AMI Labs
AMI Labs (Advanced Machine Intelligence) is a Paris-based artificial intelligence research company co-founded by Yann LeCun, the Turing Award winner and former Chief AI Scientist at Meta, to build world models rather than large language models. On March 10, 2026 the company announced a seed round of $1.03 billion (about 890 million euros) at a $3.5 billion pre-money valuation, which Crunchbase News described as the largest seed round ever for a European startup. [1][3][9] Its name stands for Advanced Machine Intelligence, is pronounced like the French word for "friend," and reflects the belief stated on its website that "real intelligence does not start in language. It starts in the world." [2][10] As of October 2026 the company had about 60 employees, had published research papers, and had not yet released a product. [14][18][19]
What is AMI Labs?
AMI Labs is a research-first AI lab built around Yann LeCun's argument that the path to genuinely capable machine intelligence runs through world models learned from sensory data rather than through ever-larger text-trained chatbots. The name stands for Advanced Machine Intelligence and is pronounced like the French word for "friend." [2] The lab emerged publicly in early 2026 and announced its roughly $1.03 billion seed round in March 2026. [1][3] Its stated goal is to build AI systems that "(1) understand the real world, (2) have persistent memory, (3) can reason and plan, (4) are controllable and safe," using world models that "learn abstract representations of real-world sensor data, ignoring unpredictable details, and that make predictions in representation space." [10] The approach is grounded in LeCun's Joint Embedding Predictive Architecture (JEPA), which he positions as an alternative to the large language model paradigm that dominates the rest of the frontier-AI field. [1][4]
When was AMI Labs founded?
AMI grew directly out of LeCun's long-running disagreement with the direction of mainstream AI research. LeCun was named head of Facebook's new artificial intelligence laboratory in December 2013 and spent twelve years at the company, seven of them as chief AI scientist. [14][23] In the months before his exit, Meta had reorganized its AI work around a new unit, Meta Superintelligence Labs, and LeCun's long-term research at FAIR had been overshadowed by that overhaul, according to TechCrunch. [25] In November 2025 LeCun told Mark Zuckerberg he was leaving, and he departed at the end of 2025. [4][15]
The French company behind AMI was registered in Paris on December 15, 2025. It was incorporated under the name PengyouCo and renamed Advanced Machine Intelligence by a shareholder decision of February 12, 2026, according to legal notices in the French company register. [13] LeCun confirmed the venture publicly on December 18, 2025, writing on LinkedIn: "Yes, AMI Labs is my new startup. I'm the Executive Chairman." [11] The company's website went up in January 2026, and it made its formal launch alongside the funding announcement on March 10, 2026. [2][9][27]
LeCun, who shared the 2018 ACM A.M. Turing Award with Geoffrey Hinton and Yoshua Bengio for "conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing," is known in particular for his work on the convolutional neural network. [22][23] He serves as executive chairman of AMI rather than chief executive. [11][21] He has kept his professorship at New York University, where he teaches one class per year and supervises PhD students and postdocs, and he holds dual French-American citizenship. [2][4] Day-to-day operations are led by chief executive Alexandre LeBrun. [4]
Who founded AMI Labs and who leads it?
The founding team draws heavily from LeCun's former colleagues at Meta and from academia. The company is led by its co-founder and CEO, Alexandre LeBrun, [16] a French entrepreneur who co-founded and ran the clinical AI company Nabla. His earlier startup, Wit.ai, which offered an API for building voice and natural-language interfaces, was acquired by Facebook in January 2015. After the Wit.ai acquisition, LeBrun worked under LeCun's leadership at Meta's FAIR lab. [2][4][26] His move to AMI was announced on December 18, 2025 as part of an exclusive partnership in which Nabla gained "first access" to AMI's world model technology; LeBrun stepped down as Nabla's CEO and became its chief AI scientist and chairman. LeCun had been a founding investor in and long-time advisor to Nabla. [12]
| Role | Person | Background |
|---|---|---|
| Executive chairman | Yann LeCun | Former Chief AI Scientist, Meta; professor, NYU; 2018 Turing Award [21][22] |
| Chief executive officer | Alexandre LeBrun | Founder of Wit.ai (acquired by Facebook); co-founder of Nabla; former Meta FAIR [2][12] |
| Chief science officer | Saining Xie | Co-founder; NYU computer science faculty; formerly Google DeepMind and Facebook AI Research [1][20] |
| Chief research and innovation officer | Pascale Fung | Former senior director of AI research at Meta [1][4] |
| VP of world models | Michael Rabbat | Former director of research science at Meta; based in Montreal [1][4] |
| Chief operating officer | Laurent Solly | Former Meta vice president for Europe [1][2] |
Solly also became the company's legal president (president of the French SAS) under a shareholder decision of February 12, 2026, according to company register notices. [13] The company's website additionally lists Min Lin among its team. [10]
What does AMI Labs work on?
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, and The Next Web summarized his view as holding that they are "a statistical illusion. Impressive, yes. Intelligent, no." [4][5]
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. The company argues that "much of real-world sensor data is unpredictable, and generative approaches do not work well," and that "action-conditioned world models allow agentic systems to predict the consequences of their actions, and to plan action sequences to accomplish a task, subject to safety guardrails." [10] 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. [4][7] Before AMI, LeCun and Rabbat were among the authors of the 2024 paper introducing V-JEPA, a family of JEPA vision models trained on video with a feature prediction objective. [24]
At launch, LeCun told The Straits Times that AMI already had prototypes with "some level of common sense": shown a video in which "something impossible occurs, like someone throws a ball and the ball turns into a cube or disappears, the system tells us this is impossible." [15] In an October 2026 interview with Fortune, he described the company's primary near-term focus as industrial applications, "systems that understand the real world, like a manufacturing plant or turbojet engine," with anomaly detection and robotics among the main uses. [14]
LeBrun has been candid that the term may get diluted quickly. He predicted that "world models" would become the next industry buzzword, saying "In six months, every company will call itself a world model to raise funding." [1]
Research output
AMI has said it will publish its research as it goes. [1] By October 2026, papers listing the company as an author affiliation had appeared on arXiv, including the two below. [18][19]
| Paper | arXiv submission | Authors | What it reports |
|---|---|---|---|
| "LpWM: A Case for Sparse Representations in World Models" | August 24, 2026 | Yilun Kuang, Yash Dagade, Quentin Le Lidec, Lucas Maes, Randall Balestriero, Yann LeCun | A JEPA world model regularized toward sparse, non-negative codes; on the PushT task it outperforms the dense LeWorldModel baseline by up to 57% in planning success at intermediate predictor sizes [18] |
| "H-JEPA: End-to-End Learning of Hierarchical World Models for Visual Planning" | October 5, 2026 | Wancong Zhang, Basile Terver, Michael Rabbat, Yann LeCun, Randall Balestriero | A hierarchy of action-conditioned JEPAs in which each level predicts farther ahead; on Visual AntMaze a three-level hierarchy raises planning success from 18% to 73% while using less planner compute [19] |
In the LpWM paper, all authors except one list AMI Labs alongside a university (NYU, Mila or Brown University), and the acknowledgements say the work was supported through AMI Labs. [18] In the H-JEPA paper, all five authors list Advanced Machine Intelligence as an affiliation, several jointly with NYU, INRIA Paris or Brown University, and the paper links to public code. [19] Both papers evaluate mainly on simulated planning benchmarks; H-JEPA also reports offline planning results on real-robot videos from the DROID dataset. [18][19]
How much funding did AMI Labs raise, and who invested?
AMI's seed round raised $1.03 billion, roughly 890 million euros. [9] TechCrunch, AIwire and Futurum reported the valuation as $3.5 billion pre-money, and Crunchbase News reported that the round "reportedly values AMI at $3.5 billion" without specifying a basis. [1][3][5][6] Crunchbase News called it the largest seed round ever for a European startup, and LeCun himself described it as "one the largest seeds ever, probably the largest for a European company." [3][27]
The company had set out to raise much less. In December 2025, TechCrunch, citing the Financial Times, reported that AMI was seeking about 500 million euros at a valuation of about 3 billion euros; in January 2026 TechCrunch, citing Bloomberg, reported that Cathay Innovation, Greycroft and Hiro Capital were among the venture firms in talks with the startup. [2][11] LeBrun said the high level of interest let AMI choose its investors. [1][4]
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][9] The company named Toyota Ventures, New Legacy Ventures, Temasek, SBVA, Nvidia, Mark Cuban, Association Familiale Mulliez, Groupe Industriel Marcel Dassault, Sea and Alpha Intelligence Capital among its long-term and strategic backers. [9] It listed further participation from Eric Schmidt, Aglaé Lab, ZEBOX Ventures, Artémis, Xavier Niel, Publicis Groupe, Samsung, Bpifrance Digital Venture, Jim Breyer, Tim and Rosemary Berners-Lee, and Mark Leslie. [9] The Abu Dhabi investment firm Shorooq separately announced that it had joined the round. [17]
| Funding detail | Reported figure |
|---|---|
| Round type | Seed [1][9] |
| Amount raised | $1.03 billion (about 890 million euros) [9] |
| Valuation | $3.5 billion pre-money [1][5][6]; Crunchbase News: "$3.5 billion," basis not stated [3] |
| Announcement | March 10, 2026 [9][27] |
| Co-lead investors | Cathay Innovation, Greycroft, Hiro Capital, HV Capital, Bezos Expeditions [1][9] |
| Initial target (reported) | About 500 million euros at about 3 billion euros, per the Financial Times via TechCrunch [11] |
Where is AMI Labs based?
AMI is headquartered in Paris. [1][2] LeCun told the MIT Technology Review that AMI "is going to be a global company [that's] headquartered in Paris," and the choice was welcomed by French President Emmanuel Macron. [2] The company says it has operated in Paris, New York, Montreal and Singapore "from day one." [9] TechCrunch reported the reasoning behind each site: New York because LeCun teaches at NYU, Montreal because Rabbat is based there, and Singapore both to recruit AI talent and to be close to future clients in Asia. [1] The Singapore office had four staff at launch, with plans to grow to 20 over the following years, according to The Straits Times. [15]
The company has grown quickly from a small founding team. Futurum described it as having "roughly a dozen employees" at the time of the seed round, and Fortune reported in October 2026 that its "now 60-some employees" were building world models. [6][14]
What is AMI Labs' commercial strategy?
The lab has framed itself as a research-first organization rather than a product company. LeBrun told TechCrunch that AMI "starts with fundamental research" and is "not your typical applied AI startup," and that it had no plans to generate revenue for the time being; LeCun told journalists the first year would be devoted to research and development. [1][4] Over a longer horizon LeCun told AFP that within three to five years the goal is to produce "fairly universal intelligent systems" that could be deployed across almost any domain. [4] The company lists industrial process control, automation, wearable devices, robotics and healthcare as target sectors, and Shorooq's announcement named manufacturing, aerospace, robotics and biomedical industries as initial enterprise targets. [10][17] In October 2026 LeCun told Fortune that the company's first product would debut "soon," and that open models might be involved. [14]
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, which LeBrun co-founded and now chairs, and which is expected to gain early access to AMI's models. [1][12] TechCrunch reported that AMI plans to license its technology to industry partners. [2]
In October 2026 the Abu Dhabi technology group G42 and its portfolio company Analog, whose founder and CEO is Alex Kipman, signed memoranda of understanding with AMI on the sidelines of the UAE-France Business Dialogue in Abu Dhabi, according to a report published on October 20. The non-binding agreements span frontier AI research, advanced compute infrastructure, "sovereign enabled cloud" and potential industrial applications; scope, commercial terms and implementation are left to separate definitive agreements. [16]
The company has also said it intends to publish research and open-source at least some of its code, with LeBrun saying "We will also make a lot of code open source," consistent with the open-research culture LeCun championed at FAIR. [1] LeCun told The Straits Times: "We plan to publish our research and open-source models, to really create a community." [15]
How does AMI Labs compare to other 2026 frontier labs?
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 systems runs through world models learned from the physical world instead. [1][4] TechCrunch described Fei-Fei Li's World Labs as a direct rival in the world-model field. [2] LeCun has also positioned AMI as a European lab, saying "We are one of the few frontier AI labs that are neither Chinese nor American." [4]
Commentary at launch was cautious about how quickly the bet could pay off. The Next Web wrote that "AMI has no product, no revenue, and no near-term prospect of either," and that whether the ambition is achievable "remains genuinely open." [4] Futurum analyst Nick Patience called the critique of LLMs "well grounded," but argued that the claim that world models are a categorical solution to LLM failure modes "is more contested," and that the path from laboratory JEPA demonstrations to reliable deployment in healthcare or robotics "is non-trivial." [6]
Continue reading
- List of neolabsThis list of neolabs covers the 106 companies on "The Ultimate Neolab List", a list kept by Debarghya "Deedy" Das, a partner at the venture capital firm Menlo Ventures.Listed in See also
- World modelA world model is an artificial intelligence system that learns an internal representation of how an environment works, enabling it to predict future states, simulate the consequences of actions, and support…Mentioned 16 times above
- Meta AIMeta AI is the name Meta Platforms uses for two related but distinct things: the company's artificial intelligence research and engineering organizationMentioned 10 times above
- Yann LeCunYann Andre LeCun (born July 8, 1960) is a French-American computer scientist, a pioneer of convolutional neural networks (CNNs) and deep learning, and a 2018 ACM A.M. Turing Award laureate often called one of…Mentioned 5 times above
- Large Language ModelA large language model (LLM) is an artificial intelligence system built on a transformer neural network with billions to trillions of parameters, trained on massive text corpora to predict the next token in a…Mentioned 3 times above
- Artificial IntelligenceArtificial intelligence (AI) is the field of computer science dedicated to creating systems that perform tasks normally requiring human intelligence, such as learning, reasoning, recognizing patterns…Mentioned twice above
More to read
References
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20 ^21Heim, Anna. "Yann LeCun's AMI Labs raises $1.03B to build world models." TechCrunch, March 9, 2026. techcrunch.com/...03-billion-to-build-world-models
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12Heim, Anna. "Who's behind AMI Labs, Yann LeCun's 'world model' startup." TechCrunch, January 23, 2026. techcrunch.com/...-yann-lecuns-world-model-startup
- ^1 ^2 ^3 ^4 ^5van Romburgh, Marlize. "Turing Winner LeCun's New 'World Model' AI Lab Raises $1B In Europe's Largest Seed Round Ever." Crunchbase News, March 10, 2026. news.crunchbase.com/...-europes-largest-seed-round
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15Stanciuc, Ana-Maria. "Yann LeCun just raised $1bn to prove the AI industry has got it wrong." The Next Web, March 10, 2026. thenextweb.com/...un-ami-labs-world-models-billion
- ^1 ^2 ^3Hampton, Jaime. "Yann LeCun's AMI Secures $1B Seed to Develop AI World Models." AIwire / HPCwire, March 11, 2026. hpcwire.com/...-1b-seed-to-develop-ai-world-models
- ^1 ^2 ^3 ^4Patience, Nick. "Yann LeCun's AMI Raises $1BN Seed Round: Is the World Model Era Finally Here?" Futurum Group, March 13, 2026. futurumgroup.com/...e-world-model-era-finally-here
- ^1 ^2 ^3 ^4LeCun, Yann. "A Path Towards Autonomous Machine Intelligence." OpenReview, June 27, 2022. openreview.net/forum
- ^1 ^2Dawid, Anna; LeCun, Yann. "Introduction to Latent Variable Energy-Based Models: A Path Towards Autonomous Machine Intelligence." arXiv:2306.02572, 2023. arxiv.org/...2306.02572
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11Advanced Machine Intelligence. "Official launch." AMI Labs Updates, March 10, 2026. amilabs.xyz/updates
- ^1 ^2 ^3 ^4 ^5Advanced Machine Intelligence. "AMI Labs: Real World. Real Intelligence." Company website, accessed October 27, 2026. amilabs.xyz
- ^1 ^2 ^3 ^4Bort, Julie. "Yann LeCun confirms his new 'world model' startup, reportedly seeks $5B+ valuation." TechCrunch, December 19, 2025. techcrunch.com/...up-reportedly-seeks-5b-valuation
- ^1 ^2 ^3"Nabla Announces Exclusive Partnership With Advanced Machine Intelligence to Pioneer the Next Era of Agentic Healthcare AI." Nabla press release, December 18, 2025. nabla.com/...the-next-era-of-agentic-healthcare-ai
- ^1 ^2"ADVANCED MACHINE INTELLIGENCE (SIREN 994675254)." Societe.com company register entry and legal notices, accessed October 27, 2026. societe.com/...nced-machine-intelligence-994675254
- ^1 ^2 ^3 ^4 ^5Forlini, Emily. "AI 'godfather' Yann LeCun has 'zero concerns' about human extinction, says Anthropic CEO Dario Amodei is 'deluded'." Fortune, October 1, 2026. fortune.com/...nthropic-ceo-dario-amodei-is-deuded
- ^1 ^2 ^3 ^4"AI 'godfather' raises S$1.3 billion for startup with Singapore as a key base." The Straits Times, March 10, 2026, republished by the Singapore Economic Development Board. edb.gov.sg/...startup-with-singapore-as-a-key-base
- ^1 ^2"G42 and Yann LeCun's AMI Labs sign deals to explore collaboration in AI for physical world." Zawya, via TradingView News, October 20, 2026. tradingview.com/...ration-in-ai-for-physical-world
- ^1 ^2"Shorooq invests in AMI Labs as Yann LeCun's AI startup raises $1.03bln." Shorooq press release via Zawya, March 10, 2026. zawya.com/...uns-ai-startup-raises-103bln-klbupz6n
- ^1 ^2 ^3 ^4 ^5Kuang, Yilun; Dagade, Yash; Le Lidec, Quentin; Maes, Lucas; Balestriero, Randall; LeCun, Yann. "LpWM: A Case for Sparse Representations in World Models." arXiv:2608.22764, August 24, 2026. arxiv.org/...2608.22764
- ^1 ^2 ^3 ^4 ^5Zhang, Wancong; Terver, Basile; Rabbat, Michael; LeCun, Yann; Balestriero, Randall. "H-JEPA: End-to-End Learning of Hierarchical World Models for Visual Planning." arXiv:2610.06805, October 5, 2026. arxiv.org/...2610.06805
- ^Xie, Saining. Personal homepage, accessed October 27, 2026. sainingxie.com
- ^1 ^2LeCun, Yann. Personal homepage, accessed October 27, 2026. yann.lecun.com
- ^1 ^2"Fathers of the Deep Learning Revolution Receive ACM A.M. Turing Award." Association for Computing Machinery. awards.acm.org/...2018-turing
- ^1 ^2Constine, Josh. "NYU 'Deep Learning' Professor LeCun Will Head Facebook's New Artificial Intelligence Lab." TechCrunch, December 9, 2013. techcrunch.com/...rtificial-intelligence-lab-lecun
- ^Bardes, Adrien; Garrido, Quentin; Ponce, Jean; Chen, Xinlei; Rabbat, Michael; LeCun, Yann; Assran, Mahmoud; Ballas, Nicolas. "Revisiting Feature Prediction for Learning Visual Representations from Video." arXiv:2404.08471, 2024. arxiv.org/...2404.08471
- ^Iyer, Ram. "Meta's chief AI scientist Yann LeCun reportedly plans to leave to build his own startup." TechCrunch, November 11, 2025. techcrunch.com/...o-leave-to-build-his-own-startup
- ^Constine, Josh. "Facebook Acquires Wit.ai To Help Its Developers With Speech Recognition And Voice Interfaces." TechCrunch, January 5, 2015. techcrunch.com/...facebook-wit-ai
- ^1 ^2 ^3LeCun, Yann (@ylecun). "Unveiling our new startup Advanced Machine Intelligence (AMI Labs)." X, March 10, 2026. x.com/...2031268686984527936
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation. Every suggestion is reviewed for sourcing before it goes live.
6 revisions · v6 · 3,277 words · full history
Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify
Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here
Reviewer note: Founding team, $1.03 billion seed round and investors, leadership and 2026 research and partnership developments checked against TechCrunch, The Next Web, Crunchbase News, AIwire and company announcements, October 2026.
Cite this page: AI Wiki. "AMI Labs." aiwiki.ai, updated 28 Oct 2026, fact-checked 27 Oct 2026. CC BY 4.0. https://aiwiki.ai/wiki/ami_labs