AMI Labs
Last reviewed
Jun 3, 2026
Sources
8 citations
Review status
Source-backed
Revision
v1 · 1,528 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 3, 2026
Sources
8 citations
Review status
Source-backed
Revision
v1 · 1,528 words
Add missing citations, update stale details, or suggest a clearer explanation.
AMI Labs is a Paris-based artificial intelligence research company founded by Yann LeCun, the Turing Award winner and former Chief AI Scientist at Meta, following his departure from the company in late 2025. The name stands for Advanced Machine Intelligence and is pronounced like the French word for "friend." [1][2] The lab emerged publicly in early 2026 and announced a roughly $1.03 billion seed round in March 2026, reported as the largest seed financing on record in Europe. [1][3] Its stated mission is to build world models grounded in LeCun's Joint Embedding Predictive Architecture (JEPA), an approach he positions as an alternative to the large language model paradigm that dominates the rest of the frontier-AI field. [1][4]
AMI grew directly out of LeCun's long-running disagreement with the direction of mainstream AI research. LeCun spent more than a decade at Meta, where he founded the FAIR research lab in 2013 and held the title of Chief AI Scientist. He left in November 2025 amid a broader reorganization at the company, which had folded much of its research effort into a product-focused superintelligence group. [4][5] The new venture was confirmed publicly around December 2025, a website went up in January 2026, and the company made its formal launch alongside the funding announcement in March 2026. [2][3]
LeCun, who shared the 2018 ACM A.M. Turing Award with Geoffrey Hinton and Yoshua Bengio for foundational work on deep learning and the convolutional neural network, serves as executive chairman of AMI rather than chief executive. [4] He has continued his professorship at New York University in parallel with the role. [2] Day-to-day leadership falls to a CEO, with LeCun setting the scientific agenda.
The founding team draws heavily from LeCun's former colleagues at Meta and from academia. The company is led by Alexandre LeBrun, a French entrepreneur who previously founded the conversational-AI startup Wit.ai (acquired by Facebook in 2015) and the digital-health company Nabla. [1][2]
| Role | Person | Background |
|---|---|---|
| Executive chairman | Yann LeCun | Former Chief AI Scientist, Meta; professor, NYU; 2018 Turing Award [4] |
| Chief executive officer | Alexandre LeBrun | Founder of Wit.ai and Nabla; former Meta [1][2] |
| Chief science officer | Saining Xie | Computer-vision researcher (NYU); prior work with Meta and Google [2] |
| Chief research and innovation officer | Pascale Fung | Former senior director at Meta AI [2] |
| VP of world models | Michael Rabbat | Former research director, Meta [2] |
| Chief operating officer | Laurent Solly | Former Meta vice president for Europe [2] |
AMI is built around LeCun's argument that scaling up large language models will not, on its own, produce systems with human-level or animal-level understanding of the physical world. He has repeatedly characterized text-only LLMs as impressive but fundamentally limited, describing the field's heavy investment in them as a distraction from more foundational research into how machines can learn how the world works. [4][6]
The lab's technical bet is on world models: systems that learn from sensory data such as video and other signals rather than from language alone, and that build an internal, predictive model of physical reality. [1][3] The core architecture is JEPA, which LeCun introduced in his 2022 position paper "A Path Towards Autonomous Machine Intelligence." [7] Rather than predicting raw pixels or tokens the way a generative transformer does, a JEPA predicts an abstract representation of a future observation from a representation of the present one. Operating in a learned latent space lets the model ignore unpredictable surface detail and concentrate on the structural, semantically meaningful parts of a signal that actually support planning and reasoning. [7][8]
This connects to LeCun's longer-standing framework of objective-driven AI and energy-based models. In that view, an intelligent agent uses a learned world model to imagine the consequences of possible actions and then chooses actions that satisfy a set of objectives, including safety constraints, instead of generating outputs token by token. [7][8] LeCun has argued that this kind of architecture, combined with self-supervised learning, is a more plausible route to capable autonomous systems than the autoregressive approach behind today's chatbots. The work is a continuation of the V-JEPA line of research LeCun's group pursued at Meta. [4][8]
LeBrun has been candid that the term may get diluted quickly. He suggested that "world models" would become the next industry buzzword within six months, with many companies relabeling existing work to attract funding. [1]
AMI's seed round closed at about $1.03 billion, roughly €890 million, at a pre-money valuation reported at $3.5 billion. [1][3] Several outlets described it as the largest seed round ever raised by a European startup, and reporting indicated that LeCun had initially sought a far smaller amount, on the order of €500 million, before investor demand pushed the total higher. [3][6]
The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, the investment vehicle of Amazon founder Jeff Bezos. [1][3] Strategic and corporate backers reported to have taken part include Nvidia, Samsung, Toyota Ventures, the Singapore state investor Temasek, and the technology group Sea. [1][2] French corporate investors named in coverage include the Mulliez family holding company, Groupe Industriel Marcel Dassault, and the advertising group Publicis. [1] A number of individual investors reportedly joined as well, among them World Wide Web inventor Tim Berners-Lee, venture investor Jim Breyer, entrepreneur Mark Cuban, French telecoms billionaire Xavier Niel, and former Google chief executive Eric Schmidt. [2][3]
| Funding detail | Reported figure |
|---|---|
| Round type | Seed [1][3] |
| Amount raised | |
| Pre-money valuation | ~$3.5 billion [1][3] |
| Announcement | March 2026 [1] |
| Co-lead investors | Cathay Innovation, Greycroft, Hiro Capital, HV Capital, Bezos Expeditions [1][3] |
AMI is headquartered in Paris, a base that fits both LeCun's French background and the strong representation of French investors and executives on its cap table and leadership team. [1][2] The company has said it plans additional offices in New York, Montreal, and Singapore, giving it a presence near major academic AI hubs in North America and a foothold in Asia. [2][3]
The lab has framed itself as a research-first organization rather than a product company. LeCun and LeBrun have said AMI does not plan to generate revenue in its early years, intending instead to spend roughly the first year on research and development. [2][6] Over a longer horizon the company has talked about producing fairly general intelligent systems that could be deployed across multiple domains, with target sectors including healthcare, robotics, industrial automation, and wearable devices. [2][6]
Even without near-term revenue, AMI has said it will engage prospective customers early so that its models are tested against real data and real evaluations rather than only in the lab. The first disclosed partner is the digital-health company Nabla, founded by LeBrun, which is expected to gain early access to AMI's models. [1][2] The company has also indicated it intends to publish research and open-source at least some of its code, consistent with the open-research culture LeCun championed at FAIR. [4][2]
AMI arrived during a wave of well-funded new AI labs and stood out for two reasons: the size of its seed round and its explicit rejection of the LLM-centric strategy pursued by OpenAI, Anthropic, Google DeepMind, and LeCun's former employer. Where those organizations have concentrated on ever-larger generative models, AMI is wagering that the path to more capable, more reliable AGI-adjacent systems runs through world models and embodied learning instead. [1][4]
Reception has been mixed. Supporters see AMI as the best-funded attempt yet to commercialize an alternative research agenda led by one of the field's most decorated scientists. Skeptics note that JEPA-style world models remain largely unproven at scale, that the approach has produced fewer headline results than LLMs, and that a billion-dollar valuation rests heavily on LeCun's reputation and a thesis that could take years to validate. [3][6] How that bet plays out is likely to be one of the more closely watched questions in AI through the rest of the decade.