Dramatic exits are par for the course in today’s volatile artificial intelligence industry. But Yann LeCun, Meta’s former chief AI scientist, has done more than simply leave the company — he’s outright rejecting the driving force behind the generative AI boom thus far: large language models, or LLMs. By departing Meta, LeCun is staging a contrarian revolt to the notion that “superintelligence” can be found in a text box.
What Is AMI Labs?
AMI Labs is a frontier research startup focused on building world models. Moving beyond text-based language models, world models develop a deep understanding of the physical world by learning concepts like cause-and-effect and spatial logic through raw sensory data. Founded by computer scientist Yann LeCun, the goal of AMI Labs is to create a new kind of artificial intelligence capable of reasoning and planning with human-level intelligence.
His new venture, AMI Labs, tackles what is known as the Moravec’s Paradox, building so-called “world models” that teach machines the physical intuition and common sense us humans so often take for granted. By grounding AI in the messy, high-dimensional reality of sensor data rather than written content, AMI Labs is attempting to turn the apparent dead end of massive, proprietary LLMs into a sovereign, open-source path toward true human-level reasoning
What Is AMI Labs?
AMI labs is a Paris-based artificial intelligence company founded by Yann LeCun, a pioneer in computer science best known for his work on neural networks and long tenure at Meta. Launched in January 2026, the startup is primarily focused on building world models, which can understand basic principles of how the physical world works thanks to the real-world data they’ve been trained on, allowing them to power drones, robotaxis and other autonomous machines.
Pronounced “a-mee” — a nod to the French word for “friend” — AMI Labs positions itself as both a scientific initiative and strategic alternative to increasingly closed frontier labs. As LeCun sees it, open source is essential to AI innovation, whether that’s in business, academic research or countries seeking greater technological sovereignty. So the company is following a dual-track model, publishing research and contributing open-source tools while also developing commercially licensable products.
That being said, AMI Labs is not aiming to compete with generative AI leaders like ChatGPT maker OpenAI, Anthropic, the startup behind Claude, or Meta’s Superintelligence Labs. In fact, LeCun has suggested that his former employer might even become one of AMI’s first customers.
Framed as a “third path” alternative to disrupt the U.S.-Chinese tech giant binary, AMI Labs also aims to showcase Europe’s talent and research in the AI space, which, apart from standouts like Mistral and ElevenLabs, has largely been overshadowed by Silicon Valley giants. The decision to make Paris AMI’s headquarters was publicly welcomed by French President Emmanuel Macron, who pledged to “do everything to ensure [LeCun] succeeds from France.”
AMI plans to open additional offices in New York, Montreal and Singapore soon as well. LeCun himself remains in New York, where he continues his professorship at New York University, teaching and supervising graduate researchers while serving as AMI’s executive chairman.
LeCun is not the CEO of AMI Labs. That role belongs to Alex LeBrun, a serial entrepreneur and former colleague who previously worked under LeCun at Meta after the acquisition of his startup Wit.ai. The division of responsibilities was intentional, allowing LeCun to focus on long-term scientific direction and research rather than day-to-day management. LeBrun’s transition was facilitated through a partnership with his previous healthtech company, Nabla, which secured preferential access to AMI’s future technology in exchange for supporting his move.
Why Did Yann LeCun Create AMI Labs?
Turing award recipient and world-renowned computer scientist Yann LeCun doesn’t share the AI community’s obsession with large language models. If AI’s ultimate goal is to simulate human intelligence, he doesn’t think LLMs are an effective means of getting there.
In fact, some say we may have already hit the limits of what LLMs are capable of. LeCun argues the field should shift to building world models instead — systems trained not on vast troves of text data, but on multimodal, real-time sensory data that allows them to understand and simulate how the physical world works. In interviews with MIT Technology Review, he described the current LLM craze as a distraction that’s pulling attention away from more foundational research. As tech companies — including his own at the time, Meta — pushed further in that direction, the gap between LeCun’s vision and the AI industry’s priorities only widened.
So, when Meta doubled down on multi-billion dollar investments in large language models, centering its LLaMA models and commercial generative AI projects on its path to “personal superintelligence,” LeCun announced in a LinkedIn post he would be stepping away from the company and the Fundamental AI Research lab he created there. Overall, it was a difference in vision that led to the creation of AMI Labs, a startup focused on building “intelligent systems that understand the real world.”
Inside AMI Labs’ Research
World models, the central character in AMI Labs’ research, are built on artificial intelligence that learns how the physical world operates by processing continuous, high-dimensional sensor data — like images, video, audio and LiDAR — rather than just predicting the next word in a sentence. Through a specialized framework called joint embedding predictive architecture (JEPA), the startup created a kind of digital intuition, where systems are able to filter out noisy, background distractions of a messy environment. This allows the AI to operate in a “representation space” where it only cares about the underlying rules that govern our world.
This approach allows world models to develop a form of machine-made common sense. It can reason, plan complex sequences of actions, then anticipate the consequences of said actions while maintaining a persistent memory of its environment.
By building action-conditioned models, AMI Labs aims to solve something called Moravec’s Paradox, teaching machines the kind of spatial intuition that comes naturally to humans. Ultimately, the company’s work is designed to move AI out of the digital sandbox and into high-stakes, real-world applications where reliability and safety are non-negotiable — whether that’s putting a cobot on the factory floor, a self-driving car on the road or a surgical robot in a hospital’s intensive care unit.
What’s Next for AMI Labs?
AMI Labs launched in January 2026, and right out of the gate it was in discussions to raise $500 million, valuing the startup at $3.5 billion. Venture firms like Cathay Innovation, Greycroft and Hiro Capital (where LeCun is on as a financial advisor) are already planning to buy in, along with potential investors like 20VC, Bpifrance, Daphni and HV Capital. Much of this early enthusiasm stems from LeCun’s credibility as one of the architects of modern artificial intelligence, and a long-time advocate of open research.
So far, no major future plans have been announced. According to CEO Alex LeBrun, AMI Labs will likely take “about one year to get the first things we can use in the product,” referring to moving its world-model technology into applied use. LeBrun has also indicated that the startup will be targeting healthcare, robotics, wearables and industrial automation first.
Frequently Asked Questions
How do world models differ from large language models?
Unlike language models that predict the next word in a sentence, world models learn to predict the next state of the physical environment by observing video and sensor data. This allows the AI to develop a foundational understanding of physics, cause-and-effect and spatial logic rather than just mimicking human speech patterns.
Why did Yann LeCun leave Meta?
LeCun left Meta to develop world models, as he is convinced that the industry’s focus on large language models won’t ever be able to achieve true human-level intelligence. He wanted a research environment where AI could learn to understand and reason about the physical world — something he felt was being sidelined by Meta’s push into generative AI and so-called “personal superintelligence.”
Who runs AMI Labs?
Yann LeCun acts as executive chairman, focusing on long-term scientific direction. CEO Alex LeBrun, a former colleague at Meta and co-founder of Nabla, manages day-to-day operations.
Where is AMI Labs based?
AMI Labs is headquartered in Paris, with additional offices in New York, Montreal and Singapore coming soon.
