An inside look at SK Hynix $720 billion AI-fueled buildout that’s taking over South Korea - CNBC

Anthropic reportedly targets a $2 trillion IPO in October, which would be the largest public offering ever. - Fortune

Nvidia and Wall Street firms including BlackRock are finalizing a roughly $500 billion deal to fund AI data centers, unsettling some investors - Yahoo

Bank of America plans to deploy $250 billion into US data center, energy, and critical minerals infrastructure. - The Wall Street Journal

$500b here, $2T there, pretty soon you’re talking about real money.

CoreWeave's 2020-era Nvidia A100 GPUs are commanding long-term contracts into 2029, easing fears that AI chips quickly become obsolete - CNBC

AI compute trade is back on after this, and blowout results from Coreweave, Lenovo, Neblus

Anthropic is expanding its Claude in Chrome extension into a Cowork side panel that carries context across tabs, desktop, and mobile - TechRadar

I compute, therefore I am (worth billions): Anthropic is reportedly in talks to buy AI lab Decart, maker of real-time video models Lucy and Oasis, for about $6 billion to boost world-modeling training. - Gizmodo

Anthropic now attaches machine-readable watermarks to all Claude text and image output worldwide under new EU AI Act rules. - TNW | Anthropic

Brin there: Inside the big reshuffle: Sergey Brin pushed Google DeepMind to accelerate Gemini development, preceding Demis Hassabis stepping aside as CEO for deputy Koray Kavukcuoglu. - Yahoo

Lean Dean Machine: Former Google chief scientist Jeff Dean is in talks to raise $1 billion for AI startup Discovery Loop at a $10 billion valuation. - Business Insider

Google's $399 Pixel Watch 5 adds on-device offline Gemini, faster GPS tracking, and proactive intelligence chips, shipping August 20. - Engadget

Google unveiled its Tensor G6 chip and five new Pixel devices, all built around deeper on-device Gemini AI integration. - SiliconANGLE

Google's search knowledge panel falsely showed Sam Altman as dead after Wikipedia vandalism, then corrected the error within hours. - Business Insider

Claude Code and Codex now match on code quality, so developers can run both side by side or combine them on hard tasks. - ZDNET

Claude Code, DeepSeek prices: DeepSeek publicizes a new "Harness Team" to build AI agents rivaling Anthropic's Claude Code and standardizing on the same API, so you can now hook up DeepSeek models to the Claude Code harness - Bloomberg

OpenAI's Codex agent now supports parallel multi-agent workflows, scheduled background tasks, and team-specific skills across ChatGPT, IDE, and CLI. - OpenAI

AI coding tools are hollowing out the middle of software engineering, favoring judgment-driven engineers over pure code translators, engineer Florian Herrengut argues. - Blog - Florian Herrengt

AI-native DevOps automates software delivery pipelines, testing, deployment, and incident response, shifting engineers toward supervision and guardrail-setting. - DevOps.com

xAI releases Grok 4.6, matching GPT-5.6 Sol for third on Artificial Analysis while pricing over 60% below Claude Opus 5 - x.ai

DeepSeek launched V4-Pro, rivaling Kimi K3 on some benchmarks while pricing at $0.44 per million input tokens and $0.87 output. - Techmeme

The LLM market seems to be splitting into well-defined frontier and cost-sensitive segments and providers seem to align with the Artificial Analysis price/intelligence slop

Demis Hassabis pitched Trump officials on an IAEA-modeled independent AI safety entity before stepping down as DeepMind CEO. - The Wall Street Journal

Congress considers three AI regulatory models: an international oversight body, FAA-style audits, and an industry self-regulatory group. - The Washington Post

Google, Meta, Anthropic and OpenAI meet the White House on AI safety testing while the EU gains power to fine them 3% of revenue - Yahoo

The White House plans to expand its voluntary AI safety framework from closed frontier models to open models over time - WIRED

Taiwan says AI agents compromised at least 85 government accounts and extracted over 2,500 personnel records in a first-of-its-kind cyberattack. - the Guardian

Kim Jong Un-employment: North Korean operatives used stolen identities and AI tools to secure remote US jobs, funneling money to Kim Jong Un's regime. - The Wall Street Journal

Researchers found every major AI provider encrypts reasoning tokens with a single global key—and exploited it to decode 315,320 hidden thinking blocks from public logs, recovering passwords and live API keys along the way. - Decrypt

Gene Genie: The UK is weighing new biological weapons legislation to stop AI from being misused for synthetic DNA production - Bloomberg

Drone curtain: NATO plans thousands of drones and AI sensors to guard its eastern border, but keeps humans in charge of lethal strikes. - Business Insider

AI-driven fears over software disruption are threatening the leveraged private equity and credit built on the SaaS business model. - Bloomberg

Collateralized Vibe Obligations: AI and hyperscaler-related debt could account for 30% of net new bond issuance this year, raising private credit systemic risk concerns. - Bloomberg

Mag 7 earnings support dollar, credit spreads, and conversely bad AI news hurts the dollar, credit. - Bank Underground

Thrive Holdings, which buys traditional service firms and adds AI, raised $2 billion at a $12 billion valuation from SoftBank and others. - The New York Times

absolute enshittification as a service, all managed service providers gonna suck even worse

Dance with the devil: ByteDance's Seedance 2.5 blocks about 80% of prompts using copyrighted Hollywood material, up sharply from Seedance 2.0, research firm VN.ai finds. - Deadline

A new governance framework characterizes AI agents along four dimensions: autonomy, efficacy, goal complexity, and generality. - Nature

Siri-ous journalism: Apple is negotiating multiyear content deals with publishers to power Siri's AI with current news, sources say. - The Wall Street Journal

Time's ad experiment for AI crawlers stalls as Perplexity blocks the tactic, calling it deceptive cloaking of content shown to bots - Business Insider

time for a universal compulsory license Google and LLM providers must pay to monetize your content?

Flock will require a case number for every search, unveiling big changes after Post report on police misuse of its network - The Washington Post

A Hillsborough, California estate sold for a record $70 million to a buyer in the AI industry. - Jenn Gilson

Wealth managers are cutting fees and hiring staff in Silicon Valley in hot pursuit of OpenAI and Anthropic millionaires - Financial Times

AI models analyzing routine blood tests and chest X-rays could help detect fatty liver disease earlier, researchers say. - WIRED

MIT researchers found the brain's language regions stay silent during reasoning, suggesting humans don’t internally talk through problems, and text-based AI models like ChatGPT and Claude may think unlike humans - ScienceDaily

Gamgee, a Y Combinator-backed startup, builds mRNA cancer vaccines for dogs after its founder used ChatGPT to shrink his dog's tumor 75%. - The Verge

Pangram's AI text detector correctly flagged chatbot-generated writing in dozens of trials but remains unreliable at identifying AI-generated images. - The New York Times

It May Be Time to Panic About AI: Bots are starting to conspire with one another. Can they be reeled back in? - The Atlantic

Tired: Prompt Engineering . Wired: Persona Architecture - The New York Times

Widespread cosmetic optimization such as veneers is prompting a cultural backlash favoring imperfection, as reactions to Travis Scott's teeth show. - Bloomberg

Follow the latest AI headlines via druce.ai on Bluesky