In the past 13 years (2010-2023), the amount of computation devoted to training the best AI models has doubled on average more than twice per year, representing a 10-billion-fold increase since 2010; this acceleration explains why large language models emerged only in the past year despite AI research spanning decades.