Yann LeCun AMI Labs: The Man Who Invented Modern AI Just Bet $1 Billion That Everyone Else Is Wrong
Yann LeCun AMI Labs is the story of the most consequential contrarian bet in AI history.
The man who won the 2018 Turing Award for creating the neural networks that made modern artificial intelligence possible, who spent a decade running AI research at Meta, who oversaw the development of Llama and PyTorch (the frameworks half the AI industry runs on), walked away from all of it. Then he raised $1.03 billion in a seed round at a $3.5 billion pre-money valuation. The largest seed round in European history. Second only globally to Thinking Machines Lab’s $2 billion raise in June 2025.
His thesis is simple and alarming. Every AI company spending billions on large language models (OpenAI, Anthropic, Google, xAI, Meta itself) is building the wrong technology. And Jeff Bezos, Nvidia, Samsung, Toyota Ventures, Temasek, Eric Schmidt, Mark Cuban and Tim Berners-Lee just agreed with him strongly enough to write the cheques.
For SMEs building AI strategies in 2026, this matters. When the person who helped invent the technology everyone else is scaling raises a billion dollars to run in the opposite direction, it’s worth understanding why.
Yann LeCun AMI Labs: The Story Behind the Billion Dollar Bet
Yann LeCun is not an outsider throwing stones at the AI industry. He is one of the three deep learning pioneers who shared the 2018 Turing Award (with Geoffrey Hinton and Yoshua Bengio) for the research that underpins essentially all of modern machine learning. He spent more than a decade as Meta’s Chief AI Scientist, founding and leading Facebook AI Research. The convolutional neural networks he developed in the 1990s are the foundation of modern machine vision. His work isn’t academic. It’s in every AI product you’ve ever used.
In November 2025, after sustained disagreement with Mark Zuckerberg over Meta’s AI direction, LeCun left. Four months later, on March 10, 2026, his new venture Advanced Machine Intelligence Labs (AMI, pronounced like the French word for “friend”) closed its seed round. The numbers are remarkable for a company with no product, no revenue, and an acknowledged research timeline measured in years rather than quarters.
The round: - $1.03 billion raised (roughly €890 million) - $3.5 billion pre-money valuation - Closed in roughly four months from company founding - Europe’s largest seed round ever - Second-largest seed round globally
Lead investors: Cathay Innovation, Greycroft, Hiro Capital, HV Capital and Jeff Bezos’s Bezos Expeditions.
Strategic backers: Nvidia, Samsung, Temasek, Toyota Ventures, Sea, SBVA, Alpha Intelligence Capital, Bpifrance, Publicis Groupe.
Individual investors: Jeff Bezos, Mark Cuban, Eric Schmidt (former Google CEO), Tim and Rosemary Berners-Lee (yes, the inventor of the World Wide Web), venture capitalist Jim Breyer, Xavier Niel.
The team: LeCun serves as Executive Chairman. Alexandre LeBrun (former Nabla CEO) is CEO. Saining Xie (ex-Google DeepMind) is Chief Science Officer. Pascale Fung (ex-Meta senior director of AI research) is Chief Research and Innovation Officer. Michael Rabbat (ex-Meta director of research science) is VP of World Models. Laurent Solly (ex-Meta VP for Europe) is COO.
Locations: Paris (HQ), New York, Montreal, Singapore.
LeCun sought €500 million initially. Demand was high enough that AMI Labs could be selective about which investors it accepted and still closed at more than double that figure.
Yann LeCun AMI Labs: The Thesis That Justifies the Cheque
Here is what LeCun has been arguing, publicly and consistently, for years. Large language models (the technology behind ChatGPT, Claude, Gemini and Grok) are statistical pattern-matching systems. They are trained to predict the next word in a sequence. Show an LLM “the cat sat on the” and it predicts “mat.” Scale that prediction across trillions of words and you get systems that produce fluent, often useful, sometimes brilliant output.
But LeCun’s point is that these systems do not understand anything. They have no model of how the world actually works. This is why AI hallucinates. It’s why a current frontier model can write a legal brief but cannot reliably predict what happens when you push a glass off a table, something a two-year-old understands instinctively. It’s why, as the recent ARC-AGI-3 benchmark results showed, frontier models score under 1% on tasks that require figuring out unfamiliar environments without instructions, while humans score 100%.
LeCun’s alternative is called JEPA, the Joint Embedding Predictive Architecture. Instead of predicting words, JEPA-based systems learn abstract representations of physical reality. They learn how the world works by observing it through sensors, cameras and video, rather than by reading about it. The goal is AI that can reason, plan and predict consequences in the real world, not AI that produces plausible-sounding text.
AMI Labs CEO Alexandre LeBrun, who previously ran the medical AI startup Nabla and saw firsthand how LLM hallucinations can have life-threatening consequences in healthcare, put it plainly: “Factories, hospitals and robots need AI that grasps reality. Predicting tokens doesn’t cut it.”
The target applications reflect this: industrial process control, healthcare, robotics, autonomous systems. Anywhere reliability and causal understanding matter more than linguistic fluency. LeCun’s thesis, in short, is that the AI industry has spent three years optimising the wrong thing, and the real breakthrough will come from a fundamentally different architecture.
This is a specific bet within the broader world models category we covered in our AI world models blog, which reviews the five key players including AMI Labs, Fei-Fei Li’s World Labs, Google DeepMind’s Genie 3 and Nvidia’s Cosmos platform.
Yann LeCun AMI Labs: Why the Investor List Matters
The investor syndicate is worth pausing on, because it tells you something about how seriously this thesis is being taken.
Bezos Expeditions is Jeff Bezos’s personal investment vehicle. Bezos does not invest indiscriminately. When he co-leads a seed round, it signals conviction. Nvidia’s participation is particularly interesting: Nvidia dominates the AI infrastructure layer (as we covered in our Nvidia AI dominance blog) and has every commercial reason to support LLM scaling. Their investment in AMI Labs suggests they see world models as complementary to their future, not a threat to it.
Samsung and Toyota Ventures bring industrial weight. Both operate at the intersection of manufacturing, robotics and sensors, exactly where JEPA-based systems could create the most value. Temasek brings Singapore sovereign wealth backing and the long-term patient capital that research-heavy ventures require.
Tim Berners-Lee’s participation is symbolically powerful. The inventor of the World Wide Web backing an AI startup that argues the dominant AI paradigm is wrong is the kind of signal that makes seasoned investors pay attention.
Two Turing Award winners (LeCun at AMI Labs and Fei-Fei Li at World Labs) have now raised over $2 billion combined in just a few months, both arguing against the LLM-only paradigm. This is not fringe dissent. This is the founding generation of modern AI quietly building the exit strategy from the technology everyone else is betting their future on.
What This Means for SMEs
Here is what UK business leaders should take from this, and what they shouldn’t.
What you shouldn’t take: This is not a reason to stop using AI tools. Current LLMs (ChatGPT, Claude, Gemini and the rest) are genuinely useful for specific, well-defined tasks. Writing emails. Summarising documents. Generating code. Analysing text. Supporting customer service. These use cases are where SMEs are extracting real commercial value from AI right now, and nothing in LeCun’s thesis changes that.
What you should take: Three practical lessons.
First, be cautious about vendor lock-in to LLM-heavy platforms. If LeCun is right, and enough serious capital and talent agrees with him that it’s worth taking seriously, the dominant AI architecture will shift in the next five to ten years. The businesses best positioned for that shift will be the ones whose AI strategies are flexible, not the ones locked into long-term contracts with today’s frontier model providers. Building flexibility into your AI Roadmap is commercially sensible.
Second, understand what AI can and cannot reliably do. LeCun’s argument is technical, but its practical implication for SMEs is obvious. Don’t deploy AI in contexts where hallucination or misunderstanding could genuinely harm your business or your customers. Legal advice, financial decisions, medical guidance, safety-critical processes. These are areas where current LLMs are not reliable enough, and no amount of marketing from the companies selling them changes that. An AI Workshop is the structured way to identify which parts of your operations are safe for AI implementation today and which aren’t.
Third, the AI landscape is restructuring faster than most SMEs realise. Between Nvidia consolidating infrastructure, Apple putting Gemini in 2.2 billion iPhones, the Musk OpenAI lawsuit testing the legal foundations of the industry, Google releasing frontier AI you can run on your own hardware, and now LeCun raising a billion dollars to build the alternative to everything everyone else is scaling, 2026 is not a settled market. SMEs that adopt AI with this reality in mind will thrive. SMEs that assume today’s AI leaders will still be tomorrow’s are taking more risk than they realise.
This is exactly why every AI investment should start with an AI Readiness Assessment. Not because you need to predict which AI paradigm wins, but because you need a plan that works regardless of which one does.
The Bottom Line
Yann LeCun AMI Labs is either the biggest wrong bet in AI history or the most prescient. Either LeCun and Fei-Fei Li are mistaken and the LLM-based AI industry keeps printing value. Or they are right and every company built on LLM-only architectures is about to face a fundamental technical reckoning.
For UK SMEs, the verdict on that question isn’t the point. The point is that $1 billion from the world’s most sophisticated investors just backed a thesis that the current AI paradigm is incomplete. Businesses that build AI strategies flexible enough to adapt, that deploy AI against use cases it can reliably handle today, and that don’t over-commit to any single vendor’s long-term survival, will come out of this restructuring period stronger than businesses that don’t.
Complete our free AI Readiness Assessment to understand where AI fits your business, and how to build a strategy that survives whichever architecture wins.
