Coinbase turns to Chinese AI models to reduce operating costs
U.S. crypto exchange Coinbase has reportedly adopted the Chinese AI models GLM 5.2 and Kimi 2.7 in an effort to lower AI-related spending. The decision highlights a growing shift in enterprise procurement logic: instead of choosing models purely on headline performance, companies are increasingly prioritizing cost efficiency, deployment practicality, and sustainable unit economics.
For a large platform such as Coinbase, AI is not just an experimental layer. Once AI usage scales across internal workflows, customer support, analytics, or product operations, model inference costs can become a meaningful operating line item. In that context, the choice of model provider becomes a financial decision as much as a technical one.
Cost savings came from routing, caching, and context engineering
The report says Coinbase cut its AI spending by about 50% through a combination of automatic routing, improved cache hit rates, and better context engineering. These are practical optimization techniques rather than a simple one-for-one model replacement.
Automatic routing allows different tasks to be assigned to the most suitable and cost-effective model. Higher cache hit rates reduce redundant computation by reusing prior outputs where appropriate. Context-engineering improvements help control how much information is passed into each query, which can directly affect inference cost and latency. Taken together, these measures show that enterprise AI optimization increasingly depends on orchestration and system design, not just model quality alone.
Other U.S. companies are following the same playbook
The trend is not limited to Coinbase. According to the report, companies such as Lindy and Snowflake are also shifting toward lower-cost models to bring down AI expenses. That suggests a broader market pattern in which enterprise buyers are reassessing whether premium-priced frontier models justify their cost across every workflow.
As more companies test and deploy cheaper alternatives, the AI market’s pricing structure is being reshaped. The result is a more competitive environment in which model vendors are judged not only by benchmark performance, but also by total cost of use, workload fit, and integration efficiency.
Pricing pressure is building for OpenAI and Anthropic
The report points to a wider repricing trend across the AI sector. As enterprises become more comfortable mixing model providers and optimizing workloads across tiers, established U.S. leaders such as OpenAI and Anthropic face a more direct price war. This pressure is especially relevant in sectors like crypto, where fast-moving companies often prioritize flexible infrastructure and cost control over vendor loyalty.
In that sense, Coinbase’s move is more than an isolated procurement change. It reflects a broader shift in enterprise AI economics: model selection is increasingly becoming a competitive sourcing decision shaped by cost, routing strategy, and operational efficiency.