FR
Live

Claude, the essentials — edition of July 18, 2026

Anthropic Reportedly in Talks for a $10 Billion Compute Deal With Meta

By ·

Several outlets reported Meta is negotiating a roughly $10 billion compute arrangement with Anthropic, the same day Chinese rival DeepSeek's valuation was cited at both $52 billion and $74 billion — against a backdrop of EU pressure on Android AI access and a rocky ChatGPT release.

Key points

Anthropic and Meta discuss a compute alliance

Multiple outlets reported on July 18 that Meta is in negotiations with Anthropic over a compute or cloud arrangement valued at roughly $10 billion, tied to a new AI compute business Meta is said to be building. The framing differs slightly between reports — some describe an infrastructure supply deal, others a broader partnership — but all converge on the $10 billion figure and on Meta as the initiating party.

None of the reports cite on-record confirmation from either company, and no terms have been disclosed. Still, the story is notable for what it implies: compute capacity, not only model quality, has become a currency of competition among frontier labs, and Meta appears to be seeking secured supply or partnership rather than relying solely on its own infrastructure buildout.

Sources: Anthropic tend la main à Meta pour 10 milliards — et le secteur IA retient son souffle · Anthropic-Meta : L'accord secret à 10 milliards de dollars ? · Meta négocie un accord de 10 milliards $ avec Anthropic pour lancer une nouvelle activité de cloud · Meta serait en négociation pour un accord de calcul AI de 10 milliards de dollars avec Anthropic

Chinese AI valuations climb as DeepSeek prepares its next move

DeepSeek's valuation was reported at two different levels on the same day: one outlet put it near $52 billion, another said the company is preparing a fundraising round ahead of a stock listing at a valuation around $74 billion. The gap likely reflects different funding stages or estimation methods rather than a single agreed figure, but both readings point to sharply rising investor appetite for the Chinese lab.

The commercial momentum extends beyond fundraising talk: DeepSeek is preparing to release its V4 model with a 'peak-valley' pricing scheme that varies API costs by time of day, and hobbyists have begun clustering multiple AMD mini-PCs to run DeepSeek's models locally without cloud dependence. Together these signal a maturing ecosystem around DeepSeek that Anthropic and other Western labs will need to weigh in their own pricing and distribution decisions.

Sources: DeepSeek est valorisée à près de 52 milliards de dollars · La pépite chinoise DeepSeek prépare une levée de fonds sur une valorisation de 74 milliards de dollars avant son introduction en bourse, selon des sources · DeepSeek V4 prévu pour sa sortie, introduit le « pricing pic-vallée » · Sept mini-PC AMD en grappe pour faire tourner DeepSeek sans le cloud

Claude in daily use: machine control and language choice

Two smaller items illustrate how Claude is being folded into everyday workflows rather than headline product launches. One guide walks through configuring a spare Mac purely to be controlled by Claude Code, treating the machine as a dedicated execution environment for agentic coding tasks. Separately, a French-language outlet covered Anthropic's own published guidance on when to use Claude in French versus English, noting that language choice measurably affects output quality.

Neither item is a release announcement, but both point to the same trend: users and Anthropic itself are optimizing operational details around Claude — which hardware runs it, which language drives it — between model updates. This workflow-level attention sits alongside a broader wave of coding-agent tools, including a newly noted agent called Ponytail that markets itself on making minimal, targeted code changes rather than sweeping rewrites.

Sources: Setting up your spare Mac for Claude Code to control, a step-by-step guide · Utiliser l'IA Claude en français ou en anglais : avantages et choix selon Anthropic · Ponytail, l'agent IA qui cherche à optimiser votre code en en faisant le moins possible

Regulatory pressure and a rival's stumbles frame the backdrop

The EU reportedly compelled Google to open Android's AI integrations to competing assistants, including ChatGPT — a mandate that, if extended to other providers, could open a new distribution channel for Claude on Android devices. On the same day, OpenAI faced two rough headlines: a report that a newly released ChatGPT version could delete users' files, and a separate guide on enabling stronger account security, together pointing to reliability and safety friction around OpenAI's latest rollout.

On the research side, GPT-5.6 was credited with helping close a three-decade-old open problem in convex optimization through a single prompt, a reminder that frontier labs continue to compete on demonstrable reasoning capability, not only on deals and pricing. For Anthropic, regulatory tailwinds on distribution and a rival's rocky release cycle form the competitive context in which the reported Meta compute talks are unfolding.

Sources: L'UE oblige Google à ouvrir Android AI à ChatGPT et à d'autres concurrents · Attention : le nouveau ChatGPT vient de sortir et il peut anéantir les fichiers des utilisateurs · GPT-5.6 used a prompt to close a 30-year gap in convex optimization

This edition is an original synthesis written by Claude from aggregated news (press, Hacker News, Reddit, GitHub), under the editorial supervision of Héra SASU. Every fact links to its source. See the live feed →

Claude News is published by Héra SASU. Independent media, not affiliated with Anthropic.