AMI Labs' CEO Says 'Superintelligence' Is a Meaningless Word. Here's What He's Building Instead.
Alexandre LeBrun raised over a billion dollars to teach AI how the physical world works. He thinks the buzzwords his rivals love are, at best, unhelpful.
Key points
- AMI Labs raised $1.03 billion in March 2025 at a $3.5 billion pre-money valuation, with no product yet on the market.
- CEO Alexandre LeBrun says terms like "AGI" and "superintelligence" have no agreed definition and are not useful.
- AMI Labs is co-founded by Yann LeCun, winner of the Turing Award, computing's closest equivalent to a Nobel Prize.
- LeBrun visited Seoul in June 2025 to court robotics, semiconductor and manufacturing partners, calling South Korea's industrial base "unique".
- South Korea announced a plan in June 2025 to mobilise roughly $880 billion for chips, AI data centres and physical AI.
When every other AI chief is racing to declare they've built "superintelligence," Alexandre LeBrun is doing something quieter. He's refusing to use the word at all.
LeBrun is the CEO of AMI Labs, a startup co-founded by Yann LeCun after LeCun left Meta. Its goal is to build a "world model," a type of AI that understands how physical reality works, not just language. Think of it as the difference between a system that can describe a glass of water and one that knows the glass will tip and spill if you nudge it.
"We never used the word AGI," LeBrun told TechCrunch AI. "And I just noticed that nobody is using it anymore; they switched to superintelligence. Next time we'll switch to something else." His verdict on the new label: "There's no good definition. What is superintelligence? I don't know. It's not a very useful word."
It is a pointed position for a founder sitting atop a $3.5 billion valuation.
LeBrun drew a clean line between two kinds of AI. A large language model, the technology behind chatbots like ChatGPT and Claude, predicts the next word in a sequence. A world model predicts the next state of a physical situation. He sees them as complementary, not competing, much like the way a human brain handles language and spatial reasoning through separate systems.
The gap, he argues, shows up most painfully in robotics. Today's factory robots repeat fixed movements in controlled spaces and work well enough in that narrow lane. Move the robot into a home, a street, or any unscripted environment, and things break down fast. "Robots are not safe right now," he said. "There's no solution for that today."
He offered a vivid example: a robot that was dancing and performing kung fu moves at a public event approached and kicked a child. Better contextual awareness, he said, is exactly what world models are meant to provide. "The hardware is very advanced, but there's no brain."
LeBrun was speaking to TechCrunch AI from Seoul, where he attended the International Conference on Machine Learning and scouted for partners in Korea's hardware-heavy industries. Chips, factories and robots are physically located in Asia, and AMI needs access to real environments to train its models. A lab alone won't cut it.
JP Lee, CEO of SBVA, one of AMI's backers in Asia, told TechCrunch AI he has been pushing LeBrun's team toward Korea for some time. Korea's speed matters as much as its factories, Lee said, noting the country was "the fastest adopter of the internet 25 years ago."
AMI has no product and no public timeline. "We'll make a surprise when we're ready," LeBrun said.
Should investors and partners worry that AMI has nothing to show yet?
That depends on what you're buying. What LeBrun is selling right now is a thesis: that physical AI is the next frontier, and that the companies and countries that build the industrial partnerships today will be best placed when world models mature. Whether the product follows is, for now, an open question.
