AI training costs are skyrocketing, with current models like ChatGPT-4 costing up to $100 million to train. However, CEO Dario Amodei predicts these costs could balloon to $10-100 billion in just three years due to the shift from generative artificial intelligence to artificial general intelligence. This increase is fueled by factors such as the need for over 30,000 GPUs for training, with each Nvidia B200 AI chip costing $30,000 to $40,000. The power required for these GPUs is equivalent to that needed for 1.3 million homes. Despite advancements like Google's JEST method to reduce training time and resources, the strain on resources and budgets is evident. Companies like Microsoft, OpenAI, and Tesla are investing heavily in AI, with plans for $100 billion data centers and millions of chips. The future of AI training costs looks daunting and exponential, costing billions.