🚨 THREAT ASSESSMENT: The world models race just got its biggest bankroll yet.
Yann LeCun has left Meta and is betting over a billion dollars that the future of AI isn’t language — it’s reality. His new venture, AMI Labs, has raised $1.03 billion at a $3.5 billion pre-money valuation, TechCrunch AI reports. The goal: build AI that learns from the physical world, not just text.
This is a massive signal. One of the three “godfathers of deep learning” is putting his reputation and research behind a fundamentally different approach to AI.
📌 The key facts
- Who: AMI Labs, cofounded by Turing Prize winner Yann LeCun (chairman) and Alexandre LeBrun (CEO, former Nabla CEO). The leadership roster is stacked with Meta’s former VP for Europe Laurent Solly as COO, plus top researchers Saining Xie, Pascale Fung, and Michael Rabbat.
- What: $1.03 billion raised, up from the €500 million originally rumored last December. Final haul: roughly €890 million.
- Investors: Co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. Strategic backers include NVIDIA, Samsung, Toyota Ventures, Sea, and Temasek. Angels include Eric Schmidt, Mark Cuban, Tim Berners-Lee, Jim Breyer, and Xavier Niel.
- Tech: Built on JEPA (Joint Embedding Predictive Architecture), LeCun’s 2022 framework for AI that predicts representations of reality rather than generating text token by token.
🔍 Why this matters
World models represent a philosophical break from the LLM paradigm that dominates today’s AI industry. Instead of predicting the next word, world models try to build internal representations of how the physical world works. LeCun has been vocal for years that LLMs alone won’t get us to human-level intelligence — and now he’s putting serious money behind that conviction.
“My prediction is that ‘world models’ will be the next buzzword. In six months, every company will call itself a world model to raise funding.”
He’s probably right. This is the third billion-dollar raise in the world models space in recent months. Fei-Fei Li’s World Labs secured $1 billion last month. SpAItial pulled in a $13 million seed round, which is large by European standards. The pattern is clear: investors are looking beyond chatbots.
🏥 Healthcare first, open research always
AMI Labs’ first disclosed partner is Nabla, a digital health startup where LeBrun previously served as CEO. The logic is straightforward: healthcare is where LLM hallucinations carry the highest stakes, making it the strongest case for models that actually understand reality.
But don’t expect products soon. LeBrun was blunt about the timeline: “AMI Labs is a very ambitious project, because it starts with fundamental research. It’s not your typical applied AI startup that can release a product in three months.”
The company will operate from four hubs: Paris (HQ), New York (where LeCun teaches at NYU), Montreal, and Singapore, prioritizing research talent quality over headcount.
One detail worth watching: AMI Labs commits to publishing papers and open-sourcing code. “We think things move faster when they’re open, and it’s in our best interest to build a community and a research ecosystem around us,” LeBrun said, according to TechCrunch AI.
📡 What to watch
- JEPA progress: Can LeCun’s architecture deliver on the promise of learning world representations? Published papers will be the scoreboard.
- The “world model” hype cycle: LeBrun’s six-month prediction about buzzword proliferation is worth tracking. If he’s right, expect a flood of rebranded startups.
- Strategic investors: NVIDIA, Samsung, Toyota — these aren’t passive checks. They signal potential deployment domains: robotics, autonomous systems, manufacturing.
- Healthcare validation: Nabla’s early access to AMI models will be the first real test of whether world models outperform LLMs in high-stakes domains.
The billion-dollar bet is placed. Now LeCun and his team have to prove that understanding reality is worth more than predicting words. Full details are available in the original TechCrunch AI report.