AI World Models Funding Explodes to $1B for LeCun Startup

Yann LeCun’s new startup raised $1.03 billion Tuesday. That’s what investors bet on AI world models—a technology that learns from reality, not language.

AMI Labs closed the round at a $3.5 billion pre-money valuation. The French AI lab initially sought €500 million last December. Ended up raising €890 million instead. Why? The team.

LeCun chairs the company. He’s a Turing Prize winner who left Meta to build this. Alexandre LeBrun runs it as CEO—he’s an entrepreneur who previously led Nabla, a digital health startup where he’s now chairman. Laurent Solly joined as COO from his role as Meta‘s VP for Europe.

Then there’s the research team: Saining Xie as chief science officer, Pascale Fung leading research and innovation, Michael Rabbat running world models development from Montreal.

Deep bench for a long game.

## What AI World Models Actually Do

World models learn from reality. Not from text scraped off the internet.

Large language models process language. They predict the next word. Sometimes they hallucinate—make up facts that sound plausible but aren’t true. In healthcare, that kills people.

LeBrun faced this problem at Nabla. He reached the same conclusion as LeCun: LLMs have fundamental limitations. You need AI that understands the actual world.

That’s where AI world models come in. The technology builds on JEPA—Joint Embedding Predictive Architecture—which LeCun proposed in 2022. Instead of predicting words, it predicts what happens next in physical reality.

“My prediction is that ‘world models’ will be the next buzzword,” LeBrun told the press. “In six months, every company will call itself a world model to raise funding.”

He said it with a smile. Because he thinks AMI Labs is different: the goal is actually understanding the real world, not just marketing hype.

## The Long Road Ahead

“AMI Labs is a very ambitious project, because it starts with fundamental research,” LeBrun explained. “It’s not your typical applied AI startup that can release a product in three months, have revenue in six months and make $10 million in ARR in 12 months.”

Could take years to go from theory to commercial applications.

Most AI startups chase quick revenue. Build a chatbot wrapper, charge $20/month, scale to $1M ARR in six months. That’s the playbook.

AMI Labs burns that playbook. No plans to generate revenue immediately. Pure research phase first.

When I ran TaskFlow, we needed revenue by month three or we died. Different game here. These investors bought a decade-long bet on fundamental AI architecture. That’s rare.

## Who Else Is Building This

The AI world models category has fewer players than generative AI. But the checks are massive.

Fei-Fei Li’s World Labs raised $1 billion last month. SpAItial closed a $13 million seed—unusually large for a European startup. Now AMI Labs joins with $1.03 billion.

Total raised by these three: over $2 billion. For technology that won’t ship products for years.

That’s how much capital believes world modeling beats language modeling.

The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. NVIDIA participated. So did Samsung, Sea, Temasek, and Toyota Ventures.

Angel investors included Tim and Rosemary Berners-Lee, Jim Breyer, Mark Cuban, Mark Leslie, Xavier Niel, and Eric Schmidt.

LeBrun said high investor interest gave AMI Labs its pick of backers. They chose based on expectation alignment and industry connections.

## Where The Money Goes

Two main cost centers: compute and talent.

AI research burns cash on GPUs. Training models requires massive compute clusters. Costs scale exponentially with model size. That’s where hundreds of millions disappear.

Talent is the other sink. Top AI researchers command $1M+ compensation packages. AMI Labs needs dozens of them across four locations: Paris (headquarters), New York (where LeCun teaches at NYU), Montreal (Rabbat’s base), and Singapore (for Asian talent and future clients).

LeBrun said he’ll prioritize quality over quantity. Smart move. One world-class researcher beats ten mediocre ones in fundamental AI work.

## The Nabla Connection

Healthcare is the first application domain. Makes sense given the hallucination problem.

Nabla will be AMI Labs’ first partner accessing early models. LeBrun chairs Nabla, so the connection is direct. Digital health needs AI that doesn’t make up drug interactions or fake symptoms.

“We are developing world models that seek to understand the world, and you can’t do that locked up in a lab,” LeBrun explained. “At some point, we need to put the model in a real-world situation with real data and real evaluations.”

Nabla won’t be the last partner. “This may explain the presence and strong interest of certain industrial players and potential partners in the investment round,” LeBrun noted.

Toyota Ventures and Samsung didn’t invest for financial returns alone. They want early access to world models for robotics and autonomous systems.

## Open Source Strategy

AMI Labs will publish papers as it goes. Will also open source significant 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.

That’s increasingly rare in AI. OpenAI abandoned open research years ago. Anthropic keeps models closed. Most AI labs treat everything as trade secrets.

LeCun built his reputation on open research at Meta’s FAIR lab. He’s not changing approach now. LeBrun worked at FAIR too—same philosophy.

Question is whether openness works when competitors keep secrets.

In my experience, open source accelerates adoption but kills pricing power. AMI Labs isn’t chasing revenue yet, so the tradeoff makes sense now. Might regret it in five years when commercializing.

## What’s Next

Timeline to product: years, not months. Timeline to revenue: longer still.

Most startups die before finding product-market fit. AMI Labs has runway to survive the research phase. $1 billion buys a lot of time.

But it also raises expectations. Investors want 10x returns. On a $3.5 billion pre-money valuation, AMI Labs needs to reach $35 billion+ for a decent exit. That’s unicorn territory times ten.

Only happens if AI world models actually work better than LLMs for major use cases. And if AMI Labs executes faster than World Labs, SpAItial, and whoever else enters the space over the next six months.

LeBrun predicts every AI company will call itself a “world model” soon. That means the category gets crowded fast. First-mover advantage disappears. Execution becomes everything.

Next milestone: published papers showing JEPA-based models outperform LLMs on real-world tasks. Until then, this is a $1 billion research bet.

For investors with decade-long time horizons, that’s fine. For founders used to shipping product in 90 days, it’s a different universe.