AI Daily Brief: June 30, 2026

Infrastructure & Compute: The AI compute shortage dominated headlines. Google cut off Meta's access to Gemini AI models due to capacity constraints — a move confirmed by multiple outlets, revealing how compute scarcity now strains even partnerships. Samsung and SK Hynix pledged over $550B to ease the 'RAMageddon' memory crisis. China's CAICT released the first-ever benchmark for AI infrastructure operations, covering five domestic chip platforms. NVIDIA introduced three workflows using synthetic data and fine-tuning to boost vision AI agent accuracy for industrial settings.

Models & Products: DeepSeek V4 officially launched with peak-hour API prices doubled, signaling strong demand and commercialization shift. Anthropic struck a deal with California Governor Newsom, giving state agencies half-price Claude access, and separately partnered with enterprise IT firm Globant for multi-year Claude deployment. Claude models became available on Microsoft Azure Foundry powered by NVIDIA GB300 Blackwell Ultra GPUs. Google published a detailed explainer on its full-stack AI approach and rolled out personalized image generation and advanced image analysis to free Gemini users, plus a new AI shopping feature.

Enterprise & Business: Amazon launched a $1B Frontier Deployment Engineering (FDE) organization, embedding engineers in enterprise customers for custom AI agent deployment. Amazon is also reportedly evaluating OpenAI's models and its own Nova as alternatives to rising Anthropic costs. HSBC partnered with Google Cloud for AI-powered banking. Crypto exchange OKX plans a marketplace for AI agents to hire and pay each other. Vibe coding platform Base44 announced its own AI model.

AI Agents & Open Source: Minglue Technology open-sourced Octo, described as an 'internet between agents.' OpenClaw released native iOS and Android apps for its AI agent platform. Coocaa launched an enterprise AI operating system with a four-layer agent architecture. The U.S. Senate proposed the AI AGENT Act to govern enterprise AI agent use.

Policy & Society: The U.S. softened its ban on Anthropic's Mythos 5 model. TIDAL cut off monetization for AI-generated music. The Washington Post published a systematic test of political bias in AI chatbots like ChatGPT. A survey found 61% of U.S. adults use AI for health information, up from 2% in 2024. A report showed high-intensity AI adopters grew headcount 10.2%, countering the job-killing narrative.

China Updates: Zhipu AI CEO Tang Jie solicited public input for GLM-5.3 features, with users overwhelmingly requesting vision capabilities. Meituan confirmed its latest AI model was trained on domestic chips. A Hong Kong-listed newcomer is betting on physical AI. AI-powered protocol standardization achieved 98.6% pass rate in biomanufacturing.