Ways to think about token pricing (www.ben-evans.com)

Recent discussions on token pricing in AI highlight a significant supply crunch and instability in the market, with experts predicting that the landscape will evolve over the next few years. The current surge in demand is primarily driven by specific use cases like software development, creating uncertainty about future applications and their token requirements. Analysts are exploring whether high-performance models can maintain sustainable pricing power, or if they will become low-margin commodities, mirroring historical tech trends where early innovators failed to capture long-term value. The complexity of the token pricing ecosystem revolves around multiple factors, including the increased capital expenditure for data centers, improvements in inference efficiency, and unpredictable training costs. Analysts suggest two approaches to tackle this uncertainty: a bottom-up analysis that models supply and demand based on various factors, or a top-down perspective that examines broader industry trends. Importantly, the conversation draws parallels with other tech evolution scenarios, such as mobile networks and semiconductor manufacturing, raising crucial questions about future competitive landscapes and the potential for dominance by a few key players versus a multitude of more accessible models. Ultimately, the AI/ML community must navigate these uncertainties to identify viable paths for development and value capture in a rapidly changing environment.