AI inference costs falling 13 times annually

AI inference costs are declining at a rate that exceeds Moore’s Law. According to a report by research institution Epoch AI, the cost of using AI models has dropped by an average of approximately 47% per quarter since 2023, representing a 13-fold decrease annually.

This decline in the cost of “artificial thought” outpaces several other major technological advancements, including DNA sequencing, traditional computing, and lithium battery improvements. For example, the report notes that while an OpenAI model might have cost $0.30 per question in early 2025, subsequent models achieving similar performance levels saw costs drop significantly.

The rapid price reduction presents challenges for AI providers, as they may struggle to maintain premium subscription models based solely on model performance. While lower costs reduce the barrier for small and medium-sized enterprises to adopt AI for tasks like customer service and document organization, the report suggests businesses should evaluate models based on accuracy, integration, and data security rather than price alone.