LeCun's AMI Labs Lands a Record Seed — and Picks a Fight With the LLM Consensus

Yann LeCun just closed what he calls the largest non-US seed round in history for AMI Labs, his new home for world-models research. Meta got 'LLM-pilled,' he says — so he left.

Yann LeCun is no longer hedging. His new venture AMI Labs has raised what he describes as the largest seed round ever closed outside the United States, and this week he made clear why he needed a home outside Meta: the company's AI strategy, he said, had become 'LLM-pilled and short-term focused' while his long-term bet on world models lost institutional oxygen.

LeCun disclosed his dual role — Jacob T. Schwartz Professor at NYU Courant Institute and Executive Chairman at AMI Labs — in a reply on X confirming the raise. He framed the Meta split carefully in a separate thread, crediting Mark Zuckerberg and Andrew Bosworth as 'always supportive' of the Joint Embedding Predictive Architecture (JEPA) and world-models program before company priorities shifted toward near-term LLM products. Industry reaction was immediate — Shawn Wang at Latent.Space posted simply: 'jesus h christ that seed round.'

The philosophical gap between LeCun and the generative-AI consensus also surfaced this week. Asked about p(doom) — the subjective probability assigned to AI causing catastrophic outcomes — LeCun pushed back on the framing itself, arguing that such estimates are 'pulled out of thin air' and that assigning probability to events over which humans have agency is a statement about collective decision-making, not technical risk. Taken together with the AMI Labs announcement, the message is coherent: he thinks the field is measuring the wrong things and optimizing for the wrong architectures.

For builders, the raise signals that well-capitalized research labs outside the LLM-scaling track remain viable as of April 2026. Industrial applications of world models — robotics, autonomous systems, high-fidelity simulation — are back in play with serious institutional funding. For product teams shipping on OpenAI and Anthropic APIs, near-term product-market fit still sits with LLMs; nothing about AMI Labs changes that shipping calculus in the next 12 to 18 months. Simon Willison meanwhile reminded developers this week that when model endpoints leak, it's credential hygiene — not URL obfuscation — that actually matters, a useful reminder that operational security questions stay constant regardless of which architectural bet wins.

The bottom line: Ship on the LLM stack that pays rent today, but put AMI Labs on your watchlist — it's the most credible public bet against the current consensus.

Sources