Coinbase CEO says AI spending nearly halved as token usage surged

Coinbase CEO Brian Armstrong said AI spending has come close to being cut in half even as token usage grows exponentially, arguing that companies can control costs through better default models, intelligent routing, cache optimization, leaner context management and usage transparency rather than strict usage caps. He said Coinbase shifted its default AI stack toward open-source models including GLM 5.2 and Kimi 2.7, while reserving frontier models for cases where they are necessary. The idea comes as enterprises face rapidly rising AI bills. Zhipu's GLM 5.2 is priced at $1.40 per million input tokens and $4.40 per million output tokens, versus Anthropic's Opus 4.8 at $5 and $25 for the same volume. A KPMG survey found only 26% of companies have full visibility into AI costs, while 22% discover spending only after the bill arrives, and Goldman Sachs projects AI token consumption could rise 24-fold by 2030 to 120 quadrillion tokens per month. Armstrong's comments have also drawn scrutiny over the security and geopolitical risks of Chinese-origin AI models. While GLM 5.2's MIT license allows companies to self-host the model and avoid routing sensitive data through an outside API, concerns remain over legal exposure, model behavior and cybersecurity. U.S. lawmakers opened a formal inquiry in May into cybersecurity risks from Chinese-origin AI models in critical infrastructure, and Anthropic said in a letter to the Senate Banking Committee that Alibaba Qwen operators ran 28.8 million Claude exchanges through about 25,000 fake accounts between April and June in what it called the largest known campaign to steal a model's capabilities.