AMI Labs

AMI Labs (Advanced Machine Intelligence) is a frontier AI research lab headquartered in Paris with additional sites in New York, Montreal and Singapore. It develops "world models": AI systems trained on continuous, high-dimensional sensor data such as images, video, audio and LiDAR rather than on text, with the stated aim of building systems that understand the real world, retain persistent memory, reason and plan, and remain controllable and safe. The company's technical program centers on the Joint Embedding Predictive Architecture (JEPA) proposed by Yann LeCun in 2022, in which models predict abstract latent representations of the world instead of reconstructing pixels or generating tokens, on the argument that much real-world sensor data is unpredictable at surface level and that generative next-token or next-pixel prediction is poorly suited to it.

Action-conditioned world models are intended to let agentic systems predict the consequences of their actions and plan action sequences under safety guardrails. Targeted application domains where reliability and controllability matter include industrial process control and automation, wearable devices, robotics and healthcare; press coverage also cites autonomous machines such as drones and robotaxis. The company was founded in late 2025 by Yann LeCun after his departure from Meta, where he had led the Fundamental AI Research (FAIR) group since 2013, and it publicly launched in March 2026. LeCun serves as executive chairman while continuing his professorship at New York University; Alexandre (Alex) LeBrun is CEO, Laurent Solly is COO, Saining Xie is chief science officer, Pascale Fung is chief research and innovation officer, and Michael Rabbat is VP of world models. The website also lists Min Lin among the founding group.

AMI describes a dual-track approach of advancing fundamental research through open publications and open-source code while developing commercially licensable products, and positions itself as a European alternative to predominantly American and Chinese frontier labs. The company had roughly a dozen employees at the time of its seed round and no shipped product, with LeBrun stating the first phase is research-led and commercial applications are years rather than quarters away.

Founding story

LeCun has stated he created AMI Labs to continue the Advanced Machine Intelligence research program he had pursued with colleagues at Meta and New York University, after concluding that large language models are not a path to human-level intelligence and that Meta's priorities had diverged from world-model research. CEO Alexandre LeBrun, whose earlier startup Wit.ai was acquired by Meta and who later founded the health-tech company Nabla, had reached similar conclusions about LLM limitations in settings such as healthcare where hallucinations carry severe consequences; his move to AMI was facilitated by an arrangement giving Nabla preferential access to AMI's future technology. The name is pronounced "ah-mee," a reference to the French word for friend.

Business model

AMI Labs operates a dual-track model: it publishes research and releases open-source code while intending to develop commercially licensable products built on its world models. It has no revenue plans in the near term and expects a multi-year path from fundamental research to commercial deployment, with capital directed primarily at compute and talent.

No revenue is planned for the near term; the intended long-term model combines open research and open-source releases with commercially licensable products and partner deployments.

Traction

At the time of its March 2026 seed round the company had roughly a dozen employees and no shipped product, and was recruiting across its four locations. Traction to date consists of the financing, a research partnership arrangement with Meta enabling continued JEPA collaboration, and Nabla as first disclosed application partner.

Latest developments

AMI Labs announced its $1.03 billion seed round on March 9, 2026 at a $3.5 billion pre-money valuation and made its official public launch on March 10, 2026. V-JEPA 2.1, described as offering sharper temporally consistent dense features, shipped in March 2026 alongside the funding close.