AMI Labs Seeks Seoul Partners for World Models, Rejecting AGI Label After $1.03 Billion Raise
Updated
Updated · TechCrunch · Jul 16
AMI Labs Seeks Seoul Partners for World Models, Rejecting AGI Label After $1.03 Billion Raise
1 articles · Updated · TechCrunch · Jul 16
Summary
Seoul became AMI Labs’ latest target as CEO Alexandre LeBrun used last week’s ICML trip to court industrial partners, researchers and global companies to help train its still pre-product world models.
Real-world access is the key ask: LeBrun said world models must learn from factories, robots and other physical environments because lab training alone cannot make AI safe or context-aware in open settings.
LeBrun also distanced AMI from AI’s branding race, saying terms like “AGI” and “superintelligence” lack useful definitions, while arguing world models should complement rather than replace LLMs.
South Korea stands out to AMI for its robotics, chip and manufacturing base and for speed of adoption, with backer SBVA pointing to Seoul’s June plan to mobilize about $880 billion for chips, AI data centers and physical AI.
AMI’s outreach comes months after the Yann LeCun-backed startup raised $1.03 billion at a $3.5 billion pre-money valuation in March, though LeBrun gave no product timeline.
Is AMI Labs' $3.5B pre-product valuation a smart bet on physical AI or a sign of a massive market bubble?
Why is a top French AI lab turning to South Korea to build the brains for the world's next generation of robots?
Can world models overcome the 'data friction' problem that language models never faced to truly revolutionize the physical world?
AMI Labs Secures $1.03 Billion to Lead the Shift from LLMs to World Models in AI
Overview
World models mark a major shift in artificial intelligence by enabling AI systems to learn directly from real-world sensor data and video, rather than just text and code. Their main goal is to build an internal, predictive model of reality, allowing AI to understand how the world works, predict future states, and anticipate the consequences of actions. This approach helps AI operate effectively in complex, dynamic environments and is seen as a new foundation for real-world AI. Unlike large language models, world models focus on deep, intuitive understanding of the physical world, opening new possibilities for advanced AI applications.