7 min readJust now

–

Press enter or click to view image in full size

Yann LeCun invented the technology that makes your phone recognize faces. He won the Turing Award, the Nobel Prize of computing. He spent 12 years building Meta’s AI empire. And now, at 65, he just left to start a company that doesn’t exist yet and is already worth $3.5 billion.

Let me tell you why that should terrify you.

Yann LeCun

Before we talk about the money, you need to understand who Yann LeCun is. This isn’t some tech bro with a pitch deck. In the late 1980s, when neural networks were considered a joke in computer science, LeCun was building convolutional neural networks at Bell Labs. His LeNet system could read handwritten digits on bank checks. That technology, refined over decades, is now the backbone of every image recognition system on the planet.

In 2018, he shared the Turing Award with Geoffrey Hinton and Yoshua Bengio. They’re called the “Godfathers of AI” because they literally created deep learning. Not in a marketing sense. They built the math, wrote the papers, and proved it worked when everyone else had given up.

LeCun joined Facebook (now Meta) in 2013 to start their AI research division from scratch. For over a decade, he led one of the world’s premier AI labs. PyTorch, the framework that powers most AI research today, came from his team. Meta’s entire AI strategy, from content recommendation to the Llama models, has his fingerprints all over it.

This November, he announced he was leaving Meta. Not to retire. To start Advanced Machine Intelligence, or AMI Labs. The goal? Build “world models” that understand physics, maintain memory, and can reason like humans.

When Builders Become Bubble Makers

Let me paint you a picture of what’s happening in AI right now.

Ilya Sutskever left OpenAI in May 2024 after the Sam Altman drama, by September, he raised $1 billion at a $5 billion valuation for Safe Superintelligence. Six months later, another $2 billion at $32 billion valuation. His strategy? The company won’t release anything until they achieve superintelligence. No product. No timeline. Just a promise.

Mira Murati left OpenAI as CTO in September 2024. By July 2025, she raised $2 billion at a $12 billion valuation for Thinking Machines Lab. By November, reports said she’s in talks for funding at a $50 billion valuation. The company launched in February 2025 saying they’d show their first product “in the next couple months.” Five months later, still nothing.

Now LeCun. One month after leaving Meta, already talking about half a billion dollars at $3.5 billion valuation.

See the pattern? These aren’t random entrepreneurs. These are the actual builders of AI. People who understand the technology at the deepest level. People who, in other contexts, will tell you honestly that AGI is decades away and incredibly difficult.

But when they’re raising money? Suddenly it’s all achievable, it’s coming soon, and it’s worth billions.

The Numbers That Don’t Add Up

Let me show you what actual AI companies with real products look like.

OpenAI generates $13 billion in revenue. They’re losing billions annually, but at least they have paying customers. Anthropic hit $7 billion in annualized revenue in 2025. Again, not profitable, but real product, real users.

Now look at companies doing less glamorous AI work. Cursor, an AI coding tool founded in 2022, generates $500 million in revenue with just 150 employees. That’s $3.2 million revenue per employee. DeepL, the translation company, is actually profitable and independent. Anduril, the defense AI company, doubled revenue to $1 billion in 2024.

These companies solve specific problems. They have customers. They generate cash flow and they’re worth a fraction of what Ilya, Mira, and Yann are raising for companies with zero products.

Nabla, the healthcare AI company that just partnered with LeCun’s AMI Labs, has raised $120 million total and tripled its revenue this year to approach $1 billion ARR. They built an actual product used by actual doctors. Their total valuation is probably around $800 million to $1.2 billion.

LeCun’s company, which has zero revenue, zero product, and zero customers, is starting at $3.5 billion. It’s worth more than companies that took years to build real businesses.

Why Smart Money Makes Dumb Bets

Here’s what I keep hearing: “But these are geniuses! They invented this stuff!”

Yes. They are. But that doesn’t change the economics.

Think about it. LeCun has been saying publicly that world models will take a decade to develop. His own research papers talk about fundamental problems we don’t know how to solve. He left Meta because they wanted to focus on shipping products, and he wanted to do long term research that might not pay off for years.

So why are investors lining up? A few reasons.

First, it’s FOMO at the highest level. If one of these companies actually achieves a breakthrough, whoever didn’t invest looks like they missed the next Google. So everyone invests.

Second, infrastructure play. Notice who’s funding these companies. Google invested in Ilya’s company. Nvidia backed both Ilya and Mira. These companies then spend that money buying computing power and chips from their investors. The money goes in circles.

Third, the talent arbitrage. You’re not betting on a product. You’re betting on genius. LeCun won a Turing Award. That has value. Even if the company fails, the team could get acqui-hired, or pivot, or do something valuable.

Fourth, valuation inflation creates liquidity events. Early investors don’t need the company to succeed. They need the valuation to keep going up long enough to sell their stake to the next round of investors. It’s musical chairs with billions.

The Warning Nobody’s Heeding

MIT released a study showing 95% of organizations investing in AI are getting zero return. Let that sink in. Nineteen out of twenty companies spending money on AI are seeing nothing back.

Companies are spending half their operating cash flow on AI infrastructure. Consultants estimate we need $2 trillion in annual AI revenue by 2030 just to break even on current infrastructure spending. That’s more than Amazon, Apple, Google, Microsoft, Meta, and Nvidia’s combined 2024 revenue.

Even Sam Altman, who has every reason to hype AI, admitted at a private dinner that investors are “overexcited” and “someone is going to lose a phenomenal amount of money.”

The problem isn’t that AI won’t be valuable. It’s that we’re funding decade long research projects at software startup valuations. Research projects, by definition, might fail. Most do.

MY TAKE

I don’t question their intentions. They genuinely want to solve hard problems. They genuinely believe in their vision. But by accepting these astronomical valuations for pre-product companies, they’re setting a dangerous precedent.

When LeCun raises at $3.5 billion with no product, it tells every mediocre AI founder that they too deserve billion dollar valuations based purely on promises. When Ilya raises $2 billion without shipping anything, it says research timelines don’t matter. When Mira talks about $50 billion with nothing to show, it says credentials trump results.

This creates a cascade effect. The next founder sees these valuations and thinks, “If they can get that, surely I can get half that.” Investors see everyone else writing checks and feel pressure to participate. The bubble inflates.

The really sad part? These people don’t need this money. LeCun could have raised $50 million at a $200 million valuation and had plenty of runway for a decade of research. That would be a massive seed round for a research project. But $586 million at $3.5 billion? That’s not research funding. That’s bubble mechanics.

What happens next is predictable. Within two years, at least one of these mega valued, zero revenue AI companies will quietly wind down or get acquired for pennies on the dollar. The warning sign will be when they start talking about “pivoting to more immediate applications” or “refining the vision.”

The companies that survive will be the ones shipping actual products. Not the ones with the best credentials or biggest promises. Cursor is worth more than its valuation suggests because developers actually pay for it. DeepL is valuable because it actually translates better than Google. Anduril is growing because defense contractors actually buy their systems.

The fundamental mistake everyone’s making is confusing research potential with business value. Yes, LeCun might crack world models in ten years. That would be worth trillions. But the probability adjusted present value of that outcome is not $3.5 billion today. It’s maybe $200 million. Maybe.

What really worries me is this: when the bubble pops, and it will, it’s going to hurt the entire AI ecosystem. Good companies with real products will get lumped in with the vaporware. Funding will dry up across the board. Talented researchers will get burned. We’ll waste years of progress.

And the saddest part? The crash is entirely avoidable. If these legendary researchers would just accept reasonable valuations for what they’re actually doing, research projects with decade long timelines, we could avoid the bubble entirely.

But they won’t. Because everyone else is doing it. Because the money is there. Because FOMO is too powerful.

I respect Yann LeCun’s work immensely. His contributions to AI are foundational. But taking $586 million at a $3.5 billion valuation for a company that doesn’t exist yet isn’t bold research funding. It’s bubble inflation.