[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-post-\u002Fnews\u002Fyann-lecun-ami-labs-890-million-funding":3,"news-related-en-\u002Fnews\u002Fyann-lecun-ami-labs-890-million-funding":337,"hreflang-yann-lecun-ami-labs":357},{"doc":4,"redirect":8},{"id":5,"title":6,"author":7,"author_avatar":8,"body":9,"category":316,"date":317,"description":318,"extension":319,"image":320,"meta":321,"navigation":322,"path":323,"seo":324,"seoDescription":325,"seoTitle":6,"stem":326,"tags":327,"translationKey":335,"updatedAt":8,"__hash__":336},"news_en\u002Fnews\u002Fyann-lecun-ami-labs-890-million-funding.md","Yann LeCun Launches AMI and Raises €890M to Reinvent AI from Paris","Paul Lefizelier",null,{"type":10,"value":11,"toc":308},"minimark",[12,29,34,45,64,68,75,90,103,106,110,113,175,178,182,189,192,204,207,212,295,297],[13,14,15,16,20,21,24,25,28],"p",{},"The Turing Award winner and former head of AI research at Meta just made a powerful move. On March 10, 2026, ",[17,18,19],"strong",{},"Yann LeCun"," announces the creation of ",[17,22,23],{},"AMI"," (Advanced Machine Intelligence), a Paris-based startup, and closes a first round of ",[17,26,27],{},"€890 million"," (approximately $1.03 billion). Valued at €3 billion pre-money, AMI instantly becomes one of Europe's best-funded AI startups — and sends a powerful signal about French ambition in the global AI race.",[30,31,33],"h2",{"id":32},"a-historic-raise-for-french-tech","A Historic Raise for French Tech",[13,35,36,37,40,41,44],{},"The amount places AMI among the largest raises in French tech history. For reference, ",[17,38,39],{},"Mistral AI"," had closed a €1.7 billion round, but across multiple successive rounds. AMI reaches €890 million ",[17,42,43],{},"in its very first round"," — unprecedented for a European company with no revenue.",[13,46,47,48,51,52,55,56,59,60,63],{},"The founding team explains this investor confidence. Around LeCun, who serves as board chairman, you'll find ",[17,49,50],{},"Alexandre LeBrun"," as CEO (serial entrepreneur, former founder of Wit.ai acquired by Facebook), ",[17,53,54],{},"Saining Xie"," as scientific director, ",[17,57,58],{},"Pascale Fung"," heading research, and ",[17,61,62],{},"Michael Rabbat"," in charge of \"world representations.\" Four offices are already open: Paris (HQ), New York, Singapore, and Montreal.",[30,65,67],{"id":66},"world-models-ai-that-truly-understands-the-world","World Models: AI That Truly Understands the World",[13,69,70,71,74],{},"AMI isn't building yet another chatbot. The startup develops ",[17,72,73],{},"world models"," — AI systems capable of understanding the physical world, not just manipulating text.",[13,76,77,78,81,82,85,86,89],{},"The key architecture is called ",[17,79,80],{},"JEPA"," (Joint Embedding Predictive Architecture), a theoretical framework LeCun has been working on for a decade at Meta. To understand the difference from current LLMs, a simple analogy: an LLM like GPT or Claude is a student who has ",[17,83,84],{},"memorized the map"," — they can describe a route, but they don't understand geography. A world model is a student who ",[17,87,88],{},"understands the terrain"," — they can anticipate that a river blocks the path even if they've never seen it on the map.",[13,91,92,93,98,99,102],{},"LeCun doesn't mince words about the limitations of current LLMs. According to him, language models — whether ",[94,95,97],"a",{"href":96},"\u002Fcompare\u002Fclaude-vs-chatgpt","GPT or Claude"," — are \"incapable of planning, reasoning, or understanding causality.\" They predict the next word brilliantly, but build no internal representation of the world. JEPA aims to bridge this gap by learning ",[17,100,101],{},"abstract representations"," of reality, usable for planning, reasoning, and action in the physical world.",[13,104,105],{},"AMI has no revenue yet and likely won't for several years. It's a long-term R&D bet — exactly the kind that only patient investors convinced by the scientific vision can support.",[30,107,109],{"id":108},"a-global-round-a-deliberately-french-anchoring","A Global Round, a Deliberately French Anchoring",[13,111,112],{},"The investor list reads like a who's who of global tech:",[114,115,116,130,140,157],"ul",{},[117,118,119,122,123,122,126,129],"li",{},[17,120,121],{},"Nvidia",", ",[17,124,125],{},"Toyota",[17,127,128],{},"Samsung"," — three industrial players betting on AI applied to the physical world (robotics, automotive, electronics)",[117,131,132,135,136,139],{},[17,133,134],{},"Jeff Bezos"," (personally) and ",[17,137,138],{},"Eric Schmidt"," (former Google CEO) — two figures who are rarely wrong when they write checks",[117,141,142,143,122,146,149,150,153,154],{},"French side: ",[17,144,145],{},"Xavier Niel",[17,147,148],{},"Dassault",", the ",[17,151,152],{},"Mulliez"," family, ",[17,155,156],{},"Cathay Innovation",[117,158,159,160,122,163,166,167],{},"Complemented by ",[17,161,162],{},"Hiro Capital",[17,164,165],{},"Greycroft",", and ",[17,168,169],{},[94,170,174],{"href":171,"rel":172},"https:\u002F\u002Fcloverfund.vc",[173],"nofollow","Clover Fund",[13,176,177],{},"Alexandre LeBrun emphasizes this \"geographic balance\" of the investor base. The goal: anchor AMI in Europe while ensuring access to American and Asian markets and talent. A pragmatic positioning, far from technological nationalism, but clearly designed so that the value created benefits the French ecosystem first.",[30,179,181],{"id":180},"macron-and-the-state-the-sovereign-ai-bet","Macron and the State: The Sovereign AI Bet",[13,183,184,185,188],{},"The political reaction was swift. On X, ",[17,186,187],{},"Emmanuel Macron"," hailed the announcement: \"Yann LeCun opens a new era for artificial intelligence. This is the France of researchers, builders, and bold thinkers. Bravo!\"",[13,190,191],{},"Beyond the tweet, AMI fits into France's national AI strategy and the \"AI Pioneers\" program launched by the Élysée. With Mistral on LLMs and now AMI on world models, France has two champions on two complementary approaches to artificial intelligence — a unique positioning against the American OpenAI\u002FAnthropic duopoly.",[13,193,194,195,199,200,203],{},"The open question remains: can France compete durably when OpenAI raises $110 billion and Anthropic $30 billion? The ",[94,196,198],{"href":197},"\u002Fblog\u002Fopen-source-vs-proprietary-ai-future-developer-tools","open source vs proprietary AI"," debate is more relevant than ever. AMI's bet is that the race won't be won by stacking GPUs, but by finding the ",[17,201,202],{},"right architecture"," — and that JEPA could be that architecture.",[205,206],"hr",{},[13,208,209],{},[17,210,211],{},"In Brief: AMI vs the Giants",[213,214,215,237],"table",{},[216,217,218],"thead",{},[219,220,221,224,228,233],"tr",{},[222,223],"th",{},[222,225,226],{},[17,227,23],{},[222,229,230],{},[17,231,232],{},"OpenAI",[222,234,235],{},[17,236,39],{},[238,239,240,254,268,281],"tbody",{},[219,241,242,246,249,252],{},[243,244,245],"td",{},"Nationality",[243,247,248],{},"France",[243,250,251],{},"United States",[243,253,248],{},[219,255,256,259,262,265],{},[243,257,258],{},"Approach",[243,260,261],{},"World models (JEPA)",[243,263,264],{},"LLMs (GPT)",[243,266,267],{},"LLMs open-weight",[219,269,270,273,276,279],{},[243,271,272],{},"Stage",[243,274,275],{},"R&D, pre-revenue",[243,277,278],{},"Products in production",[243,280,278],{},[219,282,283,286,289,292],{},[243,284,285],{},"Valuation",[243,287,288],{},"€3B",[243,290,291],{},"~$800B",[243,293,294],{},"~€6B",[205,296],{},[13,298,299],{},[300,301,302,303,307],"em",{},"Next expected steps: publication of first JEPA research results applied to robotics and video, and a hiring wave — AMI targets 200 researchers by end of 2026, with Paris as the main hub. This raise is part of the ",[94,304,306],{"href":305},"\u002Fnews\u002Fai-funding-explosion-openai-anthropic-record-2026","record AI funding wave"," redefining global venture capital.",{"title":309,"searchDepth":310,"depth":310,"links":311},"",2,[312,313,314,315],{"id":32,"depth":310,"text":33},{"id":66,"depth":310,"text":67},{"id":108,"depth":310,"text":109},{"id":180,"depth":310,"text":181},"ai","2026-03-10","Turing Award winner Yann LeCun founds AMI (Advanced Machine Intelligence) in Paris and closes a first round of €890M with Nvidia, Toyota, Samsung and Jeff Bezos. Goal: surpass LLMs with world models.","md","\u002Fami-labs-yann-lecun.jpg",{},true,"\u002Fnews\u002Fyann-lecun-ami-labs-890-million-funding",{"title":6,"description":318},"AMI, the AI startup founded by Yann LeCun in Paris, raises €890M from Nvidia, Toyota, Samsung and Jeff Bezos to develop world models capable of understanding the physical world.","news\u002Fyann-lecun-ami-labs-890-million-funding",[328,329,330,331,332,333,334],"yann-lecun","ami","fundraising","french-tech","world-models","jepa","nvidia","yann-lecun-ami-labs","B4f6Yjdl2SVIV-sasiQnYdV1fR9L7c2iveF6dczchJo",[338,344,351],{"title":339,"description":340,"path":341,"date":342,"image":343,"category":316},"Meta pauses its new ad and business terms after the AI scraping backlash","Meta has indefinitely paused updates to four sets of commercial and advertising terms that publishers said would let it crawl their sites to build AI and ad products.","\u002Fnews\u002Fmeta-pauses-business-ad-terms-ai-scraping","2026-11-09T08:00:00","\u002Fimages\u002Fnews\u002Fmeta-pauses-business-ad-terms-ai-scraping.jpg",{"title":345,"description":346,"path":347,"date":348,"image":349,"category":350},"Claude Code's “don’t ask again” left passwords in 1,122 public GitHub files","Security firm Apiiro found 1,122 public copies of Claude Code’s local settings file holding secrets, after “Yes, and don’t ask again” saved approved commands in plain text.","\u002Fnews\u002Fclaude-code-dont-ask-again-passwords-github","2026-11-08T08:00:00","\u002Fimages\u002Fnews\u002Fclaude-code-dont-ask-again-passwords-github.jpg","vibecoding",{"title":352,"description":353,"path":354,"date":355,"image":356,"category":350},"Cursor cloud agents now build, host and publish live apps","Cursor's cloud agents can scaffold a web app, provision Postgres and Cursor SSO, run security checks, then publish a shareable URL — without a separate hosting account.","\u002Fnews\u002Fcursor-live-apps-cloud-agents-deploy-postgres-auth","2026-11-07T08:00:00","\u002Fimages\u002Fnews\u002Fcursor-live-apps-cloud-agents-deploy-postgres-auth.jpg",{"en":358,"fr":551,"es":8},{"id":5,"title":6,"author":7,"author_avatar":8,"body":359,"category":316,"date":317,"description":318,"extension":319,"image":320,"meta":548,"navigation":322,"path":323,"seo":549,"seoDescription":325,"seoTitle":6,"stem":326,"tags":550,"translationKey":335,"updatedAt":8,"__hash__":336},{"type":10,"value":360,"toc":542},[361,369,371,377,387,389,393,401,407,409,411,413,450,452,454,458,460,466,468,472,534,536],[13,362,15,363,20,365,24,367,28],{},[17,364,19],{},[17,366,23],{},[17,368,27],{},[30,370,33],{"id":32},[13,372,36,373,40,375,44],{},[17,374,39],{},[17,376,43],{},[13,378,47,379,51,381,55,383,59,385,63],{},[17,380,50],{},[17,382,54],{},[17,384,58],{},[17,386,62],{},[30,388,67],{"id":66},[13,390,70,391,74],{},[17,392,73],{},[13,394,77,395,81,397,85,399,89],{},[17,396,80],{},[17,398,84],{},[17,400,88],{},[13,402,92,403,98,405,102],{},[94,404,97],{"href":96},[17,406,101],{},[13,408,105],{},[30,410,109],{"id":108},[13,412,112],{},[114,414,415,423,429,439],{},[117,416,417,122,419,122,421,129],{},[17,418,121],{},[17,420,125],{},[17,422,128],{},[117,424,425,135,427,139],{},[17,426,134],{},[17,428,138],{},[117,430,142,431,122,433,149,435,153,437],{},[17,432,145],{},[17,434,148],{},[17,436,152],{},[17,438,156],{},[117,440,159,441,122,443,166,445],{},[17,442,162],{},[17,444,165],{},[17,446,447],{},[94,448,174],{"href":171,"rel":449},[173],[13,451,177],{},[30,453,181],{"id":180},[13,455,184,456,188],{},[17,457,187],{},[13,459,191],{},[13,461,194,462,199,464,203],{},[94,463,198],{"href":197},[17,465,202],{},[205,467],{},[13,469,470],{},[17,471,211],{},[213,473,474,492],{},[216,475,476],{},[219,477,478,480,484,488],{},[222,479],{},[222,481,482],{},[17,483,23],{},[222,485,486],{},[17,487,232],{},[222,489,490],{},[17,491,39],{},[238,493,494,504,514,524],{},[219,495,496,498,500,502],{},[243,497,245],{},[243,499,248],{},[243,501,251],{},[243,503,248],{},[219,505,506,508,510,512],{},[243,507,258],{},[243,509,261],{},[243,511,264],{},[243,513,267],{},[219,515,516,518,520,522],{},[243,517,272],{},[243,519,275],{},[243,521,278],{},[243,523,278],{},[219,525,526,528,530,532],{},[243,527,285],{},[243,529,288],{},[243,531,291],{},[243,533,294],{},[205,535],{},[13,537,538],{},[300,539,302,540,307],{},[94,541,306],{"href":305},{"title":309,"searchDepth":310,"depth":310,"links":543},[544,545,546,547],{"id":32,"depth":310,"text":33},{"id":66,"depth":310,"text":67},{"id":108,"depth":310,"text":109},{"id":180,"depth":310,"text":181},{},{"title":6,"description":318},[328,329,330,331,332,333,334],{"id":552,"title":553,"author":7,"author_avatar":8,"body":554,"category":316,"date":317,"description":813,"extension":319,"image":320,"meta":814,"navigation":322,"path":815,"seo":816,"seoDescription":817,"seoTitle":553,"stem":818,"tags":819,"translationKey":335,"updatedAt":8,"__hash__":821},"news_fr\u002Ffr\u002Fnews\u002Fyann-lecun-ami-labs-levee-890-millions.md","Yann LeCun lance AMI et lève 890 M€ pour réinventer l'IA depuis Paris",{"type":10,"value":555,"toc":807},[556,569,573,583,598,602,608,622,634,637,641,644,688,691,695,701,704,716,718,723,795,797],[13,557,558,559,561,562,564,565,568],{},"Le prix Turing et ancien directeur de la recherche IA chez Meta vient de frapper fort. Ce 10 mars 2026, ",[17,560,19],{}," annonce la création d'",[17,563,23],{}," (Advanced Machine Intelligence), une startup basée à Paris, et boucle un premier tour de ",[17,566,567],{},"890 millions d'euros"," (environ 1,03 milliard de dollars). Valorisée 3 milliards d'euros avant l'opération, AMI devient instantanément l'une des startups IA les mieux financées d'Europe — et envoie un signal puissant sur l'ambition française dans la course mondiale à l'intelligence artificielle.",[30,570,572],{"id":571},"une-levée-historique-pour-la-french-tech","Une levée historique pour la French Tech",[13,574,575,576,578,579,582],{},"Le montant place AMI parmi les plus grosses levées de l'histoire de la tech française. Pour mémoire, ",[17,577,39],{}," avait bouclé un tour de 1,7 milliard d'euros, mais après plusieurs rounds successifs. AMI atteint les 890 millions d'euros ",[17,580,581],{},"dès son premier tour"," — un fait sans précédent pour une entreprise européenne sans revenus.",[13,584,585,586,588,589,591,592,594,595,597],{},"L'équipe fondatrice explique cette confiance des investisseurs. Autour de LeCun, président du conseil d'administration, on trouve ",[17,587,50],{}," au poste de CEO (serial entrepreneur, ex-fondateur de Wit.ai racheté par Facebook), ",[17,590,54],{}," comme directeur scientifique, ",[17,593,58],{}," à la direction de la recherche, et ",[17,596,62],{}," en charge des \"représentations du monde\". Quatre bureaux sont déjà ouverts : Paris (siège), New York, Singapour et Montréal.",[30,599,601],{"id":600},"les-world-models-lia-qui-comprend-vraiment-le-monde","Les world models : l'IA qui comprend vraiment le monde",[13,603,604,605,607],{},"AMI ne construit pas un énième chatbot. La startup développe des ",[17,606,73],{}," — des systèmes d'IA capables de comprendre le monde physique, pas seulement de manipuler du texte.",[13,609,610,611,613,614,617,618,621],{},"L'architecture clé s'appelle ",[17,612,80],{}," (Joint Embedding Predictive Architecture), un cadre théorique sur lequel LeCun travaille depuis une décennie chez Meta. Pour comprendre la différence avec les LLMs actuels, une analogie simple : un LLM comme GPT ou Claude, c'est un étudiant qui a ",[17,615,616],{},"mémorisé la carte"," — il peut décrire un itinéraire, mais il ne comprend pas la géographie. Un world model, c'est un étudiant qui ",[17,619,620],{},"comprend le terrain"," — il peut anticiper qu'une rivière bloque le chemin même s'il ne l'a jamais vu sur la carte.",[13,623,624,625,629,630,633],{},"LeCun ne mâche pas ses mots sur les limites des LLMs actuels. Selon lui, les modèles de langage — qu'il s'agisse de ",[94,626,628],{"href":627},"\u002Ffr\u002Fcompare\u002Fclaude-vs-chatgpt","GPT ou Claude"," — sont « incapables de planifier, de raisonner, de comprendre la causalité ». Ils prédisent le mot suivant avec brio, mais ne construisent aucune représentation interne du monde. JEPA vise à combler ce fossé en apprenant des ",[17,631,632],{},"représentations abstraites"," de la réalité, utilisables ensuite pour la planification, le raisonnement et l'action dans le monde physique.",[13,635,636],{},"AMI n'a pas encore de revenus et n'en aura probablement pas avant plusieurs années. C'est un pari de R&D long terme — exactement le type de bet que seuls des investisseurs patients et convaincus par la vision scientifique peuvent soutenir.",[30,638,640],{"id":639},"un-tour-de-table-mondial-un-ancrage-français-assumé","Un tour de table mondial, un ancrage français assumé",[13,642,643],{},"La liste des investisseurs ressemble à un who's who de la tech mondiale :",[114,645,646,655,663,675],{},[117,647,648,122,650,122,652,654],{},[17,649,121],{},[17,651,125],{},[17,653,128],{}," — trois industriels qui misent sur l'IA appliquée au monde physique (robotique, automobile, électronique)",[117,656,657,659,660,662],{},[17,658,134],{}," (à titre personnel) et ",[17,661,138],{}," (ex-PDG de Google) — deux figures qui ont rarement tort quand ils sortent le chéquier",[117,664,665,666,122,668,670,671,122,673],{},"Côté français : ",[17,667,145],{},[17,669,148],{},", la famille ",[17,672,152],{},[17,674,156],{},[117,676,677,678,122,680,682,683],{},"Complété par ",[17,679,162],{},[17,681,165],{}," et ",[17,684,685],{},[94,686,174],{"href":171,"rel":687},[173],[13,689,690],{},"Alexandre LeBrun insiste sur cet « équilibre géographique » du tour de table. L'objectif : ancrer AMI en Europe tout en s'assurant un accès aux marchés et aux talents américains et asiatiques. Un positionnement pragmatique, loin du nationalisme technologique, mais clairement pensé pour que la valeur créée bénéficie d'abord à l'écosystème français.",[30,692,694],{"id":693},"macron-et-létat-le-pari-de-la-souveraineté-ia","Macron et l'État : le pari de la souveraineté IA",[13,696,697,698,700],{},"La réaction politique n'a pas tardé. Sur X, ",[17,699,187],{}," a salué l'annonce : « Yann LeCun ouvre une nouvelle ère pour l'intelligence artificielle. C'est la France des chercheurs, des bâtisseurs et des audacieux. Bravo ! »",[13,702,703],{},"Au-delà du tweet, AMI s'inscrit dans la stratégie nationale IA française et le programme « Pionniers de l'IA » lancé par l'Élysée. Avec Mistral sur les LLMs et désormais AMI sur les world models, la France dispose de deux champions sur deux approches complémentaires de l'intelligence artificielle — un positionnement unique face au duopole américain OpenAI\u002FAnthropic.",[13,705,706,707,711,712,715],{},"La question reste ouverte : la France peut-elle rivaliser durablement quand OpenAI lève 110 milliards de dollars et Anthropic 30 milliards ? Le débat ",[94,708,710],{"href":709},"\u002Ffr\u002Fblog\u002Fopen-source-vs-ia-proprietaire-avenir-outils-developpeurs","open source vs IA propriétaire"," est plus que jamais d'actualité. Le pari d'AMI est que la course ne se gagnera pas en empilant des GPU, mais en trouvant la ",[17,713,714],{},"bonne architecture"," — et que JEPA pourrait être cette architecture.",[205,717],{},[13,719,720],{},[17,721,722],{},"En bref : AMI vs les géants",[213,724,725,743],{},[216,726,727],{},[219,728,729,731,735,739],{},[222,730],{},[222,732,733],{},[17,734,23],{},[222,736,737],{},[17,738,232],{},[222,740,741],{},[17,742,39],{},[238,744,745,757,768,781],{},[219,746,747,750,752,755],{},[243,748,749],{},"Nationalité",[243,751,248],{},[243,753,754],{},"États-Unis",[243,756,248],{},[219,758,759,762,764,766],{},[243,760,761],{},"Approche",[243,763,261],{},[243,765,264],{},[243,767,267],{},[219,769,770,773,776,779],{},[243,771,772],{},"Stade",[243,774,775],{},"R&D, pré-revenus",[243,777,778],{},"Produits en production",[243,780,778],{},[219,782,783,786,789,792],{},[243,784,785],{},"Valorisation",[243,787,788],{},"3 Md€",[243,790,791],{},"~800 Md$",[243,793,794],{},"~6 Md€",[205,796],{},[13,798,799],{},[300,800,801,802,806],{},"Prochaines étapes attendues : la publication des premiers résultats de recherche sur JEPA appliqué à la robotique et à la vidéo, et une vague de recrutements — AMI vise 200 chercheurs d'ici fin 2026, avec Paris comme principal hub. Cette levée s'inscrit dans la ",[94,803,805],{"href":804},"\u002Ffr\u002Fnews\u002Fexplosion-funding-ai-openai-anthropic-record-2026","vague de funding IA record"," qui redéfinit le capital-risque mondial.",{"title":309,"searchDepth":310,"depth":310,"links":808},[809,810,811,812],{"id":571,"depth":310,"text":572},{"id":600,"depth":310,"text":601},{"id":639,"depth":310,"text":640},{"id":693,"depth":310,"text":694},"Le prix Turing Yann LeCun fonde AMI (Advanced Machine Intelligence) à Paris et boucle un premier tour de 890 M€ avec Nvidia, Toyota, Samsung et Jeff Bezos. Objectif : dépasser les LLMs avec les world models.",{},"\u002Ffr\u002Fnews\u002Fyann-lecun-ami-labs-levee-890-millions",{"title":553,"description":813},"AMI, la startup IA fondée par Yann LeCun à Paris, lève 890 M€ auprès de Nvidia, Toyota, Samsung et Jeff Bezos pour développer des world models capables de comprendre le monde physique.","fr\u002Fnews\u002Fyann-lecun-ami-labs-levee-890-millions",[328,329,820,331,332,333,334],"levee-de-fonds","M-eTIHZFD9AEMjIbvbhoqSm9MWNHbYp--wsubz7VVtQ"]