Coinbase CEO Brian Armstrong stated that the model cost for most AI workloads could drop by 99% within the next 12 to 18 months. He believes that enterprises will increasingly offload routine tasks to older and open-source models, rather than continuing to rely on more expensive, high-end models.

Coinbase has adopted a low-cost model

Armstrong made the above assessment in response to concerns about rising API costs. He said that Coinbase has already allocated many notification requests to lower-cost models to control overall expenses.

He also revealed that AI has already generated approximately 40% of Coinbase's code. This means that AI has become a relatively important production tool within the company during the software development process, and cost control has become a prerequisite for the company to expand its use.

Open source models drive down costs

Armstrong believes the main factors driving this price decline include the continued improvement of open-source models and the rapid decrease in inference costs. As more models approach usable capabilities, enterprises may not need to use the most expensive models in many everyday scenarios.

In his opinion, higher-priced cutting-edge models will still retain a market, but will mainly be used for cutting-edge research and more complex AI applications, rather than covering all business scenarios.

Enterprise deployment paths may continue to diverge.

This statement reflects a real trend in enterprise AI deployment: striking a balance between cost, speed, and effectiveness. For most standardized tasks, lower-priced models are becoming more attractive.

If this trend continues, the structure of enterprise AI spending may become further differentiated. Some budgets will flow to high-performance models, while others will shift to cheaper, more controllable open-source or mature models.