AMI Labs

AMI Labs, formally Advanced Machine Intelligence Labs, is a Paris-headquartered AI research company founded in late 2025 to build world models using the latent / JEPA (Joint Embedding Predictive Architecture) approach. It was co-founded by Yann LeCun, who serves as chief scientist after departing Meta, and is led by CEO Alexandre LeBrun. In March 2026 it announced a $1.03 billion seed round at a $3.5 billion pre-money valuation, described by TechCrunch and by the analyst firm Futurum as the largest seed round raised by a European company (Source: techcrunch.com; futurumgroup.com).

Snapshot

Funding

Technical approach

AMI Labs represents the latent / JEPA school of world models, one of three approaches the secondary coverage distinguishes alongside World Labs (generative pixel-space) and General Intuition (action-conditioned). The approach descends from LeCun's 2022 JEPA paper. LeCun argues that predicting abstract representations is more compute-efficient and learns better-generalizing structure than predicting pixels, sidestepping the "blurriness problem" he associates with pixel-prediction architectures.

LeCun's longstanding position is that a sufficiently good JEPA-style world model is the missing component for AGI, and that LLMs, however large, cannot learn the structure of causality from token streams alone. He has described AMI Labs as building toward that thesis.

Futurum characterises the company's target as world models that learn abstract representations of real-world sensor data rather than predicting outputs token by token, aimed at industrial, robotic and healthcare applications "where the limitations of large language models are most consequential" (Source: futurumgroup.com).

Business model and timeline

LeBrun has stated that AMI Labs is not structured to produce a near-term product. "It's not your typical applied AI startup that can release a product in three months, have revenue in six months, and make $10 million in [annual recurring revenue] in 12 months," he told TechCrunch, adding that it could take years for world models to move from theory to commercial applications. The company has no plans to generate revenue for the time being, and intends instead to engage prospective customers early: "We are developing world models that seek to understand the world, and you can't do that locked up in a lab. At some point, we need to put the model in a real-world situation with real data and real evaluations" (Source: techcrunch.com).

The first disclosed deployment partner is Nabla, a digital health company of which LeBrun is chairman and was previously CEO. TechCrunch reports that LeBrun reached the same conclusion as LeCun about the limitations of large language models while running Nabla, in a setting where hallucinations could have life-threatening consequences. LeBrun attributes the presence of several industrial backers in the round to a similar interest in early access. See Healthcare — AI Deployment.

LeBrun has also said the company will publish papers as it goes and open-source a substantial amount of code: "We think things move faster when they're open, and it's in our best interest to build a community and a research ecosystem around us," a stance he acknowledged is "increasingly rare" (Source: techcrunch.com). See Open-Weight Frontier Models.

Funding and investors

The March 2026 round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital and Bezos Expeditions. Corporate and strategic backers include Nvidia, Samsung, Sea, Temasek and Toyota Ventures, alongside the French investors Association Familiale Mulliez, Groupe Industriel Marcel Dassault and Publicis Groupe. Aglaé Lab, Alpha Intelligence Capital, Artémis, Bpifrance Digital Venture, New Legacy Ventures, SBVA and ZEBOX Ventures also participated. Named individual investors include Tim and Rosemary Berners-Lee, Jim Breyer, Mark Cuban, Mark Leslie, Xavier Niel and Eric Schmidt (Source: techcrunch.com).

LeBrun has said that investor interest allowed the company to select backers on expectation alignment and background rather than on terms alone, and that the capital funds two cost centres, compute and talent.

Futurum places the round in context: World Labs raised $230 million at a $1 billion valuation at its August 2024 launch, itself considered a large pre-product round at the time, and Thinking Machines Lab was valued at $12 billion in its seed round. On that comparison the AMI raise is not the largest in absolute terms, but Futurum describes it as a structurally different profile — a company with roughly a dozen employees, no product, and a research agenda measured in years — and notes that a $3.5 billion pre-money valuation implies investors are paying for scientific credibility and long-term option value, which creates execution pressure at the next financing (Source: futurumgroup.com). This is analyst commentary rather than a disclosure. See AI Bubble Debate.

People and locations

Yann LeCun is chief scientist, a role he took after leaving Meta, where he spent more than a decade leading the FAIR group. Alexandre LeBrun is CEO; he also worked at FAIR before founding Nabla.

The company is headquartered in Paris and operates from four locations: Paris; New York, where LeCun teaches at NYU; Montreal, where the researcher Rabbat is based; and Singapore, which LeBrun describes as serving both recruitment and proximity to prospective Asian customers. LeBrun has said hiring will prioritise quality over quantity (Source: techcrunch.com).

Reception and positioning

In an interview published March 9, 2026, LeBrun said:

"My prediction is that 'World Models' will be the next buzzword. In six months, every company will call itself a World Model to raise funding."

The remark is quoted in World Models: Computing the Uncomputable — Packy McCormick & Pim De Witte (Not Boring, March 2026) and originates in TechCrunch's coverage of the seed round; it comes from the CEO of a company named for the world-model approach. TechCrunch records that LeBrun said it "with a smile" and regards AMI Labs as different on the ground that its goal is to understand the real world (Source: techcrunch.com).

Open questions

• Whether the company's headquarters and incorporation are French is treated here as established by TechCrunch, which calls it "the French AI lab" and places the headquarters in Paris, and by Futurum, which describes hubs in four cities. An earlier version of this page recorded the company as US-based; that description is not supported by either account and has been corrected.
• No follow-on financing, product, publication or model release has been recorded since the March 2026 round. Futurum names the terms of a future Series A as the first market test of whether the research output translates into commercial credibility.

Relationships

• instance-of: World Models — JEPA / latent school.
• related: World Labs (generative pixel-space), General Intuition (action-conditioned), Google DeepMind (V-JEPA work), Meta AI (where LeCun previously led FAIR).
• founded-by: Yann LeCun.
• deploys-in: Healthcare — AI Deployment — Nabla is the first disclosed deployment partner.
• anchor source: World Models: Computing the Uncomputable — Packy McCormick & Pim De Witte (Not Boring, March 2026).

Confidence

• High for the JEPA-school positioning and LeCun's involvement.
• Medium for the funding figures. The $1.03 billion raise and $3.5 billion pre-money valuation are corroborated by TechCrunch and Futurum, but valuations are a fast-decay claim and the round closed in March 2026, outside the three-month reinforcement window.
• Medium for headcount, locations and partner arrangements, each carried by a single account.