AI inference costs have rapidly decreased, with predictions of over 90% reduction for large language model providers by 2030 compared to last year, and LLMs becoming up to 100 times more cost-efficient than 2022 models due to improved hardware and model design.