OpenAI chairman Bret Taylor on AI tokenomics, token efficiency

Summary

Bret Taylor discusses the state of the AI boom, focusing on tokenomics, token efficiency, and the competitive dynamics between US frontier labs and open weight models. He argues that token efficiency, not training cost, is the real driver of value, giving OpenAI and Anthropic a durable edge. Taylor also explains how his company Sierra uses outcome-based pricing to address enterprise ROI concerns.

• Frontier models from OpenAI and Anthropic offer superior token efficiency compared to open weight models
• Open weight models like China's K3 may be cheaper to train but not necessarily cheaper to run
• Token efficiency is the critical factor for inference cost and quality
• US labs need to lead on price, performance, latency, and cost across all tasks
• Enterprises are increasingly worried about AI token spending and ROI
• Sierra charges clients per successful outcome (e.g., loan origination) rather than per token
• Outcome-based pricing helps companies build a competitive moat by compounding customer data
• The applied AI market remains immature, and maturing it will solve many tokenomics issues