Yann LeCun, one of the most influential figures in artificial intelligence research, has launched a new startup that aims to chart a fundamentally different course from today’s dominant large language models, setting an ambitious valuation and fundraising target from its inception.
LeCun confirmed on the creation of Advanced Machine Intelligence (AMI) Labs, which is seeking around €500 million ($586 million) in funding at a pre-money valuation of roughly €3 billion ($3.5 billion). The Turing Award winner and former chief AI scientist at Meta will serve as executive chairman, while Alex LeBrun, co-founder of medical transcription company Nabla, has been appointed chief executive.
A Bet on ‘World Models’ Over Language Models
AMI Labs is focused on developing so-called “world models”, artificial intelligence systems designed to understand the physical world, reason about cause and effect, retain long-term memory, and plan complex actions. The approach stands in contrast to large language models, which generate outputs by predicting the next word in a sequence and dominate current commercial AI deployments.
LeCun, a persistent critic of the view that scaling language models alone will lead to human-level intelligence, has argued that systems trained primarily on text lack grounding in physical reality and therefore fall short in reasoning and planning. World models, by contrast, are intended to learn from video and spatial data, allowing machines to build internal representations of how the world works. LeCun has previously said such systems could take a decade or more to fully mature, underscoring AMI Labs’ long-term research focus.
Paris Headquarters Signals Strategic Shift
The company plans to establish its headquarters in Paris in early 2025, a deliberate move that reflects both LeCun’s personal ties to the city and his skepticism of Silicon Valley’s prevailing AI paradigm. Speaking earlier this year, LeCun said the Valley had become “hypnotized” by generative AI, arguing that truly novel research would require stepping outside its dominant assumptions.
Paris offers access to deep academic talent, a growing European AI ecosystem, and increasing public-sector support for advanced research.
LeCun left Meta after 12 years at the company, where he led its Fundamental AI Research (FAIR) lab and later served as chief AI scientist. Meta will not invest directly in AMI Labs, avoiding potential conflicts with its own AI strategy, which has increasingly emphasized large language models and near-term commercial applications.
The two organizations are expected to maintain a collaborative relationship, potentially involving shared research or infrastructure access, while remaining financially independent.
Meta’s AI Reorientation
His departure comes amid broader changes at Meta, which has reoriented its AI leadership and priorities following the appointment of Alexandr Wang, founder of Scale AI, as chief AI officer. Former employees have said the FAIR lab’s influence diminished as the company shifted resources toward product-driven AI teams, with a significant number of researchers leaving in recent years.
One of AI’s Largest Pre-Launch Fundraises
If completed, AMI Labs’ fundraising would rank among the largest pre-launch capital raises in the AI sector. The scale reflects investor confidence in LeCun’s track record, which includes foundational contributions to modern AI.
His work on convolutional neural networks in the 1980s laid the groundwork for computer vision systems that later processed a substantial share of bank checks in the United States.
Competing on Architecture, Not Scale
AMI Labs enters a competitive field that includes major research groups at Google DeepMind as well as newer startups pursuing similar world-model approaches. However, LeCun’s stature and research legacy are expected to help the company attract elite talent and patient capital willing to support extended development timelines.
The central bet behind AMI Labs challenges the prevailing assumption that ever-larger models trained on more data will naturally lead to more intelligent systems. Instead, the company is wagering that new architectures, grounded in perception, memory, and reasoning, are essential to overcoming the limitations of today’s AI, including hallucinations, weak causal understanding, and the inability to plan over long horizons.
