Yann LeCun’s AMI Labs Raises $1 Billion for Modular AI

Yann LeCun’s Advanced Machine Intelligence Labs, a 12-person research startup, has raised $1 billion to develop modular, domain-specific artificial intelligence systems as an alternative to large language models.

LeCun left his role as chief AI scientist at Meta late last year and founded AMI Labs. The organization will operate as a research lab and does not expect to deliver a commercial product for around five years.

The lab plans to assemble AI instances from specialized modules. Each instance would include a domain-specific world model that represents the AI’s operating environment; an actor that proposes actions using reinforcement learning; a critic that evaluates proposed actions using short-term memory and rule checks; a perception system for inputs such as video, audio, text or images; short-term memory; and a configurator that routes information between modules.

Each module would be trained on directed data relevant to its environment and purpose rather than on broad internet text. The relative importance and training methods for each module would vary by use case, with greater emphasis on the critic for sensitive-data settings and on perception for systems that must respond to real-world events.

AMI Labs says these specialist modules could require far fewer parameters-potentially in the low hundreds of millions-and run on a fraction of the GPU power used by large LLMs or on-device. Large language models from major companies now use hundreds of billions of parameters and require increasing compute to train and run; iterative prompting and refinement can increase runtime costs.

Investors provided $1 billion to support AMI Labs’ research program. The funding will back development of the modular architecture and experiments with domain-focused training methods.

The lab plans to test whether smaller, purpose-built modules can carry out narrow tasks with lower resource demands while other firms continue to develop large, general-purpose models.