Overview

• OpenAI’s August launch of GPT-5 as a more adaptive, unified system drew heavy criticism, prompting the restoration of GPT-4o as Sam Altman acknowledged rollout missteps and described the new model as warm but not fawning.
• Altman said stronger models are held back by compute limits and signaled plans to spend billions on new data centers, as reports estimate Meta, Google, Amazon and Microsoft will invest more than $340 billion this year on AI infrastructure.
• Fresh research from MIT highlights modest returns for many corporate AI projects, and Nobel laureate Daron Acemoglu estimates only about 5% of tasks will be profitably automated in the next decade, implying roughly 1% GDP growth from AI in that span.
• Data-center constraints remain acute with an estimated 11,000 to 11,500 facilities worldwide and projections that electricity use could more than double by 2030, fueling interest in smaller on-device models to ease reliance on the cloud.
• Concrete scientific uses advanced as NASA and IBM unveiled Surya to predict solar eruptions up to two hours ahead and OpenAI reported a Retro Biosciences collaboration that redesigned proteins with a claimed 50-fold efficiency gain validated by external labs.