“Chamath Palihapitiya predicts that a 15 to 20 trillion parameter model will be trained on NVIDIA's Vera Rubin architecture within the next 6 to 9 months.”