Introduction

If you want to understand where AI is really going — not the hype, not the press releases, not the product launches you pay close attention to what Yann LeCun does.

LeCun is one of three people who essentially invented modern AI. He, Geoffrey Hinton, and Yoshua Bengio are known as the "Godfathers of Deep Learning" they developed the theoretical and practical foundations that everything from ChatGPT to Google Photos to self-driving cars is built on. When these three men speak about AI, the entire industry listens.

LeCun has been saying for years loudly, controversially, in direct opposition to most of his peers that the dominant approach to AI in 2026 is fundamentally wrong. That large language models like ChatGPT, Gemini, and Claude are impressive but brittle. That they'll never achieve genuine intelligence. That the entire industry is building on a foundation that will eventually hit a wall it cannot climb over.

And today literally today, March 10, 2026 — the world found out that he's not just talking anymore. He's building.

Former Meta AI chief Yann LeCun's new company, AMI Labs, has raised more than $1 billion in funding in what the Financial Times described as Europe's largest ever seed round. The startup is focused on "world models" — an approach that builds AI systems that learn from reality and physical environments rather than relying mainly on next-token prediction. Backers include Nvidia, Temasek, and Jeff Bezos-linked capital. Rawlinzdesigns

Europe's largest ever seed round. Today. Here's why it matters.

The Problem With ChatGPT — According to the Man Who Made AI Possible

To understand what LeCun is building, you first have to understand his critique of what everyone else is building.

Large language models ChatGPT, Gemini, Claude, and all their peers are trained primarily on text. Enormous amounts of text. Trillions of words scraped from the internet, digitised books, scientific papers, code repositories. They learn to predict the next word in a sequence with extraordinary accuracy, and from that seemingly simple task they develop remarkably sophisticated capabilities: reasoning, creativity, conversation, coding.

But LeCun has argued consistently that this approach has a fundamental ceiling. His argument, simplified: the world is not made of text. Human intelligence developed by interacting with a physical world touching things, moving through space, observing cause and effect, building mental models of how reality works. A child learns more about physics in their first two years of life by dropping things, falling over, pouring water than any language model learns from reading every physics textbook ever written.

AMI Labs is focused on "world models" an approach that builds AI systems that learn from reality and physical environments rather than relying mainly on next-token prediction. Rawlinzdesigns

World models are AI systems that build internal representations of how the physical world works not just statistical patterns in text, but genuine models of causality, physics, space, and time. Instead of predicting the next word, they predict the next state of the world. Instead of learning from language, they learn from observation of reality.

The vision LeCun is pursuing is AI that genuinely understands the world not AI that has read everything ever written about the world. To him, these are profoundly different things.

Why This Is More Than Academic Debate

At this point you might be thinking: this sounds like a fascinating philosophical argument about AI architectures. Why does it matter for me, my business, or my life?

Because if LeCun is right and $1 billion worth of some of the world's smartest investors believe he might be then the AI tools everyone is using today are not the final form of AI. They are a powerful but limited early chapter. And the next chapter AI that genuinely understands the physical world will be dramatically more capable in ways that current AI simply cannot match.

Think about what AI cannot do well today despite its impressive capabilities. It cannot reliably navigate a new physical environment. It cannot manipulate unfamiliar objects. It cannot plan sequences of physical actions in the real world with the ease that a five-year-old can. It struggles with spatial reasoning, with genuine causal understanding, with the kind of common-sense physical intuition that humans develop in infancy.

These limitations are not bugs that can be fixed by making current language models bigger. According to LeCun, they are fundamental architectural constraints the inevitable consequence of training AI primarily on text.

World models would address these limitations directly. And the applications would be extraordinary: robots that can genuinely navigate and operate in the physical world. Autonomous vehicles that understand causality, not just pattern-match on billions of miles of training data. AI assistants that can reason about physical tasks with the intuition of a human. Scientific AI that develops genuine theories about how the world works not just finds patterns in published papers.

The Investors Backing This Bet

The fact that this raise happened and at this scale is arguably as significant as what AMI Labs is building.

Backers include Nvidia, Temasek, and Jeff Bezos-linked capital. rawlinzdesigns.co.ke

Let's break down what each of those represents.

Nvidia is the company that makes the chips that power almost every AI system on the planet. Their investment is a signal of extraordinary strategic significance the company that physically builds the hardware for current AI is betting that the next architecture of AI will require different approaches, and they want to be positioned for it.

Temasek is Singapore's sovereign wealth fund one of the most sophisticated and disciplined long-term investors in the world. They do not make $1 billion seed bets on academic experiments. They invest in technologies they believe will reshape global industries within their investment horizon.

Jeff Bezos-linked capital the man who built Amazon and Blue Origin, who has consistently backed transformative technology bets when most of the industry was sceptical.

This is not speculative venture capital chasing a hot trend. This is patient, sophisticated capital betting that the dominant AI architecture of 2026 is not the dominant AI architecture of 2035 and that Yann LeCun has a plausible path to what comes next.

The Landscape of AI Dissent: LeCun Is Not Alone

LeCun's critique of large language models is controversial but he is far from the only credible voice raising these concerns.

Geoffrey Hinton — LeCun's fellow Godfather of Deep Learning left Google in 2023 and has spoken extensively about the risks and limitations of current AI approaches, albeit from a different angle. Where LeCun focuses on architectural limitations, Hinton focuses on safety risks.

More practically, the enterprise world is beginning to encounter the real-world limitations of current AI in ways that align with LeCun's critique. LLMs hallucinate — they generate confident, plausible-sounding falsehoods. They lack reliable causal reasoning. They cannot plan sequences of actions in dynamic environments without significant scaffolding. They are extraordinarily capable within a specific domain while being surprisingly brittle at the edges of that domain.

These are not criticisms from AI sceptics. They are observations from the companies and organisations deploying these tools at scale every day.

Deloitte's 2026 Tech Trends report captures this tension precisely: "The question used to be 'What can we do with AI?' Now it's 'How do we move from experimentation to impact?'" The focus has moved from endless pilots to real business value and for many organisations, that transition is exposing the gap between what AI demos suggest is possible and what AI deployments actually deliver. Instagram

AMI Labs is betting that bridging that gap requires a different kind of AI, not just a larger version of the current kind.

What World Models Could Change — Concretely

Let's make this tangible. What would world-model AI actually enable that current AI cannot?

Robotics that actually works. We covered humanoid robots in a previous blog — and one of the recurring themes was that current robots are impressive at specific, pre-programmed tasks but struggle enormously with the open-ended physical manipulation that humans handle effortlessly. World models could give robots the physical intuition they currently lack understanding how objects behave, how forces interact, how to handle unfamiliar items without pre-programmed instructions for every scenario.

Autonomous vehicles that understand causality. Current self-driving systems are extraordinarily good at pattern matching they've seen millions of miles of training data and can recognise road situations they've encountered before. But they struggle with truly novel scenarios. A world model-based system would understand why traffic behaves the way it does the underlying causal mechanisms rather than just recognising patterns in data.

Scientific discovery at scale. Current AI can find patterns in scientific literature and suggest hypotheses based on those patterns. A world model AI could genuinely reason about physical and chemical phenomena building theoretical models of how the world works, suggesting experiments that test those models, and iterating based on results. The implications for drug discovery, materials science, and fundamental physics research are extraordinary.

Personal AI that genuinely helps with real-world tasks. Today's AI assistants are remarkable at language tasks and increasingly capable at taking digital actions. But ask them to help you plan a physical task rearranging your office, packing for a move, organising a warehouse and the limitations become apparent quickly. World model AI would understand the physical dimensions of these problems in ways that current systems simply cannot.

The $1 Trillion Question: Can LeCun Actually Build This?

Raising $1 billion is extraordinary. Building what LeCun is describing is harder.

World models are not a new idea. Researchers have been working on them for years. The challenge is that building AI systems that genuinely learn from physical reality rather than from text requires different training data, different architectures, and different computational approaches. These are unsolved research problems, not engineering challenges with known solutions.

LeCun knows this better than anyone. He is not claiming to have built world model AI. He is claiming to have a credible research path toward it and that the potential is large enough to justify the capital required to pursue it.

The history of AI is full of approaches that were dismissed as dead ends and then proved central, and approaches that seemed promising and then hit insurmountable walls. LeCun's own work on convolutional neural networks was dismissed for years before it became the foundation of modern computer vision.

What is beyond dispute is that the bet has been placed publicly, loudly, with $1 billion behind it and that it will shape the trajectory of AI research for years to come. Even if AMI Labs doesn't achieve everything LeCun is aiming for, the research it generates will push the entire field forward.

What This Means for AI Users and Businesses Today

Does this mean you should stop using ChatGPT, Claude, and Gemini while waiting for world model AI?

Absolutely not. Current AI tools are extraordinarily powerful right now, for a wide range of tasks, and their capabilities are improving rapidly. The limitations LeCun identifies are real but they don't prevent current AI from delivering enormous value in language, code, analysis, creative work, and increasingly in digital action-taking.

What this does mean is that the AI landscape in five to ten years will look dramatically different from today. The tools you're building your workflows around now may be superseded by architectures that approach intelligence in fundamentally different ways. Businesses that treat AI as a permanent strategic investment building genuine capabilities and genuine understanding, rather than just plugging in current tools will be better positioned to adapt to that transition.

As Deloitte's 2026 Tech Trends report observes: "Innovation is compounding forces aren't simply additive, but multiplicative. Better technology enables more applications. More applications generate more data." The flywheel of AI development is spinning faster than ever and the companies that understand the direction it's spinning will make better decisions about where to invest their time, money, and attention. Instagram

At rawlinzdesigns.co.ke, we help businesses navigate exactly this kind of technological transition building digital products and strategies that are grounded in current capabilities while remaining adaptable to where the technology is heading. The worst position to be in is one where your digital infrastructure is so tightly coupled to today's tools that you can't evolve when the landscape changes.

The Bottom Line

Yann LeCun helped invent the AI that powers ChatGPT, Gemini, and Claude. And today he raised $1 billion to build something different something he believes will surpass all of them.

AMI Labs has raised more than $1 billion Europe's largest ever seed round backed by Nvidia, Temasek, and Bezos-linked capital, focused on world models: AI that learns from physical reality rather than text prediction. Rawlinzdesigns

Whether LeCun is right about the limitations of current AI, whether world models will deliver on their promise, and whether AMI Labs will be the company to build them these are genuinely open questions. The research is hard. The timeline is uncertain. The competition from OpenAI, Google, Anthropic, and Meta will be fierce.

But $1 billion Europe's largest seed round ever says that enough people who understand both the technology and the stakes believe this is worth betting on.

The next chapter of AI has a name. It's called AMI Labs. And its first page was written today.

Want to build digital products that are ready for the next generation of AI not just the current one? Talk to rawlinzdesigns.co.ke— we build smart, future-ready digital solutions for businesses that are serious about staying ahead.

Further Reading:

Yann LeCun Just Raised $1 Billion to Bet Against ChatGPT — And He Might Be Right