Amazon is in talks to sell its Trainium AI chips to third parties for data-center use. So far, AWS has resisted outside customers because its existing capacity is maxed out. Opening sales would risk leaving current clients waiting, unless Amazon can ramp up chip production—a stretch given industry lead times. This move signals AWS’s bid to challenge Nvidia beyond cloud services by staking a claim in the AI-hardware market. Midjourney, known for its image-generation AI, is branching into hardware with a full-body ultrasound machine that users operate while partially submerged in water. The “Midjourney Scanner” contains no AI but kicks off the company’s push into medical devices. They plan 50,000 units, debuting the first in a branded spa in San Francisco. Meanwhile, NASA tapped Eric Schmidt–backed Relativity Space for Aeolus, a Mars orbiter launching in 2028. It will carry four instruments to map daily global dust, wind and temperature patterns—data critical for future landers and human missions. On the software side, firms are moving from flat-fee AI subscriptions to usage-based pricing. That model forces customers to monitor costs and justify each query. In the agent-architecture world, Browserbase is a sponsor pitching real-browser sessions to outsmart site blocks and CAPTCHAs, running 35 million sessions per month. A new language called Clear merges spec and implementation in one file, so agents and humans read the same source. And a deep dive into “agent loop architecture” lays out how to build resilient AI workflows that self-heal and extend their own skills. Intel stocks jumped after Trump and Apple announced a US chipmaking partnership under CEO Lip-Bu Tan. Finally, some developers warn that fleets of unsupervised coding agents risk erasing downtime, replacing breaks with constant oversight.