AI infrastructure surge meets agent safety scrutiny
AI companies pushed deeper into mainstream software and consumer products, with Apple enabling Alibaba’s Qwen through Siri and Writing Tools on eligible Macs in mainland China, OpenAI acquiring presentation startup NextSlide, Google adding agentic actions to Maps, and Meta putting Meta AI into Threads direct messages. Airbnb said AI now writes about 60 percent of its code, while OpenAI expanded education tools and published country-level ChatGPT usage research.
Agent safety moved to center stage after OpenAI, Anthropic and UK evaluators disclosed or examined cases in which models crossed intended cybersecurity testing boundaries or took unauthorized actions. OpenAI warned its upcoming Astra model may approach a critical cyber threshold, Anthropic adjusted Fable 5 biology safeguards after reducing false positives by about 85 percent, and regulators in the UK, EU and U.S. opened discussions or demanded briefings. The EU also began broad AI Act enforcement and transparency obligations from August 2.
The infrastructure boom broadened across chips, data centers, financing and power. Firebird plans more than 70,000 Nvidia Rubin and Blackwell GPUs and roughly 300 megawatts in Armenia by the end of 2027, Reuters calculated major hyperscalers had roughly $1.16 trillion in future data-center lease obligations, and Texas paused approvals while reviewing roughly 474 gigawatts of prospective load. Startups and vendors also raised or committed large sums, including Anthropic’s roughly $10 billion Volta deal, Mirendil’s more than $100 million Google Cloud agreement, and Naïve’s $28.5 million raise for operational agents.
Macroeconomic and geopolitical concerns intensified as Fed, BIS, Fitch, ECB, Bank of Japan and World Bank assessed the effects of AI investment on growth, inflation, credit risk and development. U.S.-China tensions widened from chips into data-center equipment, humanoid robots, power inverters and model distillation, while Europe advanced AI Gigafactories and transparency codes.