“Ryan Petersen believes there is a realistic scenario where AI companies like OpenAI or Anthropic could cut off customer access to their models, deciding their compute is more valuable for training future superintelligence.”