The cost-saving AI measure Coinbase's CEO is taking to keep costs 'roughly flat' while growing token usage

As enthusiasm for token maximization wanes, AI users are increasingly searching for ways to optimize their expenditures without sacrificing performance. Coinbase CEO Brian Armstrong recently disclosed the company's approach to managing costs while boosting token utilization. In a post on X, Armstrong emphasized the importance of utilizing more affordable AI models, stating, "We're working hard on routing prompts to cheaper models where appropriate, and in some cases have been able to keep costs roughly flat, while token usage continues to grow exponentially." While cutting-edge models like Opus 4.8 and GPT-5.5 offer advanced capabilities, they often come at the expense of higher token consumption. Armstrong noted that many users experienced frustrations with rate limits after the launch of Opus 4.7. However, he anticipates a shift in usage patterns, predicting that within the next 12 to 18 months, approximately 80% of workloads will rely on models that are 99% cheaper. According to Armstrong, the latest models will primarily be reserved for situations requiring peak performance, such as scientific advancements and complex orchestration tasks. He suggested that future limitations in AI will stem more from energy and computing resources than from the availability of superior models. Armstrong's insights have sparked interest among industry leaders. Venture capitalist Marc Andreessen found his perspective intriguing, while Julien Chaumond, cofounder of Hugging Face, noted the growing trend of model routing. Box CEO Aaron Levie remarked that while Armstrong's projections might seem extreme, he believes AI usage will begin to stratify, with high-end tasks being managed by advanced models and routine tasks handled by budget-friendly alternatives. Winston Weinberg, cofounder of Harvey, pointed out the critical need for effective intelligence allocation in this evolving landscape. As the focus shifts towards efficiency, the previous trend of flaunting high token expenditures seems to be fading. Notably, Garry Tan, CEO of Y Combinator, had previously encouraged founders to embrace token usage without restraint. The sentiment appears to be changing, as Glean cofounder Tony Gentilcore remarked that Armstrong's insights were "spot on," indicating that those in the technical field have long understood these shifts, while financial markets have yet to catch up with the realities of AI pricing.