[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-post-\u002Fnews\u002Fai-brain-drain-meta-google-openai-startups-188-billion-vc-april-2026":3,"news-related-en-\u002Fnews\u002Fai-brain-drain-meta-google-openai-startups-188-billion-vc-april-2026":430,"hreflang-ai-brain-drain-big-tech-startups-2026":450},{"doc":4,"redirect":8},{"id":5,"title":6,"author":7,"author_avatar":8,"body":9,"category":407,"date":408,"description":409,"extension":410,"image":411,"meta":412,"navigation":413,"path":414,"seo":415,"seoDescription":416,"seoTitle":417,"stem":418,"tags":419,"translationKey":428,"updatedAt":8,"__hash__":429},"news_en\u002Fnews\u002Fai-brain-drain-meta-google-openai-startups-188-billion-vc-april-2026.md","Big Tech AI brain drain: Meta, Google and OpenAI hemorrhage top talent — $18.8B already poured into startups founded since 2025","Paul Lefizelier",null,{"type":10,"value":11,"toc":380},"minimark",[12,25,30,45,92,99,103,108,121,125,140,144,154,158,168,172,182,186,197,201,204,208,217,229,233,236,240,247,251,255,262,266,278,282,289,293,296,322,326,338,345,349,352,359,367,370],[13,14,15,16,20,21,24],"p",{},"On April 28, 2026, ",[17,18,19],"strong",{},"CNBC"," finally said out loud what every Big Tech HR lead refused to admit publicly: ",[17,22,23],{},"Meta, Google and OpenAI are bleeding their most valuable talent",". And we're not talking about random engineers — these are principal researchers, division leads, foundational model architects. They're leaving. And they're raising amounts that make no sense for startups with no product.",[26,27,29],"h2",{"id":28},"the-number-that-changes-everything-188b-in-four-months","The number that changes everything: $18.8B in four months",[13,31,32,33,36,37,40,41,44],{},"According to ",[17,34,35],{},"Dealroom"," data cited by CNBC, ",[17,38,39],{},"$18.8 billion in venture capital"," has already been injected in 2026 into AI startups founded after January 1, 2025. At this pace, the year will far exceed the ",[17,42,43],{},"$27.9B"," raised in 2025 by companies founded after early 2024.",[46,47,48,61],"table",{},[49,50,51],"thead",{},[52,53,54,58],"tr",{},[55,56,57],"th",{},"Period",[55,59,60],{},"VC capital into new AI labs",[62,63,64,72,82],"tbody",{},[52,65,66,70],{},[67,68,69],"td",{},"2025 (cohort founded since 2024)",[67,71,43],{},[52,73,74,77],{},[67,75,76],{},"2026 (cohort founded since 2025)",[67,78,79],{},[17,80,81],{},"$18.8B in 4 months",[52,83,84,87],{},[67,85,86],{},"2026 annualized projection",[67,88,89],{},[17,90,91],{},"~$56B",[13,93,94,95,98],{},"Let that sink in. More venture capital is now flowing into AI startups founded by DeepMind, OpenAI and Meta FAIR alumni than into ",[17,96,97],{},"the entirety of European fintech series A\u002FB\u002FC funding"," over the same period.",[26,100,102],{"id":101},"the-departures-that-hurt","The departures that hurt",[104,105,107],"h3",{"id":106},"ineffable-intelligence-david-silver-deepmind-record-seed","Ineffable Intelligence — David Silver (DeepMind → record seed)",[13,109,110,111,114,115,120],{},"On April 27, ",[17,112,113],{},"David Silver",", former head of reinforcement learning at DeepMind, architect of AlphaGo and AlphaZero, came out of stealth with ",[116,117,119],"a",{"href":118},"\u002Fnews\u002Fineffable-intelligence-david-silver-1-1-billion-seed-deepmind-superlearner-april-2026","Ineffable Intelligence and a historic $1.1B seed"," at $5.1B valuation. Sequoia, Lightspeed, Nvidia, Google, and even the UK's Sovereign AI Fund are on the cap table. No product, no revenue, no public roadmap. Just a name on a slide.",[104,122,124],{"id":123},"ami-labs-yann-lecun-meta-1b","Ami Labs — Yann LeCun (Meta → $1B)",[13,126,127,128,131,132,135,136,139],{},"In March 2026, ",[17,129,130],{},"Yann LeCun"," left Meta after publicly disagreeing with the group's LLM-centric roadmap. His new entity, ",[17,133,134],{},"Ami Labs",", raised ",[17,137,138],{},"$1 billion"," immediately after to explore non-LLM architectures (energy-based models, world models). Mark Zuckerberg loses one of Meta's three Turing Award holders.",[104,141,143],{"id":142},"recursive-superintelligence-tim-rocktäschel-deepmind","Recursive Superintelligence — Tim Rocktäschel (DeepMind)",[13,145,146,149,150,153],{},[17,147,148],{},"Tim Rocktäschel",", ex-DeepMind, is reportedly seeking up to ",[17,151,152],{},"$1B"," for Recursive Superintelligence according to Sand Hill Road rumors. The thesis: recursive self-improvement, agents that modify their own code.",[104,155,157],{"id":156},"periodic-labs-openai-deepmind-alumni-300m","Periodic Labs — OpenAI + DeepMind alumni ($300M)",[13,159,160,163,164,167],{},[17,161,162],{},"Periodic Labs",", founded by ex-OpenAI and DeepMind staff, raised ",[17,165,166],{},"$300M"," in September 2025. Just months after launching.",[104,169,171],{"id":170},"humans-reinforcement-learning-from-real-experience-480m","Humans& — reinforcement learning from real experience ($480M)",[13,173,174,177,178,181],{},[17,175,176],{},"Humans&"," secured ",[17,179,180],{},"$480M"," in January 2026 to explore reinforcement learning from real-world experiences rather than internet data — a direct bet against the LLM-first doctrine of the major labs.",[104,183,185],{"id":184},"recursive-intelligence-335m-across-two-rounds","Recursive Intelligence — $335M across two rounds",[13,187,188,189,192,193,196],{},"Finally, ",[17,190,191],{},"Recursive Intelligence"," (not to be confused with Rocktäschel's startup) accumulated ",[17,194,195],{},"$335M"," across two rounds after launching in September 2025.",[26,198,200],{"id":199},"why-researchers-are-fleeing","Why researchers are fleeing",[13,202,203],{},"CNBC points to three structural factors that explain the exodus.",[104,205,207],{"id":206},"_1-commercial-pressure-kills-foundational-research","1. Commercial pressure kills foundational research",[209,210,211],"blockquote",{},[13,212,213],{},[214,215,216],"em",{},"\"The pressure to deliver benchmark performance and maintain rapid release cycles leaves limited room for genuinely exploratory research, particularly outside the dominant LLM paradigm.\"",[13,218,219,220,223,224,228],{},"When OpenAI has to justify its ",[17,221,222],{},"$500B valuation"," and Anthropic its ",[116,225,227],{"href":226},"\u002Fnews\u002Fgoogle-anthropic-40-billion-investment-two-phase-cloud-tpu-april-2026","$380B after the Google $40B deal",", agent research, interpretability, new architectures and vertical models take a back seat. Not because they don't matter — but because they don't win the quarterly race.",[104,230,232],{"id":231},"_2-equity-has-lost-its-magic","2. Equity has lost its magic",[13,234,235],{},"When Anthropic was at $40B, a senior researcher could reasonably target a 10x on RSUs. At $380B, the upside becomes marginal. Meanwhile, founding Ineffable at $5.1B with 30% equity = $1.5B on paper as of the seed. The math is brutal.",[104,237,239],{"id":238},"_3-alternative-architectures-are-back-in-fashion","3. Alternative architectures are back in fashion",[13,241,242,243,246],{},"For three years, the market bet exclusively on scale-out LLMs. The defectors are now betting on ",[17,244,245],{},"non-LLMs",": reinforcement learning from real experience, world models, energy-based models, neuro-symbolic. These architectures require research freedom that commercial labs can no longer offer.",[26,248,250],{"id":249},"what-this-means-for-founders","What this means for founders",[104,252,254],{"id":253},"for-normal-founders","For \"normal\" founders",[13,256,257,258,261],{},"If Sequoia signs a seed at $5.1B for an ex-DeepMind researcher with no product, ",[17,259,260],{},"your $5M seed for a working MVP is actually a good deal for VCs",". The rarity of \"real AI talent\" pushes top valuations into orbit — but the tier below remains accessible.",[104,263,265],{"id":264},"for-developers-who-want-to-capitalize","For developers who want to capitalize",[13,267,268,269,272,273,277],{},"The new labs created by DeepMind\u002FOpenAI defectors need ",[17,270,271],{},"systems, infra and frontend developers"," without PhDs. Salaries there are often more generous than Meta's, because they pay equity on very low valuations (in the sense of $5.1B vs $380B). The ",[116,274,276],{"href":275},"\u002Fblog\u002Fai-full-stack-developer-essential-role-2026","full-stack profile with an AI layer"," is the first to be targeted.",[104,279,281],{"id":280},"for-ai-operators","For AI operators",[13,283,284,285,288],{},"If you're building a consumer or B2B tool on OpenAI\u002FAnthropic\u002FGoogle APIs, know that ",[17,286,287],{},"non-LLM architectures are coming",". In 18 to 36 months, your stack could be obsolete if it assumes a transformer in the backend. Watch the new entrants.",[26,290,292],{"id":291},"the-big-tech-paradox","The Big Tech paradox",[13,294,295],{},"The giants created the conditions of their own bleeding.",[297,298,299,306,316],"ul",{},[300,301,302,305],"li",{},[17,303,304],{},"OpenAI"," closed access to its internal models for its researchers. Result: they leave to rebuild elsewhere.",[300,307,308,311,312,315],{},[17,309,310],{},"Meta"," paid Alexandr Wang ",[17,313,314],{},"$14B"," for Scale AI while laying off 10% of its non-strategic AI staff. Message received by researchers: only the top 0.1% is untouchable.",[300,317,318,321],{},[17,319,320],{},"Google DeepMind"," merged Google Brain and DeepMind, creating internal political conflicts. Silver, Rocktäschel and several others voted with their feet.",[26,323,325],{"id":324},"what-about-europe","What about Europe?",[13,327,328,329,332,333,337],{},"Ineffable is in ",[17,330,331],{},"London",". Mistral ",[116,334,336],{"href":335},"\u002Fnews\u002Fmistral-ai-billion-revenue-target-2026","aims for $1B in revenue in 2026",". H Company, Poolside, and several French labs are hiring. The UK Sovereign AI Fund co-invested in Ineffable. The narrative \"all AI talent is in California\" is crumbling.",[13,339,340,341,344],{},"Europe is no longer just a consumption market for American AI products. It's becoming a ",[17,342,343],{},"production hub"," again — driven by researchers who want to stay near their home universities and no longer need Silicon Valley to raise $1B.",[26,346,348],{"id":347},"conclusion-phase-2-of-the-ai-era","Conclusion: phase 2 of the AI era",[13,350,351],{},"Phase 1 (2022-2025) was the race to scale and LLMs. A handful of centralized labs, ever-larger models, stratospheric valuations for OpenAI and Anthropic.",[13,353,354,355,358],{},"Phase 2 (2026+) will be ",[17,356,357],{},"distributed",". Dozens of labs born from the brain drain, each specialized in an architecture, a domain, a learning method. Big Tech will continue to dominate distribution (ChatGPT, Gemini, Copilot), but upstream innovation migrates to the startup ecosystem.",[13,360,361,362,366],{},"This is excellent news for those who can navigate the noise. And a nightmare for those who thought AI was already settled. To understand how to monetize this new wave, our guide ",[116,363,365],{"href":364},"\u002Fblog\u002Findie-hacker-ai-profitable-projects-no-funding","indie hacker AI: profitable projects without funding"," is more relevant than ever.",[368,369],"hr",{},[13,371,372],{},[214,373,374,375,379],{},"The Big Tech brain drain to startups is not a blip. It's a market signal: AI researchers' pricing power has exploded, alternative architectures are returning, and upstream innovation is leaving commercial labs. Founders who understand this shift today will have a 12-18 month lead on the rest of the market. To dig deeper into 2026 trends, read ",[116,376,378],{"href":377},"\u002Fblog\u002Ftech-trends-transforming-development-2026","the major tech trends transforming development",".",{"title":381,"searchDepth":382,"depth":382,"links":383},"",2,[384,385,394,399,404,405,406],{"id":28,"depth":382,"text":29},{"id":101,"depth":382,"text":102,"children":386},[387,389,390,391,392,393],{"id":106,"depth":388,"text":107},3,{"id":123,"depth":388,"text":124},{"id":142,"depth":388,"text":143},{"id":156,"depth":388,"text":157},{"id":170,"depth":388,"text":171},{"id":184,"depth":388,"text":185},{"id":199,"depth":382,"text":200,"children":395},[396,397,398],{"id":206,"depth":388,"text":207},{"id":231,"depth":388,"text":232},{"id":238,"depth":388,"text":239},{"id":249,"depth":382,"text":250,"children":400},[401,402,403],{"id":253,"depth":388,"text":254},{"id":264,"depth":388,"text":265},{"id":280,"depth":388,"text":281},{"id":291,"depth":382,"text":292},{"id":324,"depth":382,"text":325},{"id":347,"depth":382,"text":348},"ai","2026-04-29T09:00:00","Top researchers are fleeing Meta, Google and OpenAI labs to launch their own AI startups. $18.8B already invested in 2026 in startups founded after January 2025, on pace to surpass the $27.9B raised in 2025. The face of AI innovation is shifting.","md","\u002Fbrain-drain-big-tech-startups-ia-2026.jpg",{},true,"\u002Fnews\u002Fai-brain-drain-meta-google-openai-startups-188-billion-vc-april-2026",{"title":6,"description":409},"$18.8B injected in 2026 into AI startups founded since 2025. David Silver, Yann LeCun, Tim Rocktäschel — top researchers leave Big Tech labs to build their own.","Big Tech AI brain drain: Meta, Google, OpenAI lose top talent","news\u002Fai-brain-drain-meta-google-openai-startups-188-billion-vc-april-2026",[420,421,422,423,424,425,426,427],"meta","google","openai","deepmind","brain-drain","venture-capital","reinforcement-learning","ai-startups","ai-brain-drain-big-tech-startups-2026","VQscgbKeIJHb0OtkYJKa0BBR4bwiN8Ji6Sa0bV-U81U",[431,438,444],{"title":432,"description":433,"path":434,"date":435,"image":436,"category":437},"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","vibecoding",{"title":439,"description":440,"path":441,"date":442,"image":443,"category":407},"Google makes Gemini Omni 1.2 Flash generally available with 15-second video","Google DeepMind cut Omni Flash pricing by 30% to $0.07 per second at 720p, raised clip length from 10 to 15 seconds, and opened the model on the Gemini API and Google Cloud.","\u002Fnews\u002Fgoogle-gemini-omni-1-2-flash-15-second-video","2026-11-06T08:00:00","\u002Fimages\u002Fnews\u002Fgoogle-gemini-omni-1-2-flash-15-second-video.jpg",{"title":445,"description":446,"path":447,"date":448,"image":449,"category":407},"Anthropic reportedly signs its first publisher licensing deal, with the AP","Adweek reports that Anthropic struck a content and data licensing deal with the Associated Press just before the US midterms, its first with a digital publisher. Neither company has confirmed it on the record, and the terms are undisclosed.","\u002Fnews\u002Fanthropic-first-publisher-licensing-deal-associated-press","2026-11-05T12:35:00","\u002Fimages\u002Fnews\u002Fanthropic-ap-publisher-licensing-deal.jpg",{"en":451,"fr":689,"es":8},{"id":5,"title":6,"author":7,"author_avatar":8,"body":452,"category":407,"date":408,"description":409,"extension":410,"image":411,"meta":686,"navigation":413,"path":414,"seo":687,"seoDescription":416,"seoTitle":417,"stem":418,"tags":688,"translationKey":428,"updatedAt":8,"__hash__":429},{"type":10,"value":453,"toc":662},[454,460,462,470,504,508,510,512,518,520,528,530,536,538,544,546,552,554,560,562,564,566,572,578,580,582,584,588,590,592,596,598,604,606,610,612,614,630,632,638,642,644,646,650,654,656],[13,455,15,456,20,458,24],{},[17,457,19],{},[17,459,23],{},[26,461,29],{"id":28},[13,463,32,464,36,466,40,468,44],{},[17,465,35],{},[17,467,39],{},[17,469,43],{},[46,471,472,480],{},[49,473,474],{},[52,475,476,478],{},[55,477,57],{},[55,479,60],{},[62,481,482,488,496],{},[52,483,484,486],{},[67,485,69],{},[67,487,43],{},[52,489,490,492],{},[67,491,76],{},[67,493,494],{},[17,495,81],{},[52,497,498,500],{},[67,499,86],{},[67,501,502],{},[17,503,91],{},[13,505,94,506,98],{},[17,507,97],{},[26,509,102],{"id":101},[104,511,107],{"id":106},[13,513,110,514,114,516,120],{},[17,515,113],{},[116,517,119],{"href":118},[104,519,124],{"id":123},[13,521,127,522,131,524,135,526,139],{},[17,523,130],{},[17,525,134],{},[17,527,138],{},[104,529,143],{"id":142},[13,531,532,149,534,153],{},[17,533,148],{},[17,535,152],{},[104,537,157],{"id":156},[13,539,540,163,542,167],{},[17,541,162],{},[17,543,166],{},[104,545,171],{"id":170},[13,547,548,177,550,181],{},[17,549,176],{},[17,551,180],{},[104,553,185],{"id":184},[13,555,188,556,192,558,196],{},[17,557,191],{},[17,559,195],{},[26,561,200],{"id":199},[13,563,203],{},[104,565,207],{"id":206},[209,567,568],{},[13,569,570],{},[214,571,216],{},[13,573,219,574,223,576,228],{},[17,575,222],{},[116,577,227],{"href":226},[104,579,232],{"id":231},[13,581,235],{},[104,583,239],{"id":238},[13,585,242,586,246],{},[17,587,245],{},[26,589,250],{"id":249},[104,591,254],{"id":253},[13,593,257,594,261],{},[17,595,260],{},[104,597,265],{"id":264},[13,599,268,600,272,602,277],{},[17,601,271],{},[116,603,276],{"href":275},[104,605,281],{"id":280},[13,607,284,608,288],{},[17,609,287],{},[26,611,292],{"id":291},[13,613,295],{},[297,615,616,620,626],{},[300,617,618,305],{},[17,619,304],{},[300,621,622,311,624,315],{},[17,623,310],{},[17,625,314],{},[300,627,628,321],{},[17,629,320],{},[26,631,325],{"id":324},[13,633,328,634,332,636,337],{},[17,635,331],{},[116,637,336],{"href":335},[13,639,340,640,344],{},[17,641,343],{},[26,643,348],{"id":347},[13,645,351],{},[13,647,354,648,358],{},[17,649,357],{},[13,651,361,652,366],{},[116,653,365],{"href":364},[368,655],{},[13,657,658],{},[214,659,374,660,379],{},[116,661,378],{"href":377},{"title":381,"searchDepth":382,"depth":382,"links":663},[664,665,673,678,683,684,685],{"id":28,"depth":382,"text":29},{"id":101,"depth":382,"text":102,"children":666},[667,668,669,670,671,672],{"id":106,"depth":388,"text":107},{"id":123,"depth":388,"text":124},{"id":142,"depth":388,"text":143},{"id":156,"depth":388,"text":157},{"id":170,"depth":388,"text":171},{"id":184,"depth":388,"text":185},{"id":199,"depth":382,"text":200,"children":674},[675,676,677],{"id":206,"depth":388,"text":207},{"id":231,"depth":388,"text":232},{"id":238,"depth":388,"text":239},{"id":249,"depth":382,"text":250,"children":679},[680,681,682],{"id":253,"depth":388,"text":254},{"id":264,"depth":388,"text":265},{"id":280,"depth":388,"text":281},{"id":291,"depth":382,"text":292},{"id":324,"depth":382,"text":325},{"id":347,"depth":382,"text":348},{},{"title":6,"description":409},[420,421,422,423,424,425,426,427],{"id":690,"title":691,"author":7,"author_avatar":8,"body":692,"category":407,"date":408,"description":1054,"extension":410,"image":411,"meta":1055,"navigation":413,"path":1056,"seo":1057,"seoDescription":1058,"seoTitle":1059,"stem":1060,"tags":1061,"translationKey":428,"updatedAt":8,"__hash__":1063},"news_fr\u002Ffr\u002Fnews\u002Fbrain-drain-meta-google-openai-startups-ia-188-milliards-vc-avril-2026.md","Brain drain dans la Big Tech : Meta, Google et OpenAI saignent leurs cerveaux IA — 18,8 Md$ injectés dans les startups fondées depuis 2025",{"type":10,"value":693,"toc":1030},[694,704,708,722,763,770,774,778,789,793,806,808,817,821,830,834,843,847,857,861,864,868,875,887,891,894,898,905,909,913,920,924,936,940,947,951,954,975,979,990,997,1001,1004,1011,1019,1021],[13,695,696,697,699,700,703],{},"Le 28 avril 2026, ",[17,698,19],{}," a posé un constat que tous les responsables RH des géants de l'IA refusaient d'admettre publiquement : ",[17,701,702],{},"Meta, Google et OpenAI saignent leurs talents les plus précieux",". Et ce ne sont pas n'importe lesquels — chercheurs principaux, leads de divisions entières, architectes de modèles fondamentaux. Ils partent. Et ils lèvent des montants qui n'ont aucun sens pour des startups sans produit.",[26,705,707],{"id":706},"le-chiffre-qui-change-tout-188-md-en-quatre-mois","Le chiffre qui change tout : 18,8 Md$ en quatre mois",[13,709,710,711,713,714,717,718,721],{},"Selon les données de ",[17,712,35],{}," citées par CNBC, ",[17,715,716],{},"18,8 milliards de dollars de capital-risque"," ont déjà été injectés en 2026 dans les startups IA fondées depuis le 1er janvier 2025. À ce rythme, l'année dépassera largement les ",[17,719,720],{},"27,9 Md$"," levés en 2025 par les sociétés nées après début 2024.",[46,723,724,734],{},[49,725,726],{},[52,727,728,731],{},[55,729,730],{},"Période",[55,732,733],{},"Capital VC dans les nouveaux labs IA",[62,735,736,743,753],{},[52,737,738,741],{},[67,739,740],{},"2025 (cohorte fondée depuis 2024)",[67,742,720],{},[52,744,745,748],{},[67,746,747],{},"2026 (cohorte fondée depuis 2025)",[67,749,750],{},[17,751,752],{},"18,8 Md$ en 4 mois",[52,754,755,758],{},[67,756,757],{},"Projection 2026 annualisée",[67,759,760],{},[17,761,762],{},"~56 Md$",[13,764,765,766,769],{},"Il faut prendre une seconde pour digérer ce chiffre. Plus de capital-risque va aujourd'hui dans des startups IA fondées par des transfuges de DeepMind, OpenAI et Meta FAIR que dans ",[17,767,768],{},"l'intégralité du financement series A\u002FB\u002FC de la fintech européenne"," sur la même période.",[26,771,773],{"id":772},"les-départs-qui-font-mal","Les départs qui font mal",[104,775,777],{"id":776},"ineffable-intelligence-david-silver-deepmind-seed-record","Ineffable Intelligence — David Silver (DeepMind → seed record)",[13,779,780,781,783,784,788],{},"Le 27 avril, ",[17,782,113],{},", ancien lead du reinforcement learning chez DeepMind, architecte d'AlphaGo et AlphaZero, est sorti du stealth avec ",[116,785,787],{"href":786},"\u002Ffr\u002Fnews\u002Fineffable-intelligence-david-silver-1-1-milliard-seed-deepmind-superlearner-avril-2026","Ineffable Intelligence et un seed historique de 1,1 Md$"," à 5,1 Md$ de valorisation. Sequoia, Lightspeed, Nvidia, Google, et même le Sovereign AI Fund britannique sont au tour de table. Pas de produit, pas de revenu, pas de roadmap publique. Juste un nom sur une slide.",[104,790,792],{"id":791},"ami-labs-yann-lecun-meta-1-md","Ami Labs — Yann LeCun (Meta → 1 Md$)",[13,794,795,796,798,799,801,802,805],{},"En mars 2026, ",[17,797,130],{}," quitte Meta après avoir publiquement acté son désaccord stratégique avec la roadmap LLM-centric du groupe. Sa nouvelle entité, ",[17,800,134],{},", lève ",[17,803,804],{},"1 milliard de dollars"," dans la foulée pour explorer les architectures non-LLM (energy-based models, world models). Mark Zuckerberg perd l'un des trois Turing Awards historiques de Meta.",[104,807,143],{"id":142},[13,809,810,812,813,816],{},[17,811,148],{},", ex-DeepMind, recherche jusqu'à ",[17,814,815],{},"1 Md$"," pour Recursive Superintelligence selon les rumeurs de Sand Hill Road. La thèse : auto-amélioration récursive, agents qui modifient leur propre code.",[104,818,820],{"id":819},"periodic-labs-alumni-openai-deepmind-300-m","Periodic Labs — alumni OpenAI + DeepMind (300 M$)",[13,822,823,825,826,829],{},[17,824,162],{},", fondée par d'anciens d'OpenAI et DeepMind, a levé ",[17,827,828],{},"300 M$"," en septembre 2025. Quelques mois après son lancement.",[104,831,833],{"id":832},"humans-reinforcement-learning-from-real-experience-480-m","Humans& — reinforcement learning from real experience (480 M$)",[13,835,836,838,839,842],{},[17,837,176],{}," sécurise ",[17,840,841],{},"480 M$"," en janvier 2026 pour explorer l'apprentissage par renforcement à partir d'expériences réelles plutôt que de données internet — un pari direct contre la doctrine LLM-first des grands labos.",[104,844,846],{"id":845},"recursive-intelligence-335-m-en-deux-tours","Recursive Intelligence — 335 M$ en deux tours",[13,848,849,850,852,853,856],{},"Enfin, ",[17,851,191],{}," (à ne pas confondre avec Rocktäschel) cumule ",[17,854,855],{},"335 M$"," sur deux tours après son lancement en septembre 2025.",[26,858,860],{"id":859},"pourquoi-les-chercheurs-fuient","Pourquoi les chercheurs fuient",[13,862,863],{},"CNBC pointe trois facteurs structurels qui expliquent la débâcle.",[104,865,867],{"id":866},"_1-la-pression-commerciale-tue-la-recherche-fondamentale","1. La pression commerciale tue la recherche fondamentale",[209,869,870],{},[13,871,872],{},[214,873,874],{},"« La pression de livrer des performances benchmark et de maintenir des cycles de release rapides laisse peu de place à la recherche véritablement exploratoire, particulièrement en dehors du paradigme LLM dominant. »",[13,876,877,878,881,882,886],{},"Quand OpenAI doit justifier sa ",[17,879,880],{},"valorisation de 500 Md$"," et Anthropic ses ",[116,883,885],{"href":884},"\u002Ffr\u002Fnews\u002Fgoogle-anthropic-40-milliards-investissement-deux-phases-cloud-tpu-avril-2026","380 Md$ après le deal Google à 40 Md$",", la recherche d'agents, l'interprétabilité, les nouvelles architectures et les modèles verticaux passent au second plan. Pas parce qu'ils n'importent pas — mais parce qu'ils ne gagnent pas la course du trimestre.",[104,888,890],{"id":889},"_2-les-actions-nont-plus-le-même-charme","2. Les actions n'ont plus le même charme",[13,892,893],{},"Quand Anthropic était à 40 Md$, un chercheur senior pouvait raisonnablement viser un x10 sur ses RSU. À 380 Md$, le upside devient marginal. Pendant ce temps, fonder Ineffable à 5,1 Md$ avec 30% d'equity = 1,5 Md$ sur le papier dès le seed. La math est cruelle.",[104,895,897],{"id":896},"_3-les-architectures-alternatives-reviennent-à-la-mode","3. Les architectures alternatives reviennent à la mode",[13,899,900,901,904],{},"Pendant trois ans, le marché a parié exclusivement sur les LLM scale-out. Les transfuges parient désormais sur les ",[17,902,903],{},"non-LLM"," : reinforcement learning from real experience, world models, energy-based models, neuro-symbolic. Ces architectures nécessitent une liberté de recherche que les labos commerciaux ne peuvent plus offrir.",[26,906,908],{"id":907},"ce-que-ça-change-pour-les-fondateurs","Ce que ça change pour les fondateurs",[104,910,912],{"id":911},"pour-les-founders-normaux","Pour les founders « normaux »",[13,914,915,916,919],{},"Si Sequoia signe un seed à 5,1 Md$ pour un ex-DeepMind sans produit, ",[17,917,918],{},"votre seed à 5 M$ pour un MVP qui marche est en fait une bonne affaire pour les VC",". La rareté des « vrais talents IA » fait monter les valos extrêmes du top — mais le tier en-dessous reste accessible.",[104,921,923],{"id":922},"pour-les-développeurs-qui-veulent-profiter","Pour les développeurs qui veulent profiter",[13,925,926,927,930,931,935],{},"Les nouveaux labs créés par des transfuges DeepMind\u002FOpenAI ont besoin de ",[17,928,929],{},"développeurs systèmes, infra et frontend"," qui n'ont pas de PhD. Les salaires y sont souvent plus généreux qu'à Meta, parce qu'ils paient en equity sur des valos très basses (au sens 5,1 Md$ vs 380 Md$). Les profils ",[116,932,934],{"href":933},"\u002Ffr\u002Fblog\u002Fdeveloppeur-full-stack-ia-profil-indispensable-2026","full-stack avec une couche IA"," sont les premiers ciblés.",[104,937,939],{"id":938},"pour-les-operators-ia","Pour les operators IA",[13,941,942,943,946],{},"Si vous bâtissez un outil grand public ou B2B sur les API d'OpenAI\u002FAnthropic\u002FGoogle, sachez que ",[17,944,945],{},"les architectures non-LLM arrivent",". Dans 18 à 36 mois, votre stack pourrait être obsolète si elle suppose un transformer en backend. Surveillez les nouveaux entrants.",[26,948,950],{"id":949},"le-paradoxe-big-tech","Le paradoxe Big Tech",[13,952,953],{},"Les géants ont créé eux-mêmes les conditions de leur saignement.",[297,955,956,961,970],{},[300,957,958,960],{},[17,959,304],{}," a fermé l'accès à ses modèles internes pour ses chercheurs. Résultat : ils partent pour reconstruire ailleurs.",[300,962,963,965,966,969],{},[17,964,310],{}," a payé Alexandr Wang ",[17,967,968],{},"14 Md$"," pour Scale AI tout en licenciant 10% de ses effectifs IA non-stratégiques. Message reçu par les chercheurs : seul le top 0,1% est intouchable.",[300,971,972,974],{},[17,973,320],{}," a fusionné Google Brain et DeepMind, créant des conflits politiques internes. Silver, Rocktäschel et plusieurs autres ont voté avec leurs pieds.",[26,976,978],{"id":977},"et-leurope-dans-tout-ça","Et l'Europe dans tout ça ?",[13,980,981,982,332,985,989],{},"Ineffable est à ",[17,983,984],{},"Londres",[116,986,988],{"href":987},"\u002Ffr\u002Fnews\u002Fmistral-ai-objectif-milliard-revenus-2026","vise le milliard de revenus en 2026",". H Company, Poolside, et plusieurs labs français recrutent. Le UK Sovereign AI Fund a co-investi dans Ineffable. Le narratif « tous les talents IA sont en Californie » s'effrite.",[13,991,992,993,996],{},"L'Europe n'est plus uniquement le marché de consommation des produits IA américains. Elle redevient un ",[17,994,995],{},"lieu de production"," — porté par des chercheurs qui veulent rester près des universités d'origine et qui n'ont plus besoin de Silicon Valley pour lever 1 Md$.",[26,998,1000],{"id":999},"conclusion-la-phase-2-de-lère-ia","Conclusion : la phase 2 de l'ère IA",[13,1002,1003],{},"La phase 1 (2022-2025) a été la course au scale et aux LLM. Une poignée de labos centralisés, des modèles toujours plus gros, des valorisations stratosphériques pour OpenAI et Anthropic.",[13,1005,1006,1007,1010],{},"La phase 2 (2026+) sera ",[17,1008,1009],{},"distribuée",". Des dizaines de labs issus du brain drain, chacun spécialisé sur une architecture, un domaine, une méthode d'apprentissage. Les Big Tech continueront à dominer la distribution (ChatGPT, Gemini, Copilot), mais l'innovation amont migre vers l'écosystème startup.",[13,1012,1013,1014,1018],{},"C'est une excellente nouvelle pour qui sait naviguer le bruit. Et un cauchemar pour qui croyait que l'IA était déjà jouée. Pour comprendre comment monétiser cette nouvelle vague, notre guide ",[116,1015,1017],{"href":1016},"\u002Ffr\u002Fblog\u002Findie-hacker-ia-projets-rentables-sans-levee-fonds","indie hacker IA : projets rentables sans levée de fonds"," reste plus pertinent que jamais.",[368,1020],{},[13,1022,1023],{},[214,1024,1025,1026,379],{},"Le brain drain Big Tech vers les startups n'est pas un blip. C'est un signal de marché : le pricing power des chercheurs IA a explosé, les architectures alternatives reviennent, et l'innovation amont quitte les labos commerciaux. Les founders qui comprennent ce shift dès aujourd'hui auront 12-18 mois d'avance sur le reste du marché. Pour creuser les tendances 2026, lisez ",[116,1027,1029],{"href":1028},"\u002Ffr\u002Fblog\u002Ftendances-tech-developpement-2026","les grandes tendances tech qui transforment le développement",{"title":381,"searchDepth":382,"depth":382,"links":1031},[1032,1033,1041,1046,1051,1052,1053],{"id":706,"depth":382,"text":707},{"id":772,"depth":382,"text":773,"children":1034},[1035,1036,1037,1038,1039,1040],{"id":776,"depth":388,"text":777},{"id":791,"depth":388,"text":792},{"id":142,"depth":388,"text":143},{"id":819,"depth":388,"text":820},{"id":832,"depth":388,"text":833},{"id":845,"depth":388,"text":846},{"id":859,"depth":382,"text":860,"children":1042},[1043,1044,1045],{"id":866,"depth":388,"text":867},{"id":889,"depth":388,"text":890},{"id":896,"depth":388,"text":897},{"id":907,"depth":382,"text":908,"children":1047},[1048,1049,1050],{"id":911,"depth":388,"text":912},{"id":922,"depth":388,"text":923},{"id":938,"depth":388,"text":939},{"id":949,"depth":382,"text":950},{"id":977,"depth":382,"text":978},{"id":999,"depth":382,"text":1000},"Top researchers fuient les laboratoires Meta, Google et OpenAI pour lancer leurs propres startups IA. 18,8 milliards $ déjà investis en 2026 dans les startups nées après janvier 2025, sur le rythme des 27,9 Md$ levés en 2025. Le visage de l'innovation IA bascule.",{},"\u002Ffr\u002Fnews\u002Fbrain-drain-meta-google-openai-startups-ia-188-milliards-vc-avril-2026",{"title":691,"description":1054},"18,8 Md$ injectés en 2026 dans les startups IA fondées depuis 2025. David Silver, Yann LeCun, Tim Rocktäschel : les pontes des labos quittent pour bâtir leurs propres labs.","Brain drain Big Tech IA : Meta, Google, OpenAI perdent leurs cerveaux","fr\u002Fnews\u002Fbrain-drain-meta-google-openai-startups-ia-188-milliards-vc-avril-2026",[420,421,422,423,424,425,426,1062],"startups-ia","le7tDkP_4TpYZyrvRMhEALuCVCd1QvHQp1glFI-Cyio"]