AMI Labs Secures $1.03B to Advance World Model AI Research
Yann LeCun spent years at Meta arguing that large language models were the wrong path to artificial general intelligence. Now he's put $1.03 billion behind an alternative. AMI Labs, the Paris-headquartered AI research company LeCun co-founded after leaving Meta, has closed a funding round at a $3.5 billion pre-money valuation — well above the €500 million it was initially reported to be seeking late last year.
What AMI Labs is actually building
The company's focus is world models: AI systems designed to understand and reason about physical reality rather than predict the next token in a sequence. The technical foundation is JEPA — Joint Embedding Predictive Architecture — a framework LeCun proposed in 2022 as a more grounded alternative to the transformer-based models that power today's generative AI tools.
CEO Alexandre LeBrun, who previously sold his voice AI startup Wit.ai to Facebook and later ran Nabla, a digital health company, came to the same conclusion as LeCun through a different route. At Nabla, the stakes of LLM hallucinations weren't abstract — in healthcare, a confidently wrong answer can have serious consequences. That shared frustration with the limits of language-first AI is what brought the two together.
LeBrun is candid about the timeline. "AMI Labs is a very ambitious project, because it starts with fundamental research," he told TechCrunch. "It's not your typical applied AI startup that can release a product in three months." Commercial applications could be years away. Nabla is the first disclosed partner that will get early access to AMI's models, but the company plans to engage with prospective customers well before any product launch — because, as LeBrun put it, you can't build a model that understands the world while locked in a lab.
The team and the money behind it
The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. The angel list reads like a who's-who of tech: Tim and Rosemary Berners-Lee, Jim Breyer, Mark Cuban, Xavier Niel, and Eric Schmidt all participated. Strategic backers include Nvidia, Samsung, Toyota Ventures, Temasek, and Sea, alongside several French industrial groups — Publicis Groupe, Groupe Industriel Marcel Dassault, and Association Familiale Mulliez among them.
LeBrun said the level of investor interest gave AMI Labs room to be selective, prioritizing alignment on expectations over check size. That matters for a company explicitly telling investors not to expect near-term revenue.
The team itself is a significant part of the pitch. Beyond LeCun as chairman and LeBrun as CEO, AMI Labs has recruited Meta's former VP for Europe Laurent Solly as COO, NYU and Meta researcher Saining Xie as chief science officer, Pascale Fung as chief research and innovation officer, and Michael Rabbat — another Meta AI veteran — as VP of world models. The company will operate across four cities: Paris (HQ), New York (where LeCun teaches at NYU), Montreal (Rabbat's base), and Singapore, chosen both for AI talent and proximity to potential Asian clients.
Why this funding moment matters for AI's next chapter
World models are still a niche category, but the capital flowing into it is accelerating fast. SpAItial raised a $13 million seed — large by European standards — while Fei-Fei Li's World Labs pulled in $1 billion just last month. AMI Labs' raise now sits at the top of that pile, and LeBrun is already anticipating the inevitable hype cycle. "My prediction is that 'world models' will be the next buzzword," he said. "In six months, every company will call itself a world model to raise funding."
That kind of self-aware skepticism is notable coming from someone actively raising in the space. It also points to a real tension: the gap between what world models promise — AI that genuinely understands causality, physics, and context — and what most companies will actually ship under that label. The distinction matters because the use cases being targeted, healthcare chief among them, are ones where the failure modes of current LLMs aren't just inconvenient, they're potentially dangerous.
AMI Labs is also making a deliberate bet on openness. LeBrun, who worked at Meta's AI research lab FAIR before this, said the company will publish research and open-source significant portions of its code. "We think things move faster when they're open," he said — a philosophy that's become less common as AI labs increasingly treat model weights and training methods as competitive moats. Whether that openness survives contact with commercial pressure remains to be seen, but for now it's a meaningful differentiator in how AMI Labs is positioning itself within the research community.
The billion-dollar question — literally — is whether the underlying science can deliver on the ambition. LeCun has been making the case against LLMs as a path to general intelligence for years, often against the grain of industry consensus. AMI Labs is the most serious, best-funded attempt yet to prove him right.
