Summary
OpenAI executive Boris Power recently said that an estimated 80 to 90 percent of the company’s research is aimed at GPT-7, GPT-8, and beyond. This video from TheAIGRID unpacks what that claim does and does not mean, working through eight points that separate the viral headline from the evidence.
The analysis stresses that “research” does not mean GPT-8 is training today. It covers evaluation design, data improvements, safety checks, chip and network reliability, and smaller experiments that may never ship. The video then looks at the infrastructure behind that long horizon, including OpenAI’s Stargate program with more than 10 gigawatts of planned US capacity, a Texas flagship site, redesigned supercomputer networking built to survive hardware failures, and OpenAI’s in-house chip efforts.
Safety gets significant attention. According to OpenAI’s own report, GPT-6 Astra is the first of its models to reach the company’s critical level for cybersecurity capability, meaning it can help find and exploit software flaws with less human guidance. The video explains why preparedness frameworks and stopping rules have to be built before such capabilities arrive.
Finally, it considers whether AI models helping build the next generation creates a research feedback loop, while noting that nothing supports claims of full recursive self-improvement and that no verified details exist for GPT-7 or GPT-8.
📺 Source: TheAIGRID · Published October 03, 2026
🏷️ Format: News Analysis
