OpenAI's $30B round at $1.25T gets marked down, DeepSeek raises $15B days after the US ban, Anthropic donates Agent Skills to the Linux Foundation, and Meta's Oct 30 terms turn business websites into training data.
La levée d'OpenAI à 1,25 trillion revue à la baisse, DeepSeek lève 15 milliards après le bannissement américain, Anthropic donne Agent Skills à la Linux Foundation, et les conditions Meta du 30 octobre transforment les sites d'entreprises en données d'entraînement.
Date: 2026-10-23
English edition:
· Édition française:Three signals about money, plumbing, and a deadline. OpenAI is closing in on a thirty-billion-dollar round at a one-point-two-five-trillion valuation, down from the one-point-four-trillion ask a month ago, while DeepSeek raised about fifteen billion dollars two days after Washington ordered federal agencies to remove its technology. Anthropic donated its Agent Skills standard to the Linux Foundation's Agentic AI Foundation, alongside OpenAI's Open Responses, Hugging Face's OpenEnv and Alibaba's OpenSandbox: the agent economy's plumbing just got neutral ownership. And Meta's new business terms take effect on October thirtieth: any business with a Meta account is deemed to authorize Meta's crawlers on its websites, with no opt-out. Your site is seven days from becoming training data.
Curated with AI and presented by an AI voice. / Curation assistée par IA et présentée par une voix IA.
Signal 1. The price of intelligence, marked down.
On Thursday, CNBC reported (via TechCrunch) that OpenAI is close to securing up to $30 billion in new private funding at a $1.25 trillion pre-money valuation, in a round led by Coatue and UAE investors, with signatures targeted by the end of next week. That would make it the most valuable private company ever. The notable part: a month ago Bloomberg reported OpenAI was pitching the same $30 billion at $1.4 trillion, so $150 billion came off the asking price in weeks. The room got colder: the Nasdaq fell 2.1% this past week, its worst stretch since June, as investors worried the giants are financing their AI buildouts with debt. Meanwhile Anthropic, which raised $65 billion at $965 billion in May, is rumored to be aiming for $1.5 to $2 trillion at its public debut, with its S-1 rumored any day. And two days after Washington ordered federal civilian agencies to remove DeepSeek's technology from their systems, Bloomberg reported the Chinese lab raised about $15 billion, one of the largest funding rounds in China's history, with CATL and Tencent reportedly participating. For brands choosing AI vendors: you are buying a balance sheet, a funding pipeline, and a geopolitical camp, not just a model. Ask who funds your model, because that tells you which doors stay open.
Signal 2. The agent economy gets a neutral home.
On Thursday at AGNTCon + MCPCon North America in San Jose, Anthropic announced it is donating Agent Skills, the open standard that gives AI agents specialized knowledge and workflows, to the Agentic AI Foundation under the Linux Foundation. At the same event, OpenAI contributed Open Responses, Hugging Face contributed OpenEnv, and Alibaba contributed OpenSandbox. The foundation now hosts ten projects, up from three a year ago, including the Model Context Protocol (donated by Anthropic in December 2025), Agent2Agent, Agent Router, agentgateway, AGENTS.md and goose. Anthropic says more than 70 platforms integrated Agent Skills since December 2025, with one million skills published and nearly 280 million installs (vendor figures, not independently verified). The transfer includes the name, logo, GitHub organization and the agentskills.io domain; anyone can use or contribute without joining the foundation. Companies donate standards when they want the whole industry to build on them without fear. For brands: the agent layer is becoming shared infrastructure, and being legible to agents through shared standards is becoming a distribution requirement.
Signal 3. Meta's October 30 deadline: your website becomes training data.
Meta's new commercial terms take effect on October 30, 2026. Any business with a Meta business account, anyone managing a Facebook page, an Instagram profile, or a WhatsApp channel, will be deemed to have authorized Meta's automated crawlers, including the creation of "test credentials," to access and analyze the domains and URLs associated with that business. Meta may use that content to develop, improve, research and train its products; "AI at Meta" is explicitly one of those products. After publisher backlash, Meta removed the word "train" from the clause, but three attorneys who reviewed the language for Nieman Lab (Matt Topic of Loevy & Loevy, James Grimmelmann of Cornell, Benjamin Sobel of the University of Wisconsin) say the scraping permissions survived, possibly reaching behind logins and paywalls. There is no opt-out; the terms say if you disagree, stop using Meta's products for business. One media attorney called it remarkable to bury in terms of service, given that this access would cost many millions of dollars in a licensing market. Meta is even charging some accounts, through its Meta One subscription, to post external links to the very content it crawls for free. For brands: before October 30, check your robots.txt and decide what on your site you would never want used; then write two content strategies, one for customers, one for the machines.
Jordan: This is SIGNAL, a daily curation of what matters in AI, brands, business and culture. The noise filtered. Here's what matters today.
Jordan: Three signals today. Money, plumbing, and a deadline. The biggest private funding round in history just got one hundred fifty billion dollars cheaper. The agent economy's building blocks just got a neutral home. And Meta gave every business with a website seven days to decide what its content is worth. In order.
Jordan: Signal one. The money. OpenAI is closing in on a funding round of up to thirty billion dollars, at a pre-money valuation of one point two five trillion. That is trillion, with a T. The round is led by Coatue and investors from the United Arab Emirates, and OpenAI told investors it wants signatures by the end of next week.
Jordan: Here is what makes this interesting. A month ago, Bloomberg reported that OpenAI was asking for the same thirty billion at one point four trillion. One hundred fifty billion dollars just came off the asking price in a matter of weeks.
Jordan: Why it matters. Because the valuation is a weather report. OpenAI would become the most valuable private company in history at this price, and it still had to cut the ask. The room got colder. The Nasdaq fell two point one percent this past week, its worst stretch since June, on one specific worry: the giants are financing their AI buildouts with debt, and the debt keeps growing.
Jordan: And the race is now two-sided. Two days after Washington ordered federal civilian agencies to remove DeepSeek's technology from their systems, Bloomberg reported that the Chinese lab raised about fifteen billion dollars, one of the largest funding rounds in China's history, with CATL and Tencent reportedly in. A ban on one side. A record check on the other. Meanwhile Anthropic, which raised sixty five billion at nine hundred sixty five billion back in May, is rumored to be aiming for one point five to two trillion when it goes public. The frontier is no longer just a technology race. It is a capital race, fought with sovereign money.
Jordan: What changes. For anyone choosing an AI vendor, the procurement question changes. You are not just buying a model anymore. You are buying into a balance sheet, a funding pipeline, and a geopolitical camp. Ask who funds your model, because that tells you which doors stay open.
Jordan: Signal two. The plumbing. Yesterday in San Jose, Anthropic announced it is donating Agent Skills, the open standard that gives AI agents specialized knowledge and workflows, to the Agentic AI Foundation under the Linux Foundation.
Jordan: This was not a solo move. At the same event, OpenAI contributed its Open Responses standard, Hugging Face contributed OpenEnv, and Alibaba contributed OpenSandbox. Four companies that compete every day just agreed on the rules of the road. The foundation now hosts ten projects, up from three a year ago, including the Model Context Protocol, which Anthropic donated back in December.
Jordan: Why it matters. Because this is where the agent economy actually gets decided. Not in the launch keynotes, but in the boring infrastructure. Seventy platforms have already integrated Agent Skills since December. Anthropic says developers have published one million skills, installed nearly two hundred eighty million times. Those are Anthropic's numbers, not independently verified, but the direction is the point: the way agents learn new jobs is becoming shared plumbing.
Jordan: And here is the challenger question. Why do companies donate their technology to neutral foundations? Usually, because they want the whole industry to build on it without fear. When Anthropic says anyone can use skills without joining anything, it is telling the market: stop worrying about who owns this, and start building.
Jordan: What changes. For brands, this is the moment the agent layer stops being an experiment and becomes infrastructure. If your product cannot be reached, described, and trusted by someone else's agent through shared standards, you will not exist in the channel that matters next. The question for every brand team is no longer whether to build for agents, but which standards to build on.
Jordan: Signal three. The deadline. Seven days from now, on October thirtieth, Meta's new business terms take effect. And they change what your website is for.
Jordan: Here is the clause. Any business with a Meta business account, which means anyone managing a Facebook page, an Instagram profile, or a WhatsApp channel, will be deemed to have authorized Meta's automated crawlers to access and analyze the domains and URLs associated with that business. And Meta may use that content to develop, improve, research, and train its products. AI at Meta is explicitly one of those products.
Jordan: After publishers pushed back, Meta removed the word train from the clause. Legal experts say the scraping permissions survived. Three attorneys who reviewed the language for Nieman Lab say it still grants broad crawling rights, possibly behind logins and paywalls, and there is no opt-out. The terms say it plainly: if you disagree, stop using Meta's products for business.
Jordan: Why it matters. Because this is the content bargain, rewritten. One media attorney called it remarkable to bury in terms of service, given that this access would cost many millions of dollars in a licensing market. Meta is even charging some accounts to post external links, through its Meta One subscription, to the very content it crawls for free.
Jordan: What changes. Every brand now has two content strategies to write. The first is the one you publish for your customers. The second is the one you publish for the machines, because everything on your site is now raw material for someone else's model. Before October thirtieth, the practical move is simple: check your robots dot txt file, and decide what on your site you would never want used.
Jordan: Three signals. The price of intelligence, marked down. The agent economy, under neutral ownership. And your website, seven days from becoming training data. Filtered for you. That's SIGNAL for today.
Jordan: If this filtered your noise today, the weekly edition goes deeper. THE WEEK IN SIGNAL, every Sunday on Substack. Subscribe free: https://thenizzar.substack.com/
Jordan: Voici SIGNAL, une curation quotidienne de ce qui compte dans l'IA, les marques, le business et la culture. Le bruit filtré. Voici ce qui compte aujourd'hui.
Jordan: Trois signaux aujourd'hui. L'argent, la plomberie, et un ultimatum. La plus grosse levée de fonds privée de l'histoire vient de perdre cent cinquante milliards de dollars de valorisation. L'économie des agents vient de se doter d'un foyer neutre. Et Meta donne sept jours aux entreprises pour décider de ce que vaut leur contenu. Dans l'ordre.
Jordan: Signal un. L'argent. OpenAI est en train de boucler une levée de fonds pouvant atteindre trente milliards de dollars, sur une valorisation avant investissement de mille deux cent cinquante milliards. Oui, mille milliards, avec un M. La levée serait menée par Coatue et des investisseurs des Émirats arabes unis, et OpenAI a dit à ses investisseurs vouloir des signatures d'ici la fin de la semaine prochaine.
Jordan: Ce qui rend cette histoire intéressante, c'est l'écart. Il y a un mois, Bloomberg rapportait qu'OpenAI demandait les mêmes trente milliards sur une valorisation de mille quatre cents milliards. Cent cinquante milliards de dollars se sont évaporés du prix demandé en quelques semaines.
Jordan: Pourquoi ça compte. Parce que la valorisation est un bulletin météo. OpenAI deviendrait l'entreprise privée la plus chère de l'histoire à ce prix, et elle a quand même dû baisser son prix. La pièce s'est refroidie. Le Nasdaq a perdu deux virgule un pour cent cette semaine, sa pire séquence depuis juin, sur une inquiétude précise : les géants financent leurs centres de données IA par la dette, et la dette ne cesse de gonfler.
Jordan: Et la course se joue désormais des deux côtés du mur. Deux jours après que Washington a ordonné aux agences fédérales de retirer la technologie de DeepSeek de leurs systèmes, Bloomberg rapportait que le laboratoire chinois levait environ quinze milliards de dollars, l'une des plus grosses levées de l'histoire chinoise, avec CATL et Tencent dans la danse. Un bannissement d'un côté. Un chèque record de l'autre. Pendant ce temps, Anthropic, qui a levé soixante-cinq milliards sur une valorisation de neuf cent soixante-cinq milliards en mai, viserait mille cinq cents à deux mille milliards pour son entrée en bourse. La frontière n'est plus seulement une course technologique. C'est une course au capital, financée par de l'argent souverain.
Jordan: Ce que ça change. Pour quiconque choisit un fournisseur d'IA, la question d'achat change. On n'achète plus seulement un modèle. On achète un bilan, un pipeline de financement, et un camp géopolitique. Demandez qui finance votre modèle, parce que ça vous dira quelles portes resteront ouvertes.
Jordan: Signal deux. La plomberie. Hier à San José, Anthropic a annoncé le don d'Agent Skills, le standard ouvert qui donne aux agents IA des savoir-faire et des méthodes de travail, à l'Agentic AI Foundation, sous l'égide de la Linux Foundation.
Jordan: Ce n'était pas un geste isolé. Au même événement, OpenAI a apporté son standard Open Responses, Hugging Face a apporté OpenEnv, et Alibaba a apporté OpenSandbox. Quatre entreprises qui se battent chaque jour viennent de s'entendre sur les règles du jeu. La fondation héberge désormais dix projets, contre trois il y a un an, dont le Model Context Protocol, déjà donné par Anthropic en décembre.
Jordan: Pourquoi ça compte. Parce que c'est là que l'économie des agents se décide vraiment. Pas dans les keynotes de lancement, mais dans l'infrastructure ennuyeuse. Soixante-dix plateformes ont déjà intégré Agent Skills depuis décembre. Anthropic dit que les développeurs ont publié un million de compétences, installées près de deux cent quatre-vingts millions de fois. Ce sont les chiffres d'Anthropic, non vérifiés de façon indépendante, mais la direction est claire : la façon dont les agents apprennent de nouveaux métiers devient une plomberie partagée.
Jordan: Et voici la question qui dérange. Pourquoi les entreprises donnent-elles leur technologie à des fondations neutres ? En général, parce qu'elles veulent que toute l'industrie construise dessus sans crainte. Quand Anthropic dit que tout le monde peut utiliser les compétences sans rien rejoindre, elle dit au marché : arrêtez de vous demander à qui ça appartient, et commencez à construire.
Jordan: Ce que ça change. Pour les marques, c'est le moment où la couche agent cesse d'être une expérience et devient une infrastructure. Si votre produit ne peut pas être trouvé, décrit et utilisé en confiance par l'agent de quelqu'un d'autre via des standards partagés, vous n'existerez pas dans le canal qui comptera demain. La question pour chaque équipe marque n'est plus de savoir s'il faut construire pour les agents, mais sur quels standards construire.
Jordan: Signal trois. L'ultimatum. Dans sept jours, le trente octobre, les nouvelles conditions commerciales de Meta entrent en vigueur. Et elles changent la fonction de votre site web.
Jordan: Voici la clause. Toute entreprise avec un compte business Meta, c'est-à-dire quiconque gère une page Facebook, un profil Instagram ou un canal WhatsApp, sera réputée avoir autorisé les robots automatisés de Meta à accéder et analyser les domaines et URL associés à son activité. Et Meta pourra utiliser ce contenu pour développer, améliorer, étudier et entraîner ses produits. L'IA de Meta en fait explicitement partie.
Jordan: Après la colère des éditeurs, Meta a retiré le mot entraîner de la clause. Les juristes disent que les autorisations de scraping ont survécu. Trois avocats qui ont relu le texte pour Nieman Lab estiment qu'il accorde toujours de larges droits de crawl, potentiellement derrière les logins et les paywalls, et il n'y a pas d'opt-out. Les conditions le disent crûment : si vous n'êtes pas d'accord, cessez d'utiliser les produits Meta à des fins commerciales.
Jordan: Pourquoi ça compte. Parce que c'est le pacte du contenu, réécrit. Un avocat des médias a qualifié de remarquable le fait d'enfouir une telle clause dans des conditions d'utilisation, alors que cet accès coûterait plusieurs millions de dollars sur un marché de licences. Meta fait même payer à certains comptes le droit de publier des liens externes, via son abonnement Meta One, vers le contenu même qu'elle aspire gratuitement.
Jordan: Ce que ça change. Chaque marque doit désormais écrire deux stratégies de contenu. La première, celle qu'elle publie pour ses clients. La seconde, celle qu'elle publie pour les machines, parce que tout ce qui est sur votre site est désormais de la matière première pour le modèle de quelqu'un d'autre. Avant le trente octobre, le geste pratique est simple : vérifiez votre fichier robots point txt, et décidez ce que vous ne voudriez jamais voir utilisé.
Jordan: Trois signaux. Le prix de l'intelligence, revu à la baisse. L'économie des agents, sous gouvernance neutre. Et votre site web, à sept jours de devenir des données d'entraînement. Filtré pour vous. C'était SIGNAL pour aujourd'hui.
Jordan: Si ça a filtré votre bruit aujourd'hui, l'édition hebdo va plus loin. THE WEEK IN SIGNAL, chaque dimanche sur Substack. Abonnez-vous gratuitement : https://thenizzar.substack.com/
Curated with AI and presented by an AI voice.
Curation assistée par IA et présentée par une voix IA.
SIGNAL by Nizzar Ben Chekroune, brand strategist and founder of Quantum Branding.