Larry Ellison: Models such as ChatGPT, Gemini, Grok, and Llama are mostly trained on the same publicly available internet data. When everyone is training with the same information, models inevitably become homogenized. The real moat isn't the model itself, but the proprietary data behind it. Companies that can train using exclusive datasets gain a core advantage that competitors simply cannot replicate.