistency for minutes but not hours. World Labs Marble by their own co-founder's admission hits limitations after a few steps. This is not finished technology, this is a race with the finish line years away. But the money is moving, the talent is moving.

OpenAI just killed its most viable and visible customer product to fund this bet. Google DeepMind is calling it a path to AGI. The godmother and the godfather of AI both have billion-dollar startups chasing it. When that many serious people point in the same direction at the same time it's worth paying attention to.

The AI race was always going to move beyond text and images, beyond generating content. The next frontier is AI that understands reality, that can plan, act, navigate and operate in the physical world. OpenAI killing Sora wasn't a retreat, it was a bet on what comes next and right now everyone is racing to get there first. If you want to stay ahead of moves like this, understand what they mean for your business before everyone else does, the AI Profit Boardroom is where that conversation is happening every single week.

It's an AI automation community that helps you grow, learn, save time and get more customers and leads using AI automation. Link in the comments description or go to the AIProfitBoardroom.com Thanks for watching.