Coinbase CEO Brian Armstrong has pivoted the cryptocurrency exchange to Chinese AI models, deploying GLM 5.2 and Kimi 2.7 across its operations. The shift reflects mounting cost pressures on Western AI providers.
Coinbase built an automated routing system that selects the optimal model for each request based on task requirements and pricing. The system improved cache hit rates from 5 to 60 percent, a dramatic efficiency gain. The company halved its AI spending while token usage continued climbing, demonstrating the cost gap between Chinese and Western models.
The move signals a broader pattern. Chinese AI providers offer more aggressive pricing than OpenAI, Google, Anthropic, and other Western labs. For large-scale operations like Coinbase's, these economics tip the scales decisively. Armstrong's decision carries particular weight given Coinbase's prominence in crypto and tech circles.
Western AI labs face a pricing stress test. As enterprises scale AI deployments, cost differentials widen the decision calculus. Chinese models now compete directly on capability while undercutting on price. Providers like Zhipu AI (GLM) and Moon Shot AI (Kimi) have rapidly improved model quality, narrowing the performance gap that once justified premium Western pricing.
The implications ripple across the industry. Large consumers of AI compute have latitude to shop globally. Western providers cannot rely on performance alone if Chinese alternatives deliver comparable results at half the price. This pressure forces difficult choices for OpenAI, Google, and competitors: cut margins, improve efficiency, or differentiate on capabilities Western models genuinely lead on.
Coinbase's switch also highlights a geopolitical dimension. U.S. restrictions on AI chip exports aim to slow Chinese advancement, yet Chinese models still undercut American alternatives economically. Enterprise customers increasingly prioritize cost efficiency over Cold War alignment, creating tension between policy goals and market