Yann LeCun’s AMI Labs Closes Record $1.03 B Seed Round for “World Models” AI

This funding echoes the late-1980s frenzy around Danny Hillis’s Thinking Machines (which also raised a then-record $250 M by 1990 to commercialise connectionist hardware) and Japan’s Fifth-Generation Computer Project (1982-1992), both grand quests to leapfrog prevailing AI methods. Investors are again wagering that a shift—from language token prediction to sensor-grounded ‘world models’—will unlock the next platform, just as ImageNet (2012) vaulted deep learning ahead of expert systems. The round also signals Europe’s determination to keep cutting-edge AI talent on the continent, countering decades of brain drain to Silicon Valley. Whether AMI’s JEPA-style architectures can scale economically, or fall into the fate of past “embodied AI” projects like MIT’s Cog robot (1993-2003), will shape robotics, autonomous systems and possibly the nature of AI research for the next half-century; but on a 100-year horizon it is another turn in the recurring cycle of over-promising vision and foundational method shifts that has punctuated AI since the 1956 Dartmouth workshop.