Amilabs (formerly Hylé Labs) raised $1.2 billion in seed funding — the largest seed round of its kind in Europe — but according to founder and CEO Alexandre LeBrun, the principal cost isn’t equity dilution: it’s the surge in expectations that comes with the announcement. LeBrun warns that if a startup takes in a billion dollars and produces nothing tangible over two years, it becomes almost impossible to save.
Amilabs bets on methodical buildout
The capital was initially earmarked for GPU purchases to train foundational models. Rather than bow to media pressure, the team opted for an iterative, low-profile approach to avoid the “dead weight” of initial hype. For LeBrun, expectation pressure is the most dangerous variable in massive fundraising: it forces visible, fast deliverables, sometimes at the expense of deep engineering work.
LLMs versus world models: an architectural divide
LeBrun draws a clear distinction between large language models (LLMs) and so-called “world models.” He says the argument that LLMs are a path to general intelligence is largely settled: “A year ago, when we said LLMs were not AGI, most people disagreed. Today, I think most people recognize that LLMs don’t lead to general intelligence.”
He uses a metaphor to illustrate the difference: an LLM is like “someone who has never left the room where they were born, but who has read every book for centuries” — theoretically knowledgeable but lacking direct sensory experience. World models, by contrast, rely on sensory data (video, audio, tactile input through robotics) to build grounded representations of the environment that enable interaction and planning.
