AMI (Advanced Machine Intelligence)

AMI Labs is a Paris-based AI research company founded in late 2025 by Turing Award winner Yann LeCun after his departure from Meta, pursuing world models based on the JEPA architecture as an alternative to large-language models.
AMI (Advanced Machine Intelligence)

AMI (Advanced Machine Intelligence)

Advanced Machine Intelligence Labs, branded as AMI Labs, is a Paris-headquartered artificial intelligence research company founded in late 2025 after Yann LeCun's departure from Meta. The company is pursuing world models built on the Joint Embedding Predictive Architecture (JEPA), which LeCun has publicly framed as a fundamentally different approach to artificial general intelligence than the large-language-model paradigm pursued by OpenAI, Anthropic, and Google DeepMind. As of April 2026, AMI Labs is pre-product, with $1.03 billion in seed funding raised at a $3.5 billion pre-money valuation.

At a glance

  • Founded: Late 2025 in Paris, France. Public launch and seed-round announcement: March 9, 2026.
  • Status: Private. Pre-product.
  • Funding: $1.03 billion seed at a $3.5 billion pre-money valuation (approximately $4.5 billion post-money), announced March 9, 2026. Co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions.
  • Executive Chairman: Yann LeCun (2018 Turing Award co-recipient)
  • CEO: Alexandre LeBrun (co-founder of Nabla, an AI healthcare startup)
  • Other notable leadership: Michael Rabbat (VP, World Models), Saining Xie (Chief Strategy Officer), Pascale Fung (Chief Research Officer), Laurent Solly (Chief Operating Officer)
  • Open weights: None. No models released.
  • Flagship models: None publicly released.

Origins

AMI Labs was founded in late 2025, following Yann LeCun's announced departure from Meta in November 2025 after twelve years leading Facebook AI Research and the Llama model program. LeCun's departure was triggered in part by the June 2025 Meta restructuring under Alexandr Wang, after which LeCun was reportedly asked to report to Wang. LeCun has publicly stated that the conflict reflected a deeper strategic disagreement with the LLM-centric direction of the field.

The company's research thesis predates AMI itself. LeCun proposed the Joint Embedding Predictive Architecture (JEPA) in a 2022 position paper, arguing that artificial general intelligence requires AI systems that learn from observation of the physical world rather than from language alone. While at Meta, LeCun's group published successive JEPA-family models including I-JEPA (images, 2023), V-JEPA (video, 2024), V-JEPA 2 (trained on more than one million hours of internet video and integrated into physical robots for zero-shot generalization), and VL-JEPA (vision and language in a shared embedding space). The thesis is that these architectures, which predict embeddings of unseen content rather than reconstructing pixels or tokens, capture the structure of the physical world in ways that scale-and-prediction language models cannot.

The founding team was assembled rapidly after LeCun's Meta departure. Alexandre LeBrun, co-founder of the Paris-based AI healthcare company Nabla, joined as Chief Executive Officer. Michael Rabbat, a senior researcher with experience at Meta's FAIR organization, joined as Vice President for World Models. Saining Xie, a computer-vision researcher previously at NYU and Meta, joined as Chief Strategy Officer. Pascale Fung, a senior NLP researcher previously at Hong Kong University of Science and Technology and Meta, joined as Chief Research Officer. Laurent Solly, a French executive with Meta and corporate-leadership experience, joined as Chief Operating Officer.

The $1.03 billion seed round, announced on March 9, 2026, was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. Strategic corporate backers included NVIDIA, Samsung, Toyota Ventures, Temasek (Singapore's sovereign wealth fund), and Bpifrance (the French state investment bank). Notable individual investors included Tim and Rosemary Berners-Lee, Jim Breyer, Mark Cuban, Mark Leslie, Xavier Niel, and Eric Schmidt.

Mission and strategy

AMI Labs's stated mission is to build artificial intelligence that "learns from reality, not just from language." LeCun has publicly framed the strategy as a deliberate alternative to the LLM paradigm dominant at OpenAI, Anthropic, Google DeepMind, and xAI. The technical thesis is that JEPA-based architectures, by predicting embeddings of unobserved content rather than reconstructing pixels or generating tokens, can develop world models that capture causal physical structure in ways pure language models do not.

The strategy combines three threads. First, fundamental research on JEPA-family architectures, scaled to frontier-tier compute and data resources. Second, eventual integration of world models into physical agents, including robotics platforms, where the architecture's strengths in spatial and causal understanding should produce measurable advantages over LLM-based control. Third, a Paris headquarters with explicit European positioning, complementing Mistral AI's European frontier presence and Bpifrance's strategic-investor role.

The competitive premise is that the LLM paradigm has scaled into diminishing returns and that the next generation of capability will come from architectural rather than purely scale-driven advances. AMI's bet is that JEPA is that next architecture and that LeCun's research credentials, the depth of the founding team, and the $1 billion seed scale are sufficient to validate the thesis before LLM-based competitors close the gap.

The strategy carries unusual research risk. JEPA has been published academically but has not produced a frontier-tier general-purpose model at the scale of GPT-5.5 or Claude Opus 4.7. Whether the architectural advantages translate to commercial-scale capability remains an open empirical question.

Models and products

AMI Labs has not released any models, products, demonstrations, or research papers under the AMI name as of April 2026. The JEPA-family architectures developed by LeCun's group at Meta (I-JEPA, V-JEPA, V-JEPA 2, VL-JEPA) remain under Meta's ownership and have been released as open-weights research artifacts by Meta separately.

The publicly stated AMI roadmap includes scaling JEPA-based world models to frontier-tier compute and integrating them into physical agents, but no specific timeline or capability target has been disclosed.

Benchmarks and standing

There are no benchmark evaluations of AMI work, as the company has not released models or research artifacts. The benchmark profile is empty.

The company's standing rests on LeCun's research credentials (2018 Turing Award co-recipient, twelve-year leader of Meta's FAIR), the credibility of the JEPA thesis in the academic community, and the $4.5 billion post-money valuation supplied by the Bezos Expeditions and venture-capital co-leads.

Leadership

As of April 2026, AMI Labs's senior leadership includes:

  • Yann LeCun, Executive Chairman and co-founder. 2018 Turing Award co-recipient. Twelve years as Vice President and Chief AI Scientist of Meta before departing in November 2025. Public face for the company on the JEPA research thesis and the strategic critique of LLMs.
  • Alexandre LeBrun, Chief Executive Officer and co-founder. Co-founder of Nabla, the Paris-based AI healthcare startup. Provides commercial-leadership experience that the academic-leaning founding research team did not have in depth.
  • Michael Rabbat, Vice President for World Models. Previously a senior researcher at Meta FAIR working on JEPA-family architectures.
  • Saining Xie, Chief Strategy Officer. Computer-vision researcher with NYU and Meta backgrounds.
  • Pascale Fung, Chief Research Officer. Senior NLP and dialogue researcher with HKUST and Meta backgrounds.
  • Laurent Solly, Chief Operating Officer. French executive with senior Meta and corporate experience, including Vice President roles at Meta in southern Europe.

The team is concentrated in former Meta researchers but extends to French and European AI talent that AMI's Paris base puts within recruiting reach. The Paris headquartering is unusual for a frontier-tier AI lab; the strategic implications connect to European AI policy, sovereign-AI funding (Bpifrance participation), and the broader European AI ecosystem that includes Mistral AI.

Funding and backers

AMI Labs's funding history through April 2026 consists of the March 9, 2026 seed round of $1.03 billion at a $3.5 billion pre-money valuation (approximately $4.5 billion post-money). The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions.

Strategic corporate investors included NVIDIA, Samsung, Toyota Ventures, Temasek (Singapore's sovereign wealth fund), and Bpifrance (the French state investment bank). The combination of US venture capital, Singapore sovereign capital, French state capital, Korean strategic investment (Samsung), Japanese strategic investment (Toyota Ventures), and US tech-industry strategic backing produces an unusually internationally diversified investor base.

Notable individual investors include Tim and Rosemary Berners-Lee, Jim Breyer (early Facebook investor), Mark Cuban, Mark Leslie, Xavier Niel (French telecommunications billionaire and Mistral backer), and Eric Schmidt (former Google CEO). The Niel and Bpifrance participation reinforces the European AI ecosystem connection.

The company has not announced subsequent rounds. Reports have not indicated specific plans for follow-on financing in 2026 or 2027.

Industry position

AMI Labs occupies a structurally distinctive position among Insurgent labs and within the broader AI industry. The combination of Yann LeCun's reputation, the JEPA architectural thesis as a deliberate alternative to LLMs, the $1 billion seed scale, the Paris headquartering, and the unusually international investor base produces a profile no other AI company has matched.

Strategic risks are substantial. The JEPA thesis has not been validated at frontier scale; the company is pre-product; the LLM paradigm has continued to deliver capability gains that JEPA must outpace to justify the architectural bet; and the geographic distance between Paris and the dominant US AI talent pool may constrain hiring relative to peer Insurgent labs.

Strategic strengths are equally distinct. LeCun's Turing Award credentials provide research credibility no peer Insurgent matches. The international investor base provides political and commercial alignment across Europe, Asia, and the United States. The architectural thesis, if validated, would produce capability advantages in physical-world tasks (robotics, autonomous vehicles, scientific simulation) that pure LLMs cannot easily address. The Paris base aligns with European sovereign-AI policy frameworks that may produce procurement advantages over US-domiciled competitors.

Industry coverage has placed AMI Labs among the most-watched 2026 AI startups, in part because of LeCun's high public profile and in part because the JEPA thesis is the most prominent published technical alternative to LLMs from a senior researcher with industrial credibility.

Competitive landscape

AMI Labs competes with several Frontier and Insurgent labs:

  • Meta AI / FAIR. LeCun's previous lab. Meta retains ownership of the JEPA-family research artifacts published while LeCun was Chief AI Scientist; AMI's continuation of the JEPA thesis is independent.
  • OpenAI. Direct competitor on the AGI mission, though pursuing a different architectural approach.
  • Google DeepMind. Research-deep peer with similar long-horizon framing. DeepMind's robotics and world-model research overlaps with AMI's stated direction.
  • Safe Superintelligence (SSI). Closest peer Insurgent: both founded in 2024 to 2025 by senior AI researchers, both pursuing differentiated technical theses, both pre-product, both at large valuations relative to product output.
  • Thinking Machines Lab. Senior-OpenAI-departure peer Insurgent at large valuation, though Thinking Machines pursues LLM-family research rather than the JEPA alternative.
  • Mistral AI. Closest European AI peer; both Paris-based, both with French state investment, though Mistral pursues LLM-family research and AMI pursues the JEPA alternative.
  • Wayve, Physical Intelligence, and other robotics-foundation-model labs. Indirect competitors on the physical-world AI thesis that AMI's JEPA approach also targets.

Outlook

Several open questions affect AMI Labs's trajectory in 2026 and 2027:

  • The first public release. Whether AMI will produce a paper, model, demonstration, or capability claim during 2026, and whether the demonstration validates the JEPA thesis at frontier scale.
  • Compute infrastructure. The seed-round size suggests substantial compute investment is planned; the scale, vendor (NVIDIA strategic relationship implies NVIDIA silicon), and timeline have not been disclosed.
  • Personnel scaling. The founding team is concentrated in senior researchers; growth into a frontier-tier organization will require substantial hiring across research, engineering, and infrastructure.
  • Robotics and physical-world integration. The JEPA thesis emphasizes physical-world capability; whether AMI announces robotics or autonomous-system applications is a watchable signal.
  • European AI policy alignment. AMI's Paris base, French state investment, and European policy environment may produce procurement opportunities; whether these materialize commercially is open.
  • Continued LLM scale. If LLM-family models continue to deliver capability gains through 2026 and 2027, the JEPA architectural thesis becomes harder to defend; if LLMs stall, AMI's bet becomes more credible.
  • Subsequent funding rounds. The seed-round size is unusual; whether AMI raises follow-on capital and at what valuation will indicate continued investor conviction.

Sources

About the author
Nex Tomoro

AI Research Lab Intelligence

Keep track of what's happening from cutting edge AI Research institutions.

AI Research Lab Intelligence

Great! You’ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to AI Research Lab Intelligence.

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.