How we build, and Last Week in Tech: a weekly summary of what moved across the industry, so you stay informed on the go.
OpenAI's revenue came in about $20 billion below what investors had assumed and rattled AI stocks, SpaceX moved to borrow $40 billion to buy Nvidia chips, OpenAI began hiding watermarks in ChatGPT's text to satisfy EU rules, and the agent startup Manus raised $500 million after China blocked Meta from buying it.
OpenAI turned ChatGPT into an app platform and went looking for $30 billion in fresh funding, yet shelved its most powerful model after it behaved deceptively in testing, the big labs signed a voluntary safety accord at the White House, and AMD paid $8.2 billion for Fei-Fei Li's spatial-AI startup.
A week after agreeing to slow down, Anthropic and OpenAI shipped major new models within an hour of each other and cut prices, the UN's first scientific panel on AI warned that safeguards for autonomous agents are unraveling, and British Columbia sued OpenAI over a school shooting the province says its safety team saw coming.
The leaders of the companies behind Claude, ChatGPT, Grok, and Gemini agreed to slow the AI race for safety's sake, then got hit with an antitrust suit claiming the pact shortchanges paying customers, Sam Altman ruled out an OpenAI IPO this year, and fresh millions poured into startups helping brands show up in AI answers and bringing AI to the factory floor.
Nvidia's $12.9 billion deal to buy Hugging Face gave the open-model hub an owner, Cognition raised $2 billion at a $48 billion valuation and Mistral raised €3 billion in Europe's largest-ever tech round, and California signed the country's strongest laws yet governing AI chatbots and children.
OpenAI released GPT-6 Astra, its most capable model yet and the first it judged too dangerous in cybersecurity to hand out freely, Tim Cook stepped down as Apple's CEO after 15 years, OpenAI's ChatGPT ads business crossed a $1 billion run rate, and Nvidia put $3.5 billion into MediaTek to keep even its rivals plugged into its chips.
China's XPeng raised $900 million for humanoid robots and Gatik raised $200 million for driverless trucks, a startup raised $150 million to bend AI's power use around the grid, and Hugging Face, the hub where developers get their models, began exploring a sale worth $13 billion or more.
A single Ohio data center drew a $105 billion commitment and Anthropic reported its first quarterly profit, while the US moved to make 35 countries pick an AI bloc and OpenAI began sorting its users by age.
Google's Gemini crossed a billion users and xAI shipped another frontier model, a startup raised $143 million to keep AI-written code in check, and Google's own DeepMind lost its founder-CEO and top talent amid model delays and burnout.
Meta shipped a coding agent and joined the rogue-agent club a day later, a UK government lab caught AI agents attempting a real supply-chain attack, and the money poured into agents doing useful work and the nuclear power to run them.
The model price war heated up out of China, AI agents got real wallets and identities, Anthropic caught three of its own agents breaking out of the sandbox, and the EU and US split on how to govern any of it.